05 Biomedical Engineering

338
BIOELECTRIC PHENOMENA
nation of the potential field at the body surface from underlying bioelectric activity, can be formulated.
The cardiac forward problem starts with a quantitative description of the sources in the heart; the resulting body surface potentials are known as the electrocardiogram. In a similar way sources associated with the activation of skeletal
muscle lead to the electromyogram. We will also consider the
electroencephalogram and electrogastrogram, where we will
discover bases for sources other than propagating action potentials. We consider these applications of basic theory only
in an introductory way, because there are separate articles for
each. It is the goal of this article to elucidate the underlying
principles that apply to each of the aforementioned and
other applications.
MEMBRANE ELECTROPHYSIOLOGY
Excitable CellsMacroscopic Structure
The main mammalian tissues that are electrically excitable
are nerves and muscles. Although such cells vary greatly in
size, shape, and electrical properties, there are nevertheless
certain fundamental similarities. In order to illustrate their
different cellular structures, we introduce excitable cell histology in this section, although it is somewhat ancillary to the
general goals of this article, and it is very brief; the interested
reader may consult one of the references for more detailed
information. Some additional material will also be found later
in this article in the section on Applications.
The application of engineering principles and technology to
medicine and biology has had an increasing influence on the
practice of medicine. The most visible of these contributions
is in the form of medical devices. This article, however, describes the engineering introduction of quantitative methods
in the field of bioelectricity. When such contributions first became evident, in the early 1950s many physiology researchers
were already employing modern quantitative methods to develop and utilize governing equations and suitable models of
bioelectric phenomena. Today it appears that systems physiology lives on as biomedical engineering, while physiology has
become more concerned with cell and molecular biology. On
the other hand, biomedical engineering is also currently involved in efforts to develop and apply quantitative approaches
at cellular and molecular levels.
This article, which is concerned with the electric behavior
of tissues, reviews what is known about the biophysics of excitable membranes and the volume conductor in which they
are imbedded. Our approach emphasizes the quantitative nature of physical models. We formulate an engineering description of sources associated with the propagating action potential and other excitable cellular phenomena. With such
sources and a mathematical description of fields generated in
a volume conductor the forward problem, namely a determi-
Nerve. A sketch of a typical neuron is given in Fig. 1(a),

and contains dendrites, the cell body, and an axon. All elements are enclosed by a membrane (which separates the intracellular from the extracellular space) and are electrically
excitable. However the axon shown is myelinated, that is, its
membrane is surrounded by an insulating sheath except at
periodic nodes of Ranvier (to which any possible transmembrane current is restricted). We are particularly interested in
axonal propagation and the accompanying current flow fields.
The diameter of nerve fibers varies from 0.3 to 20 m, (the
very small fibers are unmyelinated), and a propagation velocity of 0.5 to 120 m/sec (higher for larger diameters) is observed. One of the main effects of myelination is an increase
in propagation velocity for the same fiber diameter. The
length of an axon can vary from micrometers (cells in the cortex) to meters (motoneurons from the spinal cord to extremities).
Skeletal Muscle. A description of skeletal muscle structure
is given in Fig. 1(b). The whole muscle is shown subdivided
into fascicles, each of which contains many fibers. An individual fiber contains myofibrils, which constitute the contractile
machinery of the muscle. The membrane surrounding each
fiber is excitable. Axial propagation of an electrical impulse
can take place over this membrane, just as for the nerve axon.
However the muscle fiber has a transverse structure namely
the T system, which is also excitable and which conducts the
electrical impulse from the surface radially inward, where it
activates the sarcoplasmic reticulum (SR); this, in turn, initiates excitationcontraction of the muscle. Because it is unmyelinated, propagation velocity is not as great as for nerve fi-
J. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering. Copyright # 1999 John Wiley & Sons, Inc.
339
Cell body
(soma)
Initial segment
of axon
Node of Ranvier Schwann cell
Axon hillock
Nucleus
Terminal
buttons
Dendrites
(a)
Muscle
Fibers
Tendon
Fibril
Tendon
(b)
bers (typically 5 m/sec) and its diameter lies between 10 m

and 100 m.
When skeletal muscle is viewed under the microscope,
characteristic cross-striations are seen. This arises from the
arrangement of thin and thick filaments, components of the
myofibrils; interaction of the filaments is the basis for the development of contractile force.
Cardiac Muscle. The cardiac muscle fiber looks superficially
like that of the skeletal muscle fiber containing, in particular,
similar contractile proteins (the T system location differs from
that of skeletal muscle) and show the same striation pattern.
Cardiac cells, however, do not comprise long fibers, as in skeletal muscle. A typical cardiac cell length is only 100 m. However cardiac cells are interconnected by gap junctions, which
are sites through which ions and hence electric activity may
easily pass. Consequently cardiac muscle propagation can
take place throughout cardiac tissue as if, functionally, cell
membranes were contiguous (syncytial).
Smooth Muscle. Smooth muscle differs from skeletal and
cardiac muscle in that it does not exhibit cross-striations. It
also has a poorly developed SR. There are thick and thin contractile filaments, however their structure is irregular, which
accounts for the absence of a banded appearance. Smooth
muscle cells range in size from 2 to 10 m diameter and 20
600 m length. Individual cells are joined together mechanically by attachment plaques and, often, electrically via gap
junctions. This structural arrangement is similar to that
found in cardiac muscle. The gap junctions are of low electric
resistance and contribute intercellular coupling, which acts to
synchronize electrical activity of adjoining cells.
Figure 1. (a) A motor-neuron is shown with typical

structure including dendrites, cell body, and axon.
Activation of the cell body arises from the summation of excitatory inputs from the dendrites. A propagating impulse travels out on the axon to terminal
buttons (neuromuscular junctions) at which the impulse is conducted to a target muscle. The axon
shows Schwann cells that provide the myelin
sheath, which effectively insulates the axon except
at the nodes of Ranvier. From W. F. Ganong Medical
Physiology, Los Altos, CA: Lange Medical Pub, 1971.
(b) The arrangement of fibers in a whole muscle and
the myofibrils contained in each fiber. The excitable
plasma membrane surrounds each fiber, and the
contractile machinery is responsible for the crossstriations seen in each fibril. From Keynes and
Aidley, Nerve and Muscle, Cambridge: Cambridge
University Press, 1981 after K. Schmidt-Nielsen,
Animal Physiology, Cambridge: Cambridge University Press, 1979.
Excitable CellsMembrane Structure

The simplest electrophysiologic model is that of a single excitable cell lying in an unbounded uniform conducting medium.
If we imagine the cell to have been activated, then action currents will be observed to flow from the activated site throughout the intracellular and extracellular space. The source of
this current is associated with the membrane (since both intracellular and extracellular regions are passive). This section
is devoted to a consideration of the structure and function of
the biological membrane.
The main constituent of the biological membrane is a lipid
bilayer, approximately 5 nm thick, as illustrated in Fig. 2.
Because this material is oily, it has a very high electrical resistance and is an effective insulator to ion movement. There
are, however, transmembrane proteins that contain aqueous
channels; it is only the presence of these channels that endows membranes with ionic permeability.
Channel proteins have been studied biophysically by electron microscopy, electron diffraction, and biologically at the
level of molecular structure. Although a fairly consistent picture emerges, we still do not have an accurate structural
model. Based on what is known, Hille (1) created the cartoon
of a channel protein given in Fig. 2. A typical such protein is
in length and 80 A
in diameter. Imporapproximately 120 A
tant for ion movement is the aqueous channel. These channels also have gates, the opening and closing of which control
ion flow. These gates are typically sensitive to electric fields
in the membrane, implying that the channel protein contains
charged regions that are influenced by the electric field to
cause a conformational change, which, in some way, controls
the gate. Another important channel property is selectivity,
340
Lipid
bilayer
+
+
Extracellular
side
Narrow
selectivity
filter
Aqueous pore
Gate
Sugar
residues
Cytoplasmic
side
Figure 2. The figure shows the membrane that

bounds an excitable cell consisting of a lipid bilayer (two layers of lipid with their hydrophillic
heads facing outward and nonpolar tails inward)
and an ionic channel that penetrates this layer.
The channel structure is based on electron microscopy and electron diffraction studies. The channel is drawn as a transmembrane macromolecule
with a hole through the center. The external surface of the molecule is glycosylated. The functional
regionsselectivity filter, gate, and sensorare
deduced from voltage-clamp experiments and are
only beginning to be charted by structural studies.
We have yet to learn how they actually look. Figure and quotation from B. Hille, Ionic Channels of
Excitable Membranes, 2nd ed., Sunderland, MA:
Sinauer Assoc., 1992.
Voltage
sensor
+
+
Channel
protein
P
Anchor
protein
+
0
nm
by which a channel may allow the passage of only one particular ion species; selectivity may depend on the channel diameter, the charges that line the channel, or other details.
An important tool in the study of membranes is molecular
genetics. These techniques have been used to determine the
primary structure of most channels of interest. Unfortunately
it has not been possible to deduce the secondary and tertiary
structure. However educated guesses lead to a determination
of which portions of the primary amino acid sequence is intramembrane, cytoplasmic, and extracellular. As noted in Fig. 2
the channel protein extends into the cytoplasm as well as the
extracellular space.
The Squid Axon
Hodgkin and Huxley (2) pioneered a quantitative study of excitable membranes in the 1950s. For their preparation, they
used the giant axon of the squid. This axon was chosen because of its large diameter (approximately 500 m), which
allowed the insertion of an axial electrode. Until this time all
measurements of the electric behavior of excitable cells utilized only external electrodes, which left much information
inaccessible. In the absence of intracellular potentials the
conventional wisdom was that the resting membrane was depolarized, meaning that it was at zero transmembrane potential (the term depolarization continues to be used, although
it now simply implies activation of an excitable membrane).
Hodgkin and Huxley measured resting potentials on the order
of 70 mV (inside minus outside).
The squid axon, like any nerve, can be activated by passing
an adequate (transthreshold) pulse of current between two
electrodes in the external bath. A propagating action potential of the shape described in Fig. 3 is initiated at the activating end and travels to the opposite end. Except for end effects
propagation is characterized by an unchanging waveshape
and uniform velocity (assuming an axially uniform preparation).
The squid axon exemplifies an unmyelinated nerve fiber.
Although this is not typical of nerve fibers in the human body,
it presents a very simple model for analysis. One may consider that the intracellular space is simply a uniform electrolyte, whereas the extracellular space (sea water) constitutes
an independent electrolyte. Both intracellular and extracellular regions are electrically passive, and consequently whatever mechanism is responsible for the action potential must
involve the membrane.
From a chemical analysis of intracellular fluid and sea water (which constitutes the extracellular environment for the
squid), Hodgkin and Huxley determined that the major ions
available for current flow are K, Na, and Cl. They also
noted that the ionic composition of the extracellular fluid differs markedly from the intracellular. The intracellular and
extracellular concentrations of the aforementioned ions associated with the squid giant axon are shown in Table 1.
The squid axon contains a very high intracellular potassium concentration. If we assume the membrane to be permeable only to potassium, then from Table 1 we would expect
potassium ions to flow out of the intracellular space to the
lower concentration extracellular space. This single ion movement can only be transient, because positive charge will accumulate at the outside of the membrane, leaving negative
Table 1. Intracellular and Extracellular Concentrations of
Ions Associated with the Squid Giant Axon
Ion
Intracellular (mM)
Extracellular (mM)
345
72
61
10
455
540
K1
Na1
Cl1
Even though the resting potassium permeability exceeds

that of the chloride and sodium ions, the latter are not negligible. However from the numerical values found previously,
it is clear that there is no transmembrane potential that will
equilibrate all ions. Consequently the condition that must be
met at rest is that of a zero net transmembrane current that
is, a steady-state condition where
Source. Hodgkin-Huxley (2).
IK + ICl + INa = 0
charge at the inside surface of the membrane and thereby

establishing an electric field that acts inward and inhibits further potassium efflux. (Note that the lipid bilayer constitutes
an ideal capacitive dielectric, and a typical membrane capacitance for charge storage is 1 F/cm2). Equilibrium requires
that the outward flux due to diffusion be balanced by the inward flux due to the resultant electric field (additional details
will be presented shortly). Writing and equating such equations permits the derivation of the Nernst equation, which
evaluates the single-ion transmembrane potential for an ion
to be in equilibrium. For the potassium ion example, it is
given as
Vm =
[Ko ]
[Ko ]
RT
ln
= 25.2 ln
in mV
F
[Ki ]
[Ki ]
(1)
where Vm is the transmembrane potential defined as the intracellular minus extracellular potential across the membrane, R is the gas constant, F Faradays constant, and T the
absolute temperature. The coefficient RT/F evaluates to 25.2,
for Vm in mV, assuming T at room temperature (20 C). Note
that for anions the ratio in Eq. (1) must be inverted, giving
(for chloride) Vm 25.2 ln ([Cli]/[Clo]). For the numerical values in Table 1 we evaluate the potassium Nernst potential as
89.2 mV, the chloride as 54.09 mV, and the sodium as
46.5 mV.
Stimulus
Recording
Transmembrane potential, mv
50
(2)
Each ionic current based on diffusion and electric field has

been evaluated (3,4). The flux component due to diffusion or
the presence of a concentration gradient is described by Ficks
law. This gives the ith ion flux as
jdi = Di dCi /dx
(3)
where Di is Ficks constant and x is the direction across the

membrane; the flux is from high to low concentration. The
flux due to the electric field is given by
jei = ui (zi /|zi |)Ci
(4)
where ui is the ions mobility, zi its valence, and the potential gradient. Goldman assumed a constant electric field in
the membrane and hence set
= Vm /d
(5)
where d is the membrane thickness; the outward electric field

is evaluated in Eq. (5).
The electric current is evaluated by multiplying the aforementioned flux (in moles/cm2) by Faradays constant, F,
times zi /zi to take account the sign of the ion flow. Thus with
ji jdi jei the electric current density, Ii, is
Ii = F (zi /|zi |) ji
(6)
To add Eq. (3) and Eq. (4), a relationship between Di and ui

is needed; this was furnished by Einstein (5) in the expression
Nerve
S
341
Di =
ui RT
|zi |F
(7)
Evaluating the potassium, sodium, and chloride electric currents arising from Eqs. (3), (4), and (7) and inserting each into
the steady-state constraint of Eq. (2) leads to the following
equation for the steady-state transmembrane potential Vm
namely
Stimulus
artifact
1 msec
100
Figure 3. The transmembrane potential measured with an intracellular microelectrode registers the stimulus artifact and an elicited
action potential on a nerve axon. The electrode configuration is shown
in the inset above. The transmembrane potential does not return to
baseline smoothly but shows a positive after-potential.
Vm = 25.2 ln
P
+ PNa [Na]o + PCl [Cl]i

PK [K]i + PNa [Na]i + PCl[Cl]o
K [K]o

(8)
where RT/F is replaced by 25.2 for T 293 C and hence Vm

is assumed to be in mV (4). In Eq. (8) Pi is the permeability
of the ith ion equal to Di /d in this simple model. Note that, if
we assume PNa PCl 0, Eq. (8) degenerates into Eq. (1).
Hodgkin and Katz (4) applied Eq. (8) to their experiments and
found a good fit utilizing PK : PNa : PCl 1.0 : 0.04 : 0.45, which
yields a value of Vm 59.5 mV. As expected the resting
potential does not equilibrate any individual ion, although the
342
chloride is close to its Nernst potential in this illustration.

The sodium ion has a driving force (i.e., there is a difference
between the actual transmembrane potential and the value
that would result in equilibrium) of 59.5 ( 46.5) 106
mV (which, being negative represents a net inward electric
field), whereas potassium has a driving force of 29.7 mV [
59.5 (89.2)] which, being positive is an outward field. In
spite of the much larger sodium driving force its low permeability results in a lower sodium influx than potassium efflux.
The excess potassium efflux is balanced by the chloride efflux
(driven by the voltage 59.5 54.9 4.6 mV) (where chloride, being negatively charged, moves outward under the net
inward electric field) bringing the total current to zero.
Low-resistance seal
(50 M)
Suction
Seal, G
Cell attached
KCI/Ca+free
pulse of suction
or voltage
Pull
Patch Clamp
The Nobel prize-winning work of Neher and Sakmann (6) was
for the development of the Patch Electrode. This is a glass
micropipette with a tip diameter of 1 m or less. It is carefully
fire-polished so that when placed against a cell membrane
and with the application of gentle suction a very high resistance (gigaOhms) seal may be achieved. (Special cleaning of
the cell membrane may also be required.) Once this very high
resistance seal is achieved, then, as described in Fig. 4, four
configurations can subsequently be obtained. In the cellattached configuration the patch electrode measures transmembrane currents over the small membrane area contacted;
the cell itself remains intact. Other configurations include
only the membrane patch itself or the entire cell membrane
(patch removed).
The results of an experiment with the cell-attached electrode are shown in Fig. 5, which gives the transmembrane
current response to a transmembrane voltage step; nine successive trials are described. In each case it is seen that in
the small accessed area only a single channel (identified as
potassium) contributes to the measured current; the current
is either zero or 1.5 Pa depending on whether the channel is
closed or open. The need for a gigaOhm seal can now be understood as necessary to prevent currents from entering the
patch electrode via an extracellular pathway (leakage currents); without the gigaseal even small amounts of such extraneous currents could easily obscure the extremely small
desired transmembrane current.
A number of important characteristics of the single channel can be deduced from the experiments shown in Fig. 5.
We note that the channel has only two stateseither open or
closed. From the successive trials we also see that the response is different each time and is hence stochastic. On the
other hand when a large number of successive trials are
summed we see that a definite pattern emerges, so that the
probability, n(t), that a channel is open can be described as a
function of time; this is the same function as the averaged
channel conductance as a function of time as seen in the ensemble average curve [Fig. 5(b)]. Although this curve can be
found by averaging 40 successive trials of the same single
channel, we would also expect this result were we to conduct
a single trial in which the simultaneous current from 40 channels were measured. In fact in the whole-cell recording configuration (Fig. 4) because both the intracellular and extracellular regions must have the same potential, the entire cell
membrane is at the same transmembrane voltage and all
channels are in parallel; the measured transmembrane cur-
Pull
Pull
Using a
small cell
Pull
low Ca+
Pull
Air
exposure
Pull
10 m
Whole-cell
recording
Outside-out
patch
Inside-out
patch
Figure 4. Four configurations for patch clamping are described. The

clean pipette is pressed against a cell to form a tight seal using light
suction, and producing the cell-attached or on-cell configuration. Pulling the pipette away from the cell establishes an inside-out patch.
Application of a suction pulse disrupts the membrane patch, allowing
electrical and diffusional access to the cell interior for whole-cell recording. Pulling away from the whole-cell arrangement causes the
membrane to reform into an outside-out configuration. From O. P.
Hamill, A. Marty, B. Sakmann, and F. J. Sigworth, Improved patch
clamp techniques for high resolution current recording from cells and
cell-free membrane patches, Pflugers Arch., 391: 85100, 1981.
rent for a single ion will then be proportional to the probability that a single channel is open.
We also note from Fig. 5 that although the onoff intervals
change as a function of time (these are the aforementioned
random variables) the magnitude of channel current is fixed.
Scaling the curve in Fig. 5 we deduce that the channel conductance is roughly 20 pS. Thus, when the channel is open
we can describe its current as
iK = 20(Vm EK ) pA
(9)
where EK is the potassium equilibrium (Nernst) potential.

Furthermore, assuming a first-order rate process and letting
the aforementioned n(t) be the probability of a channel being
open we have
dn
= (1 n) n
dt
(10)
where is the probability of a transition from open to closed

and is the probability of a transition from closed to open.
Equation (10) describes the time rate of increase in probability of a channel being open as the probabilty of a change from
the closed to the open state [namely times the probability
of being closed (1 n)] minus the probability of a change from
open back to closed (namely times the probability of being
in the open state n). If the potassium channel density is C
channels per square centimeter then the membrane-specific
conductance will be 20nC pS/cm2, because nC will be open for
a large number of channels.
HodgkinHuxley Membrane Model
In developing a practical membrane model one must take into
account a very large number of channels of several ion species. A macroscopic model could then be developed, in principle, from single-channel studies. However, to date, membrane
Em
+50
2 pA
100 mV
343
models have been formulated from experiments conducted

with macroscopic membranes. Illustrative of this approach is
the now classical work of Hodgkin and Huxley (2), who constructed the first mathematical model of excitable membrane
behavior in the early 1950s. This model was so successful that
it continues to be utilized today, although we now have improved models for cardiac tissue and for myelinated nerve.
In the 1950s electrode sizes were much larger than today.
To accommodate a larger electrode, Hodgkin and Huxley
chose the giant squid axon as their preparation. This cell has
a diameter of approximately 500 m and was large enough to
accommodate an axial electrode. Their second electrode was
a concentric cylinder, which was placed in the extracellular
electrolyte (sea water). This configuration insured no axial
variation in potential so that the entire cell membrane behaved synchronously (i.e., like a large number of membrane
patches in parallel). They described this arrangement as
space clamped. From their studies Hodgkin and Huxley concluded that the transmembrane current was essentially carried by sodium and potassium ions (chloride being at or close
to equilibrium). But while a patch electrode is capable of measuring a single ion current, the transmembrane current in the
whole axon preparation would necessarily contain both sodium and potassium contributions. To separate these they implemented the voltage clamp.
To set the stage for the Hodgkin and Huxleys experiments
we first describe the basic model they chose. As weve already
noted they assumed that the transmembrane current basically consisted of sodium and potassium. Recognizing the contributions from both electric field and diffusion they assumed
the relationships introduced earlier, namely that
Potassium:
IK = gK (Vm EK )
(11a)
INa = gNa (Vm ENa )
(11b)
Em
Sodium:
where the conductivities gK and gNa were expected to be functions of time and transmembrane potential. In Eq. (11) EK
and ENa are the potassium and sodium equilibrium (Nernst)
potentials, the difference from Vm being the net ion driving
force. IK and INa are ensemble current densities per unit area
of memberane. To account for a small additional leakage current they added
(a) Unitary K currents
1 pA
I1 = g1 (Vm E1 )
10
20
30
Time (msec)
40
(b) Ensemble average

Figure 5. Patch-clamp recording of single-channel potassium currents in a squid giant axon for a voltage clamp of 50 mV. Nine
consecutive trials are shown in (a). Based on a driving voltage of
100 mV and an open channel current of 2 pA, a channel conductance of 20 pS results. The recordings are low-pass filtered with a
cutoff of 2 kHz. The ensemble mean of 40 trials is given in (b). T
20 C. Data provided by F. Bezanilla and C. K. Augustine. Figure
from B. Hille, Ionic Channels of Excitable Membranes. 2nd ed., Sunderland, MA: Sinauer Assoc., 1992.
(11c)
where gl is an experimentally determined constant and El is

chosen so that the total ionic current reduces to zero at rest.
Although Hodgkin and Huxley could only guess at the
presence of separate single channels of potassium and sodium, their model is consistent with current understanding.
To fit their measurements they assumed that the potassium
conductivity satisfied
gK = gK n 4
(12)
We may interpret this to describe a functional potassium

channel with four subunits, each of which must be open for
the channel to be open (hence the probability of an open channel is n4 if n is the probability of an open subunit). Because n
describes a probability, then 0 n 1, while gK is the largest
344
possible value of potassium conductance (all potassium channels open). The temporal behavior of the probability n was
assumed to follow Eq. (10) where the rate constants and
depend solely on Vm.
From their measurements Hodgkin and Huxley learned
that sodium showed second-order kinetics. To deal with this
they let
gNa = gNa m h
3
(13)
and considered m as an activating gating variable and h an

inactivating one where each is governed by the same firstorder process described in Eq. (10), namely
dm
= m (1 m) m m
dt
(14)
dh
= h (1 h) h h
dt
(15)
In these expressions we have 0 m 1 and 0 h 1 while

the s and s depend only on Vm.
With the use of feedback electronics, Hodgkin and Huxley
applied a step transmembrane voltage (away from rest) to
their space-clamped squid axon. The response to a constant
voltage was useful in several ways. For one, the capacitive
current, CdVm /dt was zero (except for a brief transient) and
did not confound the measured transmembrane current. Secondly the n(t), m(t), h(t) can be easily obtained from solutions
of Eqs. (10), (14), (15) since the and coefficients, which are
functions of Vm, are constants during a voltage clamp. To illustrate, the measured gK(t) determines an n(t) from Eq. (12)
and, from the solution to Eq. (10), this must have the form
n(t) = n (n n0 )et/ n
(16)
where n n /(n n) and n 1/(n n). The initial value

of n n0 is obtained from the n of the rest period. From a
curvefit of gK(t) for a clamped voltage Vm, n(Vm) and n(Vm)
are obtained and from these (Vm), (Vm) are found. Measurements over a range of voltage clamp values of Vm give corresponding values of the gating variables and ; Hodgkin and
Huxley used these sample values to define a continuous
function.
It remains only to explain how Hodgkin and Huxley separated their measured transmembrane current into its sodium
and potassium components. This was done by running a pair
of experiments for each voltage clamp. In the first they used
a normal extracellular composition. Then this was repeated
with the same value of voltage clamp but using a low-sodium
extracellular medium (they replaced 90% of the sodium with
choline, because the latter is nonionizing but retains isotonicity). This results in a changed sodium Nernst potential and,
hence, driving force. They assumed no other changes and that
only the sodium current would change in proportion to its
change in driving force. Thus the following two equations
were obtained at time t, namely for the first experiment
Im (t) = INa (t) + IK (t)
(17a)
where Im(t) is the total measured transmembrane current as

a function of time, and for the second trial (primed notation)
Im (t) = INa (t)

Vm ENa
+ IK (t)
Vm ENa
(17b)
Note that IK(t) and INa(t) appearing in Eq. (17b) are assumed
the same as in Eq. (17a). From these two equations, the two
unknown values of INa(t) and IK(t) are obtained.
The continuous functional expressions chosen by Hodgkin
and Huxley to approximate their discrete measurements of
the s and s are given below:
n =
vm
0.01(10 vm )
, n = 0.125 exp
exp[(10 vm )/10] 1
80
m =
0.1(25 vm )
vm
, m = 4 exp
exp[(25 vm )/10] 1
18
vm
h = 0.07 exp
,
20
h =
(18)
(19)
(30 vm )
exp
+1
10
1
(20)
In these expressions vm is the transmembrane potential variation from the resting value, that is vm Vm Vrest, so that it
reflects the true signal apart from a dc component. The numerical values in Eqs. (18) to (20) are based on vm in millivolts.
The total transmembrane current per unit area is evaluated by summing Eqs. (11a) and (11b) and adding the capacitive component namely cmdvm /dt where cm is the specific capacitance in farads per unit area. The capacitance has
already been noted to correspond to the physical membrane
structure and is consequently fairly uniform among various
membrane types. For the squid axon, and many other membranes, it equals 1.0 F/cm2. As might be expected it does not
vary with time or transmembrane voltage. A complete expression for membrane current density, im, may thus be given as
i m = cm
dvm
+ IK + INa + Il
dt
(21)
The remaining values are needed to implement a Hodgkin

Huxley simulation:
gK = 36 36 mS/cm2 , gNa = 120 mS/cm2 , g1 = 0.3 0.3 mS/cm2
(22)
Although the system of equations given by Hodgkin and
Huxley were derived from voltage-clamped conditions, they
successfully describe general temporal behavior very closely.
In particular a correct subthreshold and transthreshold response is quantitatively described. Many observed electrophysiological phenomena such as anode break, refractoriness,
and so on, are correctly and quantitatively simulated. To implement these equations fast computers and numerical techniques are normally needed. The methodology may be found
in one of numerous current references (7,8).
The HodgkinHuxley equations describe ionic flow based
on the unequal composition of extracellular and intracellular
regions; the resulting flux is entirely due to the stored energy
associated with the concentration gradient (i.e. a passive response). But the membrane also contains a sodium
potassium pump. This is an active process that maintains normal composition by transporting sodium out and potassium
into the cell. This active transport exactly compensates for
the passive flux of sodium into and potassium out of the cell
during an action potential and at rest. The pump requires
energy to drive ions against their electrochemical gradient
and this is provided by high energy phosphates (ATP). The

inclusion of the Na-K pump in our model is important not
only to adequately describe axon electrophysiology over a long
time interval (when otherwise the system would run down)
but also because the pump exchanges three sodium for two
potassium and hence adds to the total transmembrane current (though, on average, usually a relatively small amount).
It is nevertheless electrogenic.
The HodgkinHuxley model has been applied successfully
to tissue other than squid axons. For myelinated nerve it is
applied at the Nodes of Ranvier, assuming the myelin sheath
to insulate the intracellular from extracellular space elsewhere. It has also been used in the simulation of activation of
striated muscle. However in both aforementioned applications
the modification introduced by Frankenhaeuser (9) is generally more satisfactory. For cardiac muscle the action potential
differs from the aforementioned by a very long plateau and
slow recovery (each phase lasting for roughly 100 msec). This
plays an important functional role in protecting the heart by
introducing a long refractory period and hence inhibiting the
re-entry of excitation (since activation can be present for perhaps 60 msec). Present day electrophysiological models of the
cardiac action potential (10) differ considerably from the simple HodgkinHuxley model in that they contain as many as
11 current sources. The unique plateau arises from a delicate
balance of component currents which are capable of adapting
to changes in the heart rate. These models also include the
calcium ion flow; this ion is important since it contributes to
maintaining the plateau and also since it triggers the coupling from electrical to mechanical activity of the heart
muscle.
Electrical Stimulation
If a very low amplitude pulse of current is passed between
the inner and outer electrode of the HodgkinHuxley spaceclamped preparation the equations describe a membrane response that corresponds to a passive resistancecapacitance
(RC) parallel circuit. The membrane resistance is essentially
its resting value, which can be found by evaluating n, m, and
h when vm 0 and then introducing this into the expressions
for gK(0), gNa(0), and gl, whereupon Rm A[gK(0) gNa(0)
gl]1, where A is the total membrane area in square centimeters. The associated capacitance is simply Cm A(1.0)F.
Simple electric circuit theory describes the membrane response to the current pulse, I0, as
vm = I0 Rm (1 et/ m )
(23)
where m RmCm is the membrane time constant.

If the pulse amplitude is increased, a point will be reached
where an action potential is elicited. This marks the membrane threshold. It is frequently assumed that this is a fixed
value arising from the intrinsic membrane excitability. Assuming this to be so, then Eq. (23) could be used to find combinations of current strength and duration that would result
in an action potential. Thus calling VT the threshold voltage
and using Eq. (23) we have the threshold current as a function of duration, T, namely
I0 (T ) = IT (1 eT / m )1
(24)
345
where IT VT /Rm is the threshold current amplitude for T

. Equation (24) is called the strengthduration curve and is
seen to have a characteristic hyperbolic shape. For T we
note that the minimum stimulus amplitude is found (namely
IT VT /Rm), and this is described as the rheobase. For a stimulus of twice rheobase the duration, according to Eq. (24), is
Tc 0.6931m where Tc is referred to as the chronaxie.
An experimental determination of the strengthduration
curve can be performed on the space-clamped squid axon, as
described above. A similar experiment can be performed by
simulation using the above HodgkinHuxley equations. In either case the outcome will be similar and can also be interpreted as describing the behavior of a membrane patch. One can
verify that the assumption of a constant threshold is supported provided the duration satisfies 0.05 T 5 msec
hence validating Eq. (24) for this condition (a threshold of approximately 8 mV is found). Outside this range other factors
enter the functional threshold. For very small durations the
membrane needs to be brought to higher threshold voltages,
so that following termination of the stimulus, as the membrane charge leaks away and the voltage diminishes, an adequate transmembrane potential remains to continue the process of opening sodium gates (the time constant for m is in
tenths of a millisecond). For long-duration stimulii membrane
inactivation, reflected in diminishing values of h, needs to be
overcome with higher effective thresholds.
The above limitations regarding threshold apply only to
the membrane patch. However, electrical stimulation of interest normally involves extracellular electrodes placed near or
in extensive tissue membranes. In this case depolarizing voltages are also accompanied by hyperpolarizing voltages, both
of which may be distributed over a complicated geometry. Interestingly, even in such circumstances, experimentally determined strengthduration curves are still seen to follow a behavior not too different from Eq. (24). In fact a better fit often
arises with a simpler form of Eq. (24), obtained by approximating (1 eT/ )1 m /T valid for small T. This approximation can be made to hold also for large T by taking (1
eT/ )1 1 m /T giving
m
I0 (T )
=1+
IT
T
(25)
Equation (25) is known as the WeissLapique formula. Multiplying Eq. (25) by T and describing QT I0(T)T as the threshold charge results in
QT = IT (T + m )
(26)
which predicts a linear relationship with duration for the

threshold charge QT. This is, in fact, frequently seen; the
value of m is determined experimentally by curve-fitting
strengthduration measurements and is surprisingly found
to approximate the membrane time constant.
SOURCE-FIELD RELATIONSHIPS
In electrocardiography, electromyography, and so on, electrical signals originating in a specific excitable tissue is measured with extracellular electrodes. In fact the usual case is
that electrodes lie at the body surface (i.e., so that measurements are noninvasive) in which case a clear separation ex-
346
ists between the excitable tissues and the recording electrode(s). To simulate this situation quantitatively there are
two main considerations. The first is to find an engineering
(quantitative) description of the sources that are generated by
tissue activation. The second, based on such a source description, is to evaluate the currents that will flow in the surrounding passive volume conductor. One is particularly interested in the associated electrical potential field, which will be
sampled by the recording electrodes. These two goals are the
subject of this section.
ExampleLong Fiber in an Unbounded Volume Conductor
We begin by considering an excitable, infinitely long fiber lying in an unbounded volume conductor. This idealized model
could approximate a long skeletal muscle or squid axon fiber
in a volume conductor whose extent is large compared to fiber
dimensions (well be more precise about this later). We assume that a propagating action potential has been initiated,
so that at the moment a full spatial action potential vm(z) is
present on the fiber. Since fibers are very long compared to
their small diameter we may assume no intracellular radial
variation of current density or potential and consequently
that we can describe i(z) as the intracellular potential, a
function only of the axial coordinate, z. Ohms law links the
total intracellular axial current, Ii(z), with the intracellular
field, i(z)/z, according to
i (z)
= Ii ri
z
(27)
where ri is the intracellular resistance per unit length. (The

intracellular resistance per unit length, ri, can be evaluated
from the intracellular resistivity, i, and the fiber radius a,
using ri i /(a2).) Continuity requires that whatever current
is lost from the intracellular space appear as an outward
transmembrane current and consequently
im =
Ii
z
(28)
where im is the transmembrane current per unit length. Combining Eqs. (27) and (28) gives the classical linear-core-conductor expression namely
im (z) =
1 2 i (z)
ri z2
(29)
The starting point for HodgkinHuxley simulation of a propagating action potential is equating Eq. (29) with Eq. (21).
Since physiological fibers are very long compared to their
diameters, then for points in the volume conductor that are
well outside the fiber we can consider the transmembrane
current to arise from a line source on the fiber axis. A short
fiber element of length dz will behave like a point source of
current (imdz). Again, because fibers are normally very thin,
the fibers presence within the volume conductor may be ignored, and we may consider the current from the aforementioned element to flow into an unbounded conductor. Now the
current density from a point current source, I0, through a concentric sphere of radius R, where all lie in a uniform, unbounded conductor, is by symmetry simply I0 /(4R2). The
electric field at that point, e /R, from Ohms law is
I0 /(4eR2) where the extracellular medium has a uniform

conductivity of e. Integrating with respect to R gives the potential field generated by a point source as
e = I0 /(4e R)
= im dz/(4e R)
(30)
where we replaced I0 by the point source element imdz. Substituting Eq. (29) into Eq. (30) and integrating over z gives an
expression for the field of a fiber lying in an unbounded uniform volume conductor and which is carrying an action potential as
e =
a2 i
4e
2 vm 1
dz
z2 R
(31)
where ri (a2i)1. In Eq. (31) we replaced i by vm since for

an unbounded volume conductor, in view of the extensive
short-circuiting, e 0, and consequently vm i e i.
In the above source-field considerations the finite fiber radius has been neglected, an approximation that is reflected in
the one-dimensional integral in Eq. (31). We may anticipate
that quite close to the fiber it may be necessary to take into
account that the current is not emerging from an axial source
but from a finite cylindrical surface. In addition, each current
source element on the membrane surface does not actually
see an unbounded medium but rather one in which the fiber
itself is embedded. Later on we will obtain a rigorous solution
from which this approximate one can be examined.
In the derivation of Eq. (31) we assumed a steady current
source and that the electric field can be obtained as the gradient of a scalar potential function, that is, E , which
also arises from static considerations. Since biological sources
of interest are normally time-varying the aforementioned procedure is open to question. However, it has been shown that
for signals of physiologic origin quasistatic conditions are in
effect (11) and these validate the aforementioned expressions.
Fundamental Source-Field Concepts
Equation (31) illustrates the fundamental concept of sourcefield relationship for a simple preparation, but, in fact, this
also applies conceptually to distributed sources in multicellular tissues (heart, brain, muscle, etc.) which generate electrical fields in bounded nonuniform volume conductors (the human body). In Eq. (31), behaving like a point current source
is the quantity isourcedz, where
isource = a2 i
2 vm
z2
(32)
The line source density, isource, is thereby identified as a current source being generated by the propagating action potential vm(z). The relationship of isource and vm is specified by Eq.
(32) describing the source quantitatively. In turn isource will
generate an electric field in the surrounding volume conductor, and this is given by Eq. (31). The total field at any point
arises from a summation of contributions from every source
element Isourcedz.
To more clearly distinguish source points from field points,
since the same coordinate system is being utilized for both,
we choose here unprimed coordinates to describe the source
and primed coordinates for the field. We may therefore write
Eq. (31) as
e (x , y , z ) =
1
4e
p(x x ) i + (y (x, yy,)z)+ (z z )

source
2
2
2
dz
(33)
where we have written out

R=
p(x x )
2
+ (y y )2 + (z z )2
While an extension of Eq. (32) to include multicellular tissues must still be considered, it seems reasonable to expect
that an expression similar to Eq. (32) will arise except that
the source needs to be described more generally as a volume
source, Iv, in which case isource is a degenerate case when the
source can be considered to be one-dimensional. For a volume
source Eq. (33) generalizes to
1
(x , y , z ) =
4
Iv (x, y, z)
2
(x x ) + (y y )2 + (z z )2
dv (34)
where Iv is a volume source density. In the literature Iv is also

referred to as an applied or an impressed source.
An important feature of the field (x, y, z) can be found
by evaluating its Laplacian, which can be accomplished by
taking the Laplacian of both sides of Eq. (34). The Laplacian
operates at the field point [i.e., with respect to the primed
coordinates on the left-hand side of Eq. (34)], so on the righthand side of Eq. (34) the operator can be taken under the
integral sign because the latter is with respect to the unprimed coordinates. In fact, in the integrand of Eq. (34) only
R is a function of the primed coordinates, and one can show
that 2 (1/R) 4v where v is a unit (volume) delta function. Consequently (we now drop the prime notation)
fer (surely it cannot be zero everywhere since there is a

source)? We would find that, while S and its derivatives are
well behaved and, in fact, satisfy Laplaces equation everywhere, either S or its derivative behaves discontinuously in
crossing the source surface (the discontinuity depending on
the type of surface source).
One way of investigating the behavior of S is to evaluate
it as we proceed along a path that crosses KS at right angles,
as shown in Fig. 6. To facilitate the calculation, we have removed a very small circular disc from the source, where the
center of the disc is the point of intersection of the arbitrary
path across the source surface, as is also described in Fig. 6.
Now in the absence of the disc the integral in Eq. (36) is well
behaved everywhere (since R 0 everywhere), and both S
and its derivatives are also necessarily well-behaved because
there are no singularities. It only remains to investigate the
contribution to S from the circular disc. This disc is so small
that we may take it to be planar. If we choose the z axis to
lie on the disc axis with its origin at the disc, then the radial
cylindrical coordinate lies in the plane of the disc. Application of Eq. (36) gives
KS (0)
4
KS (0)
(
=
2
S (z) =
pa
0
(35)
which is Poissons equation. Even though we understand how

to obtain isource for a long, isolated fiber [namely by applying
Eq. (32)], we await subsequent material that describes the
evaluation of Iv for multicellular preparations. One can also
confirm that Eq. (34) does, indeed, apply.
(1)
r=
(x, y, z)
KS (S)
dS
R
(37)
+ z |z|)
2
;
;
;
;;
(x x)2 + ( y y)2 + (z z)2
Po
The field from surface sources have useful properties, which,

as we will see, correspond to the fields arising from excitable
cells and from volume conductor inhomogeneities. This subsection is consequently devoted to a description of surfacesource fields; their application will be given subsequently.
The field of a point source is described by Eq. (30). For a
surface source described by KS(S)mA/cm2, then KS(S) dS constitutes a point source and application of Eq. (30) gives
1
4
P(x, y, z)
Surface Sources and Field Discontinuities
S =
2
2 + z2
where KS(0) is the value of the surface-source density at the

disc origin and a is the disc radius. From an examination of
Eq. (37), we conclude that S is continuous across the source
surface; however S(z)/z is discontinuous. Specifically, from
(2)
Iv
2 =
347
(36)
Since a surface source occupies zero volume, the S obtained

from Eq. (36) should satisfy Laplaces equation everywhere.
But in a completely source-free universe would necessarily
be zero everywhere. In what way does the behavior of S dif-
(a)
(b)
Figure 6. (a) A surface source KS or double-layer lies in the arbitrary surface S. The two sides are denoted 1 and 2 and the positive
normal n is from 1 to 2. P is an arbitrary field point. (b) The behavior
of the field at P0 is examined by decomposition of S into a small source
disc centered at P0 and the remaining sources. The field of the latter
are well behaved at P0 and hence whatever discontinuities might be
present can be studied by examining the behavior of the disc field
alone. From R. Plonsey, The formulation of bioelectric source-field relationships in terms of surface discontinuities. J. Franklin Inst., 297:
317324, 1974.
348
Eq. (37) we get
Eq. (37) we now have

z
1 = 2

z
K
= S
(38)
where positive z is directed from 1 to 2.

This result could have been anticipated based on a physical argument. The current source gives rise to a current flow
in opposite directions from the source surface. The potential
must therefore rise to a peak value at the surface and diminish in both directions away from the sheet to give this expected gradient. So, a continuity of potential results; but, because the derivative of potential is oppositely directed, it is
discontinuous. We learn from Eq. (38) that the magnitude of
the discontinuity in the normal derivative equals the strength
of the surface source (divided by the conductivity) at that
point. This is an important equivalence that we subsequently utilize.
A second surface source also must be introduced, which
consists of a sheet of dipoles. As a brief review, the dipole
consists of two point sources of equal magnitude but opposite
sign, which are displaced a very short distance dl. If dl 0,
the fields of each source combine and cancel. Then we may
consider the dipole to be created by displacing the positive
source by the small displacement dl. The two fields no longer
cancel precisely by the change in the field from the positive
source. Applying Eq. (30) we have
dipole =
I0 (1/R)
4 l
(39)
The partial derivative in Eq. (39) is known as a directional

derivative and can be found using the gradient operator as
dipole =
I0
(1/R) dll
4
(40)
In spherical coordinates (1/R) aR /R2 where aR is a unit

vector from the source to the field. Equation (40) may now be
written
dipole =
aR p
4 R2
(41)
where p I0dl is the dipole moment (namely the product of

monopole strength, I0, and separation dl in the direction
dl/dl al.
A surface of dipoles is known as a double layer. If we assume such a source surface having a strength (dipole moment per unit area), then dS constitutes a dipole element.
As with the single layer its fields are well behaved everywhere except in crossing the source surface. Also, as before,
the discontinuity can be examined by considering the field behavior along the axis of a very small double-layer disc. In
place of Eq. (36) we now have
D =
1
4
a R dS
S
a
R2
(42)
where we use dS dS because the dipole orientation for a

double layer is everywhere normal to the surface. In place of
D (z) =
(0)
4
a
0
2z d
( 2 + z2 )3/2
(43)
where we utilized aR az z/ 2 z2 because the axis of the

disc is z. We note from Eq. (43) that the potential is antisymmetric on z. However the normal derivative is an even function of z and is continuous across the disc (across the surface).
By evaluating Eq. (43) we can confirm that

D
=
n 2
D |2 D |1 =
D
n 1
(44)
The vector double-layer strength is n where n is the unit

surface normal from 1 to 2.
Single Cell Sources
An arbitrary shaped single cell lying in an unbounded volume
conductor is described in Fig. 7. The intracellular conductivity
is i and the extracellular conductivity e. We assume that an
action potential has been initiated at a membrane site and
that it is propagating distally eventually to reach all points
on the cell surface. The transmembrane potential is consequently a function of position on the cell surface vm(S) (a twodimensional function); and this is a more general condition
compared with the one-dimensional single-fiber case considered earlier.
Because the membrane is extremely thin compared to all
other dimensions of interest we consider it as an interface
between the intracellular space [whose potential is designated i(S)] and the extracellular space [where e(S) designates the potential]. We let be the potential field generated
by the action potential in either space and define a new scalar
potential function namely
=
(45)
Now satisfies Laplaces equation everywhere, because, except for the membrane that we assume to occupy zero volume,
all space is source-free and passive (i.e., there are no volume
sources). Consequently, since is piecewise constant
2 = 2 = 0
o
i
o
i
(46)
n
Jm
Figure 7. The single cell lies in an unbounded volume conductor.

The positive surface normal and transmembrane current is outward.
In the figure the transmembrane potential vm equals the intracellular
membrane potential, i, minus the extracellular membrane potential, o. The intracellular conductivity is i and the extracellular
is o.
and also satisfies Laplaces equation. From Eq. (35), we

note that in the general case
2 = Iv
(47)
so that the solution for from Iv has the interesting property

that it does not depend on the conductivity (although Iv, or a
degenerate surface or point source, may so depend). For the
single active cell satisfies the following boundary conditions
in crossing the membrane
i S e S = i i S e e S = 0
(48)
an inequality that reflects both the nonzero transmembrane

potential and difference in intracellularextracellular conductivity. [Note that while the difference in Eq. (48) might be
zero at some points it is not identically zero.] A second boundary condition follows from the continuity of current across the
membrane. This gives
i
e
i
e
e
=
=0

n S
n S
n S
n S
(49)
The notation in Eqs. (48) and (49) emphasizes that the and
its normal derivative are evaluated at the membrane surface;
n is the outward membrane normal.
An examination of Eqs. (48) and (49) shows that the scalar
function , which satisfies Laplaces equation, has a continuous normal derivative across the membrane, but the function
itself is discontinuous. A comparison with Eq. (44) reveals
that the membrane behaves like a double layer, lying in the
membrane surface, whose strength is
=
D
2
D
1
= e i
(50)
a result that is obtained by associating region 1 in Fig. 7 with

the intracellular, whereas region 2 corresponds to extracellular. Based on Eq. (41)
= =
1
4
a R dS
S
a
R2
and substituting Eq. (50) results in

=
1
4
Z
(e i )
S
a R dS
R2
(52)
As noted earlier we see that the conductivity is absent from

the source-field expression for the scalar function ; its
source-field relationships appear as if the medium were uniform and homogeneous [but, of course, the discontinuity in
conductivity at the membrane interface influences the source
strength e i according to Eq. (50)]. By applying Eq.
(45) to Eq. (52) the desired result is obtained, namely
p =
1
4 p
Z
(e e i i )
S
S
a R dS
R2
membranes if its resistivity (which is high) and thickness

(which is very small) is taken into account the result given by
Eq. (53) is, nevertheless, unchanged. Thus, for a multicellular
preparation consisting of cells of arbitrary shape, there will
be a double-layer source lying in each cell membrane whose
strength and orientation is given by
n
= (e e i i )n
(53)
In Eq. (53) we designated the field point with the subscript p

so that if it is intracellular (or extracellular) we substitute
i(e). Although we obtained Eq. (53) by ignoring the finite
thickness of a membrane, it can be shown that for biological
(54)
where n is the outward unit vector normal to the cell. We

designate these sources to be secondary sources. This is because a true primary source should be directly associated with
a source of energy. Here, we note that in Eq. (54) the source
strength depends significantly on the discontinuity in conductivity. We show in the following that a source does arise at
conductivity discontinuities, but in view of its passive nature
must be secondary. A detailed analysis can be found in
Plonsey (12). What is important, here, is that a rigorous
equivalent source description is that of a distribution of surface double layers.
Inhomogeneous MediaSecondary Sources
Any actual volume conductor normally contains several tissues and hence will be necessarily inhomogeneous. A first approximation is to consider each tissue or organ to be uniform,
so that the volume conductor has a conductivity that is
piecewise constant. The boundary conditions at the interface
between any two regions of different conductivity is based on
the continuity of potential and the continuity of current flow
normal to the interface. This is written as
1 = 2
1 /n|1 = 2 /n|2
(55)
The function , defined in Eq. (45), then satisfies the following boundary conditions
2 1 = (2 1 ) = 0
/n 2 /n 1 = 0
(51)
349
(56)
where we designate 1 2 at the interface [applying Eq.

(55)]. As shown above, satisfies Laplaces equation; its continuous normal derivative and discontinuous function at the
boundary can be considered to arise from a double layer at
the interface. The strength of the double layer is given by the
discontinuity in at the interface, which, from Eqs. (44) and
(56) amounts to
= (2 1 ) Sn
(57)
where n is the normal vector from side 1 to 2. This source

meets the criteria to be a secondary source, because it comes
into existence only when a field is first generated by a primary source. In a volume conductor the presence of an electric
field gives rise to conduction currents whose Joule heating
necessitates the continual influx of energy to maintain the
quasistatic system. This expenditure of energy comes from
the primary fields, which can be recognized by their coupling
to energy sources such as ATP. (An example is the membrane
ion pumps, which maintain the ion composition responsible
for membrane ion flow and hence identify the latter as a primary current source.)
350
In principle we can apply this treatment to a complex inhomogeneous volume conductor. Along each interface between
regions of different conductivity Eq. (57) is used. The inhomogeneous volume conductor is, in this way, replaced by a uniform homogeneous medium for the scalar function [as discussed in connection with Eq. (47)] except that at all
interfaces a double layer of the form given in Eq. (57) exists.
Clearly the result is a scalar function, , satisfying Laplaces
equation and all boundary conditions [in view of Eq. (44)];
the resultant is necessarily the correct solution. Assuming
multiple cells and an inhomogeneous medium the previous
results can be summarized as
1
4
XI
i
(e i )
cells
S
a R dS
1
+
R2
4
XZ
2 1
Sj
S
a R dS
R2
(58)
source coordinates (for simplicity in notation we now assign

unprimed coordinates to the source). Now we extend our
definition of Vf to
Vf (, z) = Vf (a, z) = Vf (z)
(where a is the fiber radius), so that whereas Eq. (60) limits

Vf (z) to the membrane surface [because the right-hand side of
Eq. (60) is defined only at that surface], in Eq. (62) this function is defined throughout the fiber volume, although it retains its assigned value at the membrane surface. Consequently this new definition can be given to Vf in Eq. (61),
because the surface integral retrieves Vf (z) from Vf (, z) from
Eq. (62) and is hence unchanged. But now Gausss theorem
can be applied, converting the surface integral into a volume
integral (throughout the fiber volume) which is
1 i
(p) =
4 p
where i enumerates all cells and j all interfacial surfaces. Using Eq. (45) and solving for the potential at the point p
results in
p =
1
4 p
+
XI
(e e i i )
cells
Z
1 X
i
4 p
(j
Sj
S
a R dS
2
R
S
a dS
j ) R 2
R
(59)
Long Cylindrical FiberEquivalent Sources

In our earlier treatment of the field generated by a single active fiber in an unbounded conductor, approximations were
made that limited the result to extracellular field points sufficiently far from the fiber to justify the assumption that the
sources be considered localized on the axis. But, now, we have
in Eq. (53) an expression that is valid for all field points. This
section is devoted to the application of Eq. (53) to the earlier
long-fiber geometry. One goal is to quantitatively examine the
approximations that lead to Eq. (31). But, in the process, we
shall also arrive at other source-field formulations that are
mathematically equivalent but have different, possibly useful,
physical interpretations. Although all this may seem like too
great an effort to devote to one particular geometry, in fact
much excitable tissue consists of fibers or, as in the case of
cardiac tissue, may be treated as consisting of fibers.
We begin with the single cylindrical fiber as described earlier, where, as before, potentials at the fiber surface depend
only on z (i.e., axial symmetry is assumed). If we let
iVf (z) = e e (z) i i (z)
(60)
where e(z) and i(z) are extracellular and intracellular potentials at the membrane interface, then, using Eq. (53) we
have
(p) =
1 i
4 p
Vf (z)
S
1
R
S
dS
(61)
In Eq. (61) weve replaced aR /R2 by (1/R), where aR is from

source to field, consistent with operating on unprimed,

1
R
Vf (, z)
V
dv
(63)
Since Vf now participates in the volume integration, it is clear

that its definition within the volume [e.g., Eq. (62)] is required. If the vector identity (A) A A is applied to the integrand in Eq. (63), one obtains

Vf (, z)
where the conductivity in region 1 is designated with a prime,

in region 2 with a double prime, and p designates the conductivity at the field point p.
(62)
1
R
Vf (z)
z

az
1
R

+ Vf (z) 2
1
R
(64)
where we recognized that Vf (, z) is a function only of z and

furthermore that it equals Vf (z). We have commented earlier
[see Eq. (35)] that the Laplacian of 1/R is a delta function.
For exterior field points R 0 and the Laplacian is zero. Because our emphasis here is on extracellular fields generated
by cellular sources, this will always characterize our situation. Accordingly Eq. (63) can be written
(p) =
1 i
4 p
Z V (z)
f
V
az
1
R
dv
(65)
Incorporating Eq. (60) into Eq. (65) gives

e (p) =
1
4e
Z
z
[e e (z) i i (z)]
dz
z
Z
A
(1/R)
dA
z
(66)
where we have broken the volume integral into a cross-sectional integration and an axial integration.
Equation (66) is a new source-field relationship that evaluates the extracellular field from sources arising from the propagating action potential on a single fiber in an unbounded
volume conductor. Using Eq. (41) we can identify the crosssectional integral [including the coefficient 1/(4e)] as evaluating the field of a unit magnitude double-layer disc (with the
fiber radius a) oriented in the z direction. If z is the source
location the result of the cross-sectional integration can be
written Wd[, (z z)] where (, z) are the coordinates of the
field point. The quantity in Eq. (66) namely (z) [ee(z)
ii(z)]/z constitutes the double-layer (disc) density (a function of z). So, we also can write Eq. (66) as
e ( , z ) =
(z)Wd [ , (z z )] dz
z
(67)
a convolution integral that can be evaluated using the fast

Fourier transform (FFT). For extracellular field points, Eq.
(66) identifies the source as a volume dipole distribution that
fills the intracellular space of the fiber. Such a source is not a
real source, it does not correctly give the intracellular field,
but is an equivalent source. It does give the correct field for
any z and for a, that is, for any extracellular field point.
Equation (66) can be integrated by parts, and this gives
two additional expressions. (Note that, in this integration, the
integrated parts go to zero because at the ends of a long fiber
field quantities are all zero.) The results are
e ( p) =
1
4e
Z
z
2
[ (z) e e (z)] dz
z2 i i
Z
A
1
dA
R
(68)
and
e ( p) =
1
4e
Z
[i i (z) e e (z)] dz
z
2 (1/R)
dA
z2
(69)
In Eq. (68) the cross-sectional integral is the field of a singlelayer disc lying in the fiber cross section. The extracellular
field is the convolution of this field with the source density
function given by
K(z) =
2
[ (z) e e (z)]
z2 i i
(70)
The extracellular field in this formulation arises from a volume current-source density Iv a2K(z) with K(z) given in
Eq. (70).
In Eq. (69) the cross-sectional integral can be identified as
the field from a disc of axial quadrupoles (a single such quadrupole consists of two axial dipoles displaced in the z direction). In this formulation the function we called iVf (z) is itself the source-density function. All the aforementioned
sources are equivalent sources and all give the same answer
in the extracellular region. Depending on Vf and the geometry
one of these may be particularly attractive either for simplicity in calculation or for its clear physical interpretation or
both. But, clearly, all formulations will give identical results.
For source-field distances that are large enough the singlelayer disc in Eq. (68) can be approximated by a localization of
the source on the axis. One can examine this approximation
by comparing the field of the disc with that of a point source
of the same total strengh at the disc origin. For R/a 5 the
error will be under 1% (13). Given this condition, we may replace
Z
A
a2
1
dA =
R
R
(71)
One can also show that for the unbounded volume conductor
e is very small and can be neglected. (Under these conditions
we can also replace i by vm.) Assuming both approximations
permits the conversion of Eq. (68) into
( , z ) =
a2 i
4e
Z
z
1 2 vm
dz
R z2
(72)
351
which corresponds to Eq. (31). In the derivation of Eq. (72)

however, we have a clearer picture of the underlying approximations.
Multicellular PreparationsCardiac Muscle and Bidomain
For a small bundle of muscle fibers that are approximately of
similar diameter and on which an action potential is propagating from the same origin, one could apply Eq. (66), (68), or
Eq. (69) to each fiber and sum their contributions to evaluate
the total extracellular field. (Fortunately physiological volume
conductors are linear and superposition applies.) If e at the
deep fibers is small, as is true at the periphery, an assumption that would be correct only for a small bundle (see Ref.
14 for details), then the total source is distributed uniformly
through the entire bundle (reduced by the usually small volume fraction occupied by source-free intercellular space). The
result is similar to a single fiber with the diameter of the
bundle (except that the actual action potential velocity and
spatial extent is determined by a single component fiber diameter and not the bundle diameter). For thick bundles an
analytic approach is needed (14), some of the groundwork for
which follows.
Cardiac muscle cells may be thought of as cylindrical with
a length of approximately 100 m and a diameter of approximately 12 m. These cells are stacked together leaving only
perhaps 15% of the volume for the extracellular (interstitial)
fluid. Cells are interconnected at many points with junctions
that have a relatively low resistance. The connections (called
connexons) are proteins that include an aqueous channel so
that, regarding ion movement, the intracellular space of all
cells are directly interconnected. Cells are organized into fibers, which, macroscopically, encircle the heart in a double
spiral. If such a tissue is activated at a point, propagation
spreads outward in all directions, although the velocity along
the fibers will be some three times greater than across them
(i.e., the wavefront will be ellipsoidal). But that propagation
away from the stimulus site is relatively uniform (a consequence of the many intercellular junctions) gives rise to the
view of cardiac tissue to be syncytial (continuous).
Following activation Eq. (53) can be applied to every cell
to establish the distribution of cardiac sources. At each membrane there is a double-layer given by Eq. (53). Because each
cell is relatively small (compared to a likely distance to field
points of interest) the single net resultant dipole arising from
the vector sum of the cells double-layer elements are a good
approximation to that distributed double-layer source. Furthermore, from the fact that there are perhaps 1010cells in the
heart, a distributed dipole volume density function (designated as Ji) can be determined. In evaluating this density
function at every point the contributions from a small volume
is summed and divided by the volume. Ideally the volume
should be very small (approaching zero), but in this case although cells are small their size cannot be ignored. So in evaluating the density function the volume should be small
enough for a reasonable resolution but not so small that there
are insufficient cells to obtain a satisfactory average. This is
referred to as a coarse-grain average. We assume that the
underlying physiological processes insure smooth spatial
transitions so that Ji is well behaved mathematically. The
field from this source is described using Eq. (41), as
e =
1
4
J i (1/R) dv
V
(73)
352
(because, as noted earlier, (1/R) aR /R2). The volume V in

Eq. (73) should include all sources, and this is assured by
taking the integral over the entire heart. We now apply the
following vector identity (A) A A to the scalar vector product (1/R)Ji and get
J i /R) = (1/R) J i + (1/R) J i
(J
(74)
Substituting from Eq. (74) into Eq. (73) yields
e =
1
4
Z
BidomainMathematical Description
J i /R) dv
(J
(1/R) J i dv
(75)
The first integral on the right-hand side of Eq. (75) can be

converted into a surface integral using the divergence theorem. Because the volume is the entire heart, this integral
evaluates to zero since at the surface of the heart Ji 0. Consequently, an alternate formulation to Eq. (73) is
e =
1
4
Z
(1/R) J i dv
(76)
An examination of the form of Eq. (76) [compare with Eq.

(34)] identifies Ji as a volume source density (with previous notation weve shown that Iv Ji).
Substituting the expression for Iv into Eq. (35) and rearranging results in
J i ) = 0
(J
(77)
In Eq. (77) the vector quantity in parenthesis is solenoidal;

therefore, we can identify it as the total current, Jt, which has
the appropriate zero divergence by virtue of the continuity of
current. Consequently
J t = J i
(it is actually the average over a small surrounding region),

and this value is a solution to the membrane equations at
that same point. That is, the membrane in the bidomain is
also distributed throughout the tissue space providing a link
between intracellular and interstitial domains. In summary,
we give up the fine detail for the simplification of a continuous medium where continuum mathematics may be applied.
(78)
showing that the total current density is the sum of the ohmic
(conduction) current, E , plus Ji. But whereas Ji was
introduced as a dipole source density it now appears as an
applied current density (both interpretations have the same
dimensions, of course).
Although cardiac muscle is actually made up of a collection
of discrete cells we have introduced Ji as a source function
that is continuous. The same simplification can be introduced
in regard to the cardiac tissue structure. The fine details of
this structure include the collection of cells where each cell is
connected to its neighbors by roughly ten junctional elements
(15). But on a global basis the intracellular space can be regarded as a continuum. A similar argument can be applied to
the interstitial space, which, although containing many convolutions on a subcellular scale, yet macroscopically may be
considered through an averaged, continuous medium. Such a
model is known as a bidomain since it consists of an intracellular and an interstitial (continuous) domain. In view of the
noted preferential propagation along fiber axes (reflecting
higher conductivity values in this direction) the cardiac bidomain can be expected to be anisotropic. A further simplification is to define the intracellular and interstitial domains on
the same tissue space. For a given (bidomain) coordinate an
intracellular and extracellular potential would be retrieved;
their difference is the transmembrane potential at that point
Possibly a more satisfactory description of the bidomain utilizes the language of mathematics. In each domain we require
that Ohms law be satisfied. Accordingly, letting i refer to intracellular and e interstitial, we get
J i = gix
and
J e = gex
i
i
i
a x + giy
a y + giz
az
x
y
z
e
e
e
a x + gey
a y + gez
az
x
y
z
(79)

(80)
In Eqs. (79) and (80) the conductivities are assumed to be

different in each of the principle directions. For cardiac tissue
normally one of the principle directions is along the fiber axis,
say z (for both regions), whereas in the transverse direction
we often expect gix giy and gex gey based on structural symmetry. The currents evaluated in Eqs. (79) and (80) are constrained by the continuity of current, so that whatever leaves
one domain must appear in the other (except when current is
introduced or removed with an electrode) and thus
J i = J e = Im
(81)
where Im describes the transmembrane current per unit

volume.
If the microscopic intracellular conductivity is i and microscopic extracellular conductivity is e, then the bidomain
conductivities can be estimated from the tissue structure. Let
us assume that there is a uniform fiber orientation in the z
direction where the relative cross-sectional area occupied by
the intracellular region is p, and hence that occupied by the
interstitial region is (1 p). In this case wed get
giz = i p
gez = e (1 p)
(82)
Here we have taken into account that the actual relative intracellular cross-sectional area is p 1, whereas in the bidomain, because the full tissue space is occupied, it is now
raised to 1 and hence the bidomain conductivity must be proportionately reduced. A similar argument applies to the interstial bidomain conductivity. For the transverse directions
the geometrical factor is more difficult to evaluate. For circular cylindrical fiber arrays the transverse interstitial conductivity (16) has been found to be
gex = gey =
1 p
e
1+ p
(83)
In view of the complex structure of cardiac tissue experimental determination of bidomain conductivities is normally re-
quired. However there are only two such investigations, at

present, and these have significant disagreements (17,18).
353
(86) the volume fraction of cells, designated p, is included

since the equivalent sources only lie within the cellular volume as pointed out in Eq. (66).
ElectrodesReciprocity
We have focused attention on the evaluation of volume conductor fields from sources in excitable tissues. If the potential
field is evaluated at the surface of the body then a pair of
small electrodes will sample this field and record the difference in potential. But if the electrode is large compared to
spatial variations in potential then the measured potential is
an averaged value. How should the average be taken? It can
be shown that relative to a remote site an electrode with a
conducting surface Se lying in an (applied) potential field a
(in the absence of the electrode) measures a voltage, V, (19)
given by
V =
a Jr dS
(84)
Se
where Jr is the volume conductor current density at the electrode surface Se that arises when a unit current is put into
the electrode and removed at the remote reference. This reciprocal current, Jr, behaves as a weighting function. For a
spherical or circular cylindrical electrode Jr may be uniform
and the weighting similarly uniform. For a surface electrode
one expects an increased weighting to arise near the edges.
Another useful formulation that is helpful in considering
the signals measured by an electrode comes from the application of reciprocity. Consider a bounded volume conductor of
volume V at the surface of which are placed two recording
electrodes, a and b, which yield a measured voltage Vab. The
source is described by a volume distribution Ji within V. The
reciprocity theorem states that
Vab =
J i r dv
(85)
where r is the potential field arising from the introduction

of a unit current into electrode a and its removal from electrode b. This reciprocal potential field is associated with a
current density Jr r where is a conductivity function
of position [i.e., Eq. (85) applies to an arbitrary inhomogeneous volume conductor]. In electrocardiography Jr is defined
as the lead field of electrodes a and b, one of which may be a
reference electrode. Equation (85) provides an interpretation
of the measured voltage as a weighted average of the dot
product of the dipole volume source density with the lead vector field (Jr). One can seek to emphasize or deemphasize certain source regions or to emphasize no region (uniform sensitivity), but since r necessarily satisfies Laplaces equation
there are severe limitations on what can actually be done.
The bidomain model provides a suitable basis for evaluating the aforementioned sources given intracellular and interstitial potential distributions. A mathematically well-behaved
gradient during both activation and recovery within a local
region (consisting of a small number of cells within which uniformity can be assumed) permits setting, based on Eq. (66),
J i = p(e e i i )
(86)
Appropriately since Ji is an averaged dipole moment density

it is derived from averaged (bidomain) fields (e, i). In Eq.
APPLICATIONS
This section is devoted to a presentation of four systems in
which electrical signals arising from specific organs are recorded; the goal in each case is to obtain information about
the sources of each signal. The aforementioned systems are
the electrocardiogram (ECG), electromyogram (EMG), electroencephalogram (EEG), and electrogastrogram (EGG). Each is
treated in detail as a separate chapter in this volume; the
consideration here is limited solely to a discussion of sourcefield relationships introduced in this chapter. Our interest
centers on the quantitative evaluation of sources and on pertinent aspects of the volume conductor for each of the aforementioned systems. This introduces more advanced material
as well as illustrate the application of the earlier material of
this chapter.
Electrocardiography
Information on the electrical activity within the heart itself
comes, mainly, from canine studies where multipoint (plunge)
electrodes are inserted into the heart. The instant in time
that an activation wave passes a unipolar electrode is marked
by a rapid change in potential (the so-called intrinsic deflection) and, based on recordings from many plunge electrodes,
it is possible to construct isochronous activation surfaces. The
cardiac conduction system initiates ventricular activation at
many sites nearly simultaneously and this results in an initial broad front. The syncytial nature of cardiac tissue appears to result in relatively smooth, broad, activation surfaces, and because fibers lie parallel to the endocardium and
epicardium the anisotropy insures wavefronts to also lie parallel to these surfaces.
The temporal cardiac action potential has a rising phase of
approximately 1 msec, followed by a plateau of perhaps 100
msec and then by slow recovery, which also requires approximately 100 msec. Because activation is a propagated phenomena, a spatial action potential can be obtained from the temporal version since the space-time function must be of the
form of a propagating wave vm(s t) where s is the local direction of propagation and the velocity. Thus behind the
isochronal wavefront is a region undergoing depolarization
(for 50 cm/sec and a rise time of 1 msec its thickness is
0.5 mm (it is hence quite thin and often approximated as a
surface). Behind that, the tissue is uniformly in the plateau
state while ahead of the activation wave it is uniformly at
rest. Application of Eq. (86) shows that there are no sources
except in the region undergoing activation (the gradient being
zero wherever vm is unvarying). Thus the activation source is
a volume double layer with a thickness of around 0.5 mm lying behind the activation isochrone. The total strength of the
double layer is given by integrating Eq. (86) in the direction
of propagation; using the bidomain model this come out
= (Vpeak Vrest )
re
ri + re
(87)
354
where ri and re are bidomain intracellular and extracellular

resistance per unit length in the direction of propagation and
Vpeak and Vrest describe the peak and resting values of the action potential. The value of in Eq. (87) has been found from
the component potentials and resistances and also from direct
measurement of the potential across an activation wave and
consistent values of 40 mV found (20). Because the activation wave is only 0.5 mm thick the double layer may be considered to lie in the surface corresponding to the activation
isochrone.
That activation sources are limited essentially to a surface
is a consequence of an activation time of 1 msec. Recovery,
on the other hand occupies 100200 msec and consequently
recovery sources are distributed throughout the heart. To
make things even more complicated, recovery is not propagated (although cells undergoing recovery can and do influence neighboring cells). Of course Eq. (86) continues to apply,
but the spatial distribution of potentials now depends on spatial variations in waveshape. Assuming that cells recover earlier at the epicardium than at the endocardium would result
in equiphase surface propagation from epicardium to endocardium, and hence the dipole density found from Eq. (86) which
is outward during activation is also (on average) outward during recovery (and this would account for the observed upright
QRS and T waves). Although action potential morphology can
be readily examined at the epicardium and endocardium
(with good resolution using optical or microelectrode techniques) in vivo intramural action potential waveforms are not
accessable (although aspects, like refractory period, can be
sampled).
In vitro and isolated cell electrophysiology, although less
reliable quantitatively (since cellular interactions are abnormal), reveal that the variation in action potential duration
from endocardium to epicardium is not monotonic, as assumed above. Recent work of the Antzelevitch group (21) describes a mid-wall region containing M cells which have the
longest action potential durations. Consequently the T-wave
sources are more complex in distribution and orientation.
While they are not uniformly outward, indeed they appear to
be inward in the subendocardial region, the collective dipole
source direction is outward.
In connection with recovery, interest over the years has
developed in the time-integrated electrocardiogram. This can
be interpreted as the algebraic area of the QRS and T waves
and is consequently designated AQRST. For the jth lead, with
lead vector field lj(v), it has been shown that, based on Eqs.
(85) and (86) (22)
Z
AQRST
= C
j
l j dv
(88)
heart
where is the area of the action potential (a function of position) and the volume integral in Eq. (88) is taken throughout
the heart. If the cardiac action potentials all had similar
shapes but the duration of the plateau was a variable (possibly this is the leading difference in morphology), then
Z
AQRST
j
= C
d l j dv
(89)
heart
where d is the action potential duration (23). The dependence

of the integrated electrocardiogram on the recovery gradient,
described in Eq. (89), led to its designation as the ventricular

gradient. Dispersion of recovery has been linked to a propensity for arrhythmias; so the ventricular gradient has been examined as a possible evaluative tool.
We have concentrated most of our attention on cardiac
sources, but to complete a forward simulation one must also
consider the volume conductor. This is clearly inhomogeneous
the most important inhomogeneity being the finite torso itself. Other components are the blood cavities, the lungs, and
the surface muscle layer. The latter is anisotropic but is usually taken into account by increasing its thickness by a factor
of three. Assuming that each tissue is uniform limits the secondary sources to the various interfacial surfaces. This formulation lends itself to a forward solution by the boundary element method (BEM). A number of studies have appeared in
the literature mostly demonstrating inhomgenieties to be of
importance, although the effect is more pronounced on the
quantitative body surface potentials and less on their temporal and spatial potential patterns (24,25).
Electromyography
In electromyography the source arises from action potentials
propagating in whole muscle. Muscles of greatest interest are
those at the extremities. Potentials may be measured noninvasively at the body surface or minimally invasive with electrodes inserted into the muscle itself with a hypodermic needle. The latter electrodes may be macroscopic (from the
uninsulated portion of the shaft of the needle), or from a small
wire electrode whose tip lies within a very small hole in the
needle wall (called a side-arm), which is useful in recording
single-fiber EMGs (SFEMG). Needle electrodes are also available with multiple holes to achieve multiple leading off points
that may be recorded simultaneously. In addition a single
concentric electrode is also available that is sensitive to only
a few fibers or possibly an SFEMG. The clinical goal is to
evaluate pathologies such as atropic or hypertropic fibers or
changes in fiber distribution from abnormalities in the EMG.
For the macro-EMG, as noted above, one channel is derived from the needle electrode canula (with an uninsulated
tip 15 mm long). The sideport is located 7.5 mm from the tip
and contains a single electrode 25 m in diameter. This records an SFEMG. It may be used to trigger the canula signal,
which when averaged over a succession of signals selects the
activity of a single motor unit (that associated with the
SFEMG). Accordingly the canula signal, which ordinarily reflects many motor units, may then be said to yield a macro
motor unit potential (MMUP).
Simulation is a useful tool to investigate the properties of
EMG signals. For the MMUP a number of factors must be
considered. The motor unit contributing to the potential field
are fibers activated by a single motor neuron that may actually have one or more branches introducing possible latencies
between groups of inervated muscle fibers. In addition, muscle fiber endplates are dispersed resulting in a desynchronization of action potentials traveling on each fiber. A third factor
contributing to a variation among the individual fibers in a
motor unit is their difference in fiber diameter (and hence in
velocity which is usually assumed to be linearly proportional).
Finally there is the effect of fiber geometry within the motor
unit (the fiber density is not uniform) and the relative location
of the recording electrode.
Nandedkar and Stalberg (26) has suggested the use of a

line source model, such as described by Eq. (72). If we assume
that symmetry is axial the weighting function in cylindrical
coordinates is given as
Ws ( z ) =
4 K 2 + (z z )2
(90)
where K z / r. This result differs from 1/R in Eq. (33) because it is assumed that a single fiber in a muscle bundle lies
in an anisotropic monodomain medium, which can be described by the radial, axial conductivity parameters r, z. A
further weighting arises from the large canula surface area,
and this necessitates a surface integration of Eq. (90) of the
type shown in Eq. (84), the result of which we designate WS.
(As an approximation the canula may be considered uniform
and its surface integral estimated from a line integral giving
a simple average). The potential function is given by
e = im (z) WS
(91)
where denotes convolution and im ia22i /z2.

Because there are many active fibers in electromyography
it is of particular interest to know the extent of the sensitivity
of a particular electrode when inserted into the muscle. This
is designated the pickup area. It may be examined through
the application of reciprocity and a description of the lead
field, as pointed out with Eq. (85).
For simulations of the EMG at the surface of a limb the
volume conductor comes into consideration. To date this has
been assumed to be uniform in conductivity and only the
boundary with air interface acknowledged. The simplest
model is one that assumes the bounding tissueair interface
to be semiinfinite in extent. This permits a consideration of
the contribution of the secondary sources by the method of
images. A better approximation is to treat the limb as circular
cylindrical. Simulations using the latter model demonstrate
the shortcomings of the former, particularly for deep fibers (27).
355
gradient of transmembrane potential exists. For nerve cells

this occurs when an action potential has been elicited and
also when synaptic potentials are developed. In addition
transmembrane potentials can also arise from the passive
flow of current across membranes (as is expected in cardiac
and skeletal muscle tissue). The effect of this factor may instead be approximated by including the passive membrane in
an averaged specification of the volume conductor impedance
(e.g., by defining an appropriate anisotropic bidomain).
The response to sensory stimulation and some types of pathology is a region of cellular activity that often is circumscribed ( focal). In this case we may assume that the source
(were it in an unbounded region) would generate the field,
based on Eq. (59), given by
1 aR
=
4 R2
"
XZ
Sj
(e e i i ) dS
(92)
requiring a summation over all cells Sj. Equation (92) neglects

the distribution of cellular sources, an approximation that improves as the region diminishes in size. Defining a continuous
volume dipole source function, Ji, as was done leading to Eq.
(73), permits the simplification of Eq. (92) into
=
1 aR
4 R2
Z
J i dv
(93)
In order that Ji not average out to zero, synchronization of

sorts appears to be required. The result, according to Eq. (93),
is a single focal dipole source. A number of articles have been
directed to determining such a source (its magnitude, direction, and location) (28).
Whether or not the dipole density distribution can be considered to approximate a single dipole, the measured surface
voltage can be interpreted with the application of reciprocity.
According to Eq. (85) a distributed source Ji contributes to
the lead voltage Vab according to
Vab =
J i r dv
(85)
Electroencephalography
The electrical activity of the brain can be detected using scalp
electrodes; its spontaneous sources generate the electroencephalogram (EEG). Such sources arise from a large number
of cells. However, in contrast with the ECG and EMG, the
EEG apparently does not significantly involve action potentials of neuronal cells but rather is associated with the synaptic activity that precedes or follows activation. Actually EEG
sources are believed to arise mainly from postsynaptic potentials (PSP) on cortical pyramidal cells. A large number of cells
appear to underlie a detectable signal, and this implies temporal (and spatial) synchronization. If invasive intracellar recording is undertaken, both spike (action potential) and wave
type activity is found; abolition of the latter terminates the
EEG. In addition to the aforementioned spontaneous activity
giving rise to the EEG as measured at the scalp, one can also
obtain signals that are a response to stimuli using auditory,
visual, and tactile modalities. These are described as evoked
potentials.
As we learned from Eq. (86), a region consisting of excitable cells will establish an electrical source provided that a
In the present context included in Ji are the neuronal sources

arising from action potentials as well as synaptic potentials.
The reciprocal field must, at a minimum, include the influence of the brain tissue, the skull, and the scalp. In view of
the large differences in conductivity the reciprocal field actually available at brain sources is quite different from what
would arise were aforementioned regions homogeneous. If the
skull resistivity is taken as 80 times that of the brain and
scalp, then the lead field is diminished 20-fold (a measure of
the shunting effect of the scalp-skull) based on a concentric
spheres volume-conductor model (29).
Electrogastrography
Electrogastrography (EGG) refers to the measurement of electrical signals at the surface of the abdomen whose sources lie
in the smooth muscle of the stomach. The system is thus basically the same as in the previous applications. Internal gastric electrical activity (GEA) have also been investigated, but
because such measurements are truly invasive there is only
limited human data. However even these measurements are
356
macroscopic; intracellular data is even harder to obtain. As a

consequence, what is available about the electrical activity of
the smooth muscle is inadequate for a quantitative evaluation
of sources, such as described for the EMG and ECG. In its
place are empirical dipole source models, which are then combined with a volume conductor model.
The stomach is bean-shaped with food entering from the
esophagus at the top passing through the fundus and body
regions into the lower region (the antrum), from which it then
passes through the pylorus and enters the duodenum (small
intestine). The stomach is lined with a longitudinal muscle
layer at the outside and within which is a circular muscle
layer. Electrical activity starts at a pacemaker region somewhat near the fundus and travels distally via the longitudinal
muscle. Activity spreads inward from the longitudinal muscle
through the thicker circular muscle. A background electrical
activity (electrical control activity, ECA) is always present; a
second component, which it may trigger, is the electrical response activity (ERA). The ERA is associated with contractions and spiking (action potentials). Because it is noninvasive, the EGG is of interest in gastroenterology, but the signal
is low (100500 mV) and must be separated from other bioelectric signals and noise (motion and respiration sources).
The EGG does not present a recognizable morphology, as
with the ECG, and tends to be quite noisy. Accordingly the
clinical approach is to evaluate only three parameters,
namely amplitude, frequency, and time duration. The use of
the FFT provides a useful measure of the signal frequency
and this can be implemented with a running window to provide a view of the spectrum as a function of time. (The slow
wave noted above shows a frequency of around 3 cycles/min.)
The EGG has been examined for a determination of gastric
motility, but surface recordings of such activity have proven
unreliable (such information can be found with electrodes
placed directly on the stomach yielding gastric electrical signals). In fact the recent study by Mintchev and Bowes (30)
concludes that only the frequency dynamics forms the basis
for quantitation.
There are few biophysical models of the EGG and what is
available is ad hoc. The most recent of these is due to Mintchev and Bowles (31). The human stomach is represented as
a truncated conoid. The sources are assumed to be an annular
double layer with a width that is relatively narrow (around
23 mm) and with a dipole orientation normal to the surface.
Sources, as such, outside this band are assumed to have random orientations and to generate no net field. The source-field
equation of Mintchev and Bowles (31) is given as
Z
D )/ 3 ] dS
[(D
VQ = 1/(4)
(94)
where D is the doublelayer. Equation (94) is equivalent to Eq.

(73). The ECA is generated by the distal movement of the
double-layer band where the velocity along the greater and
lesser stomach curvatures are considered to be different. The
ERA is assumed to have no influence on the model or the ECA
pattern. Their model gave good predictibility, particularly on
the effect of electrode position, reflecting volume-conductor influence, although no effect of inhomogeneity is included in
the model.
BIBLIOGRAPHY
1. B. Hille, Ionic Channels of Excitable Membranes. 2nd ed., Sunderland, MA: Sinauer Assoc., 1992.
2. A. L. Hodgkin and A. F. Huxley, A quantitative description of
membrane current and its application to conduction and excitation in nerve, J. Physiol., 117: 500544, 1952.
3. D. E. Goldman, Potential, impedance, and rectification in membranes, J. Gen. Physiol., 27: 3760, 1943.
4. A. L. Hodgkin and B. Katz, The effect of sodium ions on the electrical activity of the giant axon of the squid, J. Physiol., 108:
3777, 1949.
ber die von der molekularkinetischen theorie die
5. A. Einstein, U
warme gefordertebewegung von in ruhenden flussigkeiten suspendierten teilchen, Ann. Physik, 17: 549560, 1905.
6. E. Neher and B. Sakmann, The patch clamp technique, Sci. Am.,
266: 2835, 1992.
7. J. E. Randall, Microcomputers and Physiological Simulation. 2nd
ed, New York: Raven Press, 1987.
8. J. Dempster, Computer Analysis of Electrophysiological Signals.
London: Academic Press, 1993.
9. B. Frankenhaeuser, Quantitative description of sodium currents
in myelinated nerve fibers of Xenopus laevis. J. Physiol. 151: 491
501, 1960.
10. C-H. Luo and Y. Rudy, A dynamic model of the cardiac ventricular action potential, Circ. Res., 74: 10711096, 1994.
11. R. Plonsey and D. B. Heppner, Considerations of quasistationarity in electrophysiological systems, Bull. Math. Biophys., 29: 657
664, 1967.
12. R. Plonsey, The formulation of bioelectric source-field relationships in terms of surface discontinuities, J. Franklin Inst., 297:
317324, 1974.
13. R. Plonsey, The active fiber in a volume conductor, IEEE Trans.
Biomed. Eng. BME-21: 371381, 1974.
14. C. S. Henriquez, N. Trayanova, and R. Plonsey, Potential and
current distributions in a cylindrical bundle of cardiac tissue, Biophys. J., 53: 907918, 1988.
15. R. H. Hoyt, M. L. Cohen, and J. E. Saffitz, Distribution and threedimensional structure of intercellular junctions in canine myocardium, Circ. Res., 64: 563574, 1989.
16. B. J. Roth and F. L. Gielen, A comparison of two models for calculating the electrical potential in skeletal muscle, Ann. Biomed.
Eng., 15: 591602, 1987.
17. L. Clerc, Directional differences of impulse spread in trabecular
muscle from mammalian heart, J. Physiol., 255: 335346, 1976.
18. D. E. Roberts and A. M. Scher, Effect of tissue anisotropy on extracellular potential fields in canine myocardium in situ, Circ
Res, 50: 342351, 1982.
19. R. Plonsey, Dependence of scalar potential measurements on
electrode geometry, Rev. Scientific Instrum., 36: 10341036, 1965.
20. R. Plonsey and A. van Oosterom, Implications of macroscopic
source strength on cardiac cellular activation models, J. Electrocard., 24: 99112, 1991.
21. D. W. Liu, G. A. Ginant, and C. Antzelevitch, Ionic basis for electrophysiological distinctions among epicardial midmyocardial,
and endocardial myocytes from the free wall of the canine left
ventricle, Circ. Res., 72: 671687, 1993.
22. D. B. Geselowitz, The ventricular gradient revisited: Relation to
the area under the action potential, IEEE Trans. Biomed. Eng.,
BME-30: 76, 1983.
23. R. Plonsey, Recovery of cardiac activitythe T-wave and ventricular gradient. In J. Liebman, R. Plonsey, and Y. Rudy (eds.), Pediatric and Fundamental Electrocardiography. Boston: Martinus
Nijhoff, 1987.
BIOLOGICAL EFFECTS OF ELECTROMAGNETIC FIELDS

24. R. M. Gulrajani, F. A. Roberge, and G. E. Mailloux, The forward
problem of electrocardiography. In P. W. Macfarlane and T. D.
Lawrie (eds.), Comprehensive Electrocardiology. New York: Pergamon Press, 1989.
25. Y. Rudy, R. Plonsey, and J. Liebman, The effects of variations in
conductivity and geometrical parameters on the electrocardiogram using an eccentric spheres model, Circ Res., 44: 104111,
1979.
26. S. Nandedkar and E. Stalberg , Simulation of macro EMG unit
potentials, Electroenceph. Clin. Neurophys., 56: 5262, 1983.
27. S. Xiao, K. C. McGill, and V. R. Hentz, Action potentials of curved
nerves in finite limbs, IEEE Trans. Biomed. Eng., BME-42: 599
607, 1995.
28. J. C. De Munck, B. W. van Dijk, and H. Spekreijse, Mathematical
dipoles are adequate to describe realistic generators of human
brain activity, IEEE Trans. Biomed. Eng., BME-35: 960966,
1988.
29. S. Rush and D. A. Driscoll, EEG-electrode sensitivityan application of reciprocity, IEEE Trans. Biomed. Eng., BME-16: 15
22, 1969.
30. M. P. Mintchev and K. L. Bowes, Extracting quantitative information from digital electrogastrograms, Med. Biol. Eng. Comput.,
34: 244248, 1996.
31. M. P. Mintchev and K. L. Bowes, Conoidal dipole model of electric
field produced by the human stomach, Med. Biol. Eng. Comput.,
33: 179184, 1995.
Reading List
R. Plonsey and R. C. Barr, Bioelectricity: A Quantitative Approach.
New York: Plenum Press, 1988.
J. Malmivuo and R. Plonsey, Bioelectromagnetism. New York: Oxford
Press, 1995.
D. Junge, Nerve and Muscle Excitation. 3rd ed., Sunderland, MA: Sinauer Assoc., 1992.
T. F. Weiss, Cellular Biophysics. Cambridge, MA: MIT Press, 1996.
R. D. Keynes and D. J. Aidley, Nerve and Muscle. Cambridge: Cambridge University Press, 1981.
F. Rattay, Electrical Nerve Stimulation. Vienna: Springer-Verlag,
1990.
C. A. Brebbia and J. Dominguez, Boundary Elements. New York: second edition, McGraw-Hill, 1992.
ROBERT PLONSEY
Duke University
BIOELECTRONICS. See MOLECULAR ELECTRONICS AND HYBRID COMPUTERS.
BIOENGINEERING. See BIOMEDICAL ENGINEERING.
357
394
BIOMAGNETISM
ing magnetic energy to the body it induces electric currents

that stimulate electrically excitable cells or it magnetizes
magnetic material in the body. Both in measurement and in
stimulation, the bioelectric currents in the body are the main
subject of interest of the scientific work in biomagnetism.
The biomagnetic fields and the bioelectric currents are, of
course, directly connected through Maxwells equations.
Therefore, to get new information from bioelectric sources by
a magnetic method, the measurement sensitivity of the magnetic measurement must have a distribution different from
that of an electric measurement. For more accurate information of the source distribution, the magnetic measurement
must concentrate its measurement sensitivity on a smaller
region than the electric measurement.
It is not self-evident that these two requirements are met
by a biomagnetic measurement system. These questions are
discussed in more detail here.
In addition to the theoretical requirements, there are technical reasons for using biomagnetic methods. They follow
from the different technology used in magnetic detection.
First, the magnetic field detector does not contact the body
surface. Secondly, because superconducting technology is
used, the magnetic detector is capable of measuring dc currents.
THEORY, BIOELECTROMAGNETIC BACKGROUND

Sources of Bioelectric Currents
Let us introduce the concept of the impressed current density
J i (x, y, z, t). This is a nonconservative current that arises
from the bioelectric activity of nerve and muscle cells due to
the conversion of energy from chemical to electric form. The
individual elements of this bioelectric source behave as electric current dipoles. Hence the impressed current density
equals the volume dipole moment density of the source. Note
that J i is zero everywhere outside the region of active
cells (1).
If the volume conductor is infinite and homogeneous and
the conductivity is , the primary sources J i establish an electric field E and a conduction current E. As a result, the total
current density J (2) is given by Eq. (1):
J = Ji + E
BIOMAGNETISM
Biomagnetism describes the electromagnetic and magnetic
phenomena that arise in biological tissues. These phenomena include
Magnetic field at and beyond the body
Response of excitable cells to magnetic field stimulation
Intrinsic magnetic properties of the tissue
The magnetic field is generated either by the bioelectric currents or by magnetic material in the body. Similarly, by feed-
(1)
The quantity E is called the return current. This current is

necessary to avoid buildup of charges due to the source
current.
Because the electric field E is quasistatic, it can be expressed at each instant of time as the negative gradient of a
scalar potential , and Eq. (1) may be rewritten as
J = J i
(2)
Because tissue capacitance is negligible (quasistatic conditions), charges redistribute themselves in a negligibly short
time in response to any source change. Because the divergence of J evaluates the rate of change of the charge density
with respect to time and because the charge density must be
zero, the divergence of J is necessarily zero. (We refer to the
total current J as being solenoidal, or forming closed lines of
BIOMAGNETISM
current flow.) Therefore, Eq. (1) reduces to Poissons equation:

J i = + J = 2
(3)
The solution of Eq. (3) for the scalar function for a region
that is uniform and infinite in extent (3) is

1
4 =
J i dv
(4)
r
V
Because a source element J i dv in Eq. (4) behaves like a
point source in that it sets up a field that varies as 1/r, the
expression J i is defined as a flow source density IF. Because we seek the solution for field points outside the region
occupied by the volume source, Eq. (4) may be transformed
(3) to

1
4 = J i
dv
(5)
r
v
This equation represents the distribution of potential due
to the bioelectric source J i within an infinite, homogeneous,
volume conductor that has conductivity . Here J i dv behaves
like a dipole element (with a field that varies as its dot product
with (1/r), and hence J i can be interpreted as a volume dipole density).
By using Greens theorem (4), Geselowitz (2) developed Eq.
(6) which evaluates the electric potential anywhere within an
inhomogeneous volume conductor containing internal volume
sources:

1
1

4 (r) = J i
dv +
( j j )
dS j
r
r
v
sj
j
(6)
The current density J throughout a volume conductor gives
rise to a magnetic field given by the following relationship
(3,5);

1
4H = J
dv
(7)
r
v
where r is the distance from an external field point at which
H is evaluated on an element of volume dv inside the body,
Jdv is a source element, and is an operator with respect to
the source coordinates. Substituting Eq. (2) in Eq. (7) and dividing the inhomogeneous volume conductor into homogeneous regions with surfaces Sj,

1
1

4H = J i
dv
j
dv
(8)
r
r
v
vj
j
Again using Greens theorem and making some vector manipulations, we obtain Eq. (9)

1
1

4H(r) = J i
dv +
( j j )
dS j
r
r
v
sj
j
(9)
This equation describes the magnetic field outside a finite volume conductor containing internal (electric) volume sources
J i and inhomogeneities (j j). It was first derived by Geselowitz (6).
395
It is important to note that the first term on the right-hand

side of Eq. (9), involving J i, represents the contribution of the
volume source, and the second term is the effect of the boundaries and inhomogeneities. The impressed source J i arises
from cellular activity and hence has diagnostic value,
whereas the second term is a distortion due to the inhomogeneities of the volume conductor. These same sources were
identified earlier when the electric field generated by them
was being evaluated by Eq. (6). (Just, as in the electric case,
these terms are also called primary and secondary sources.)
Similarly, as discussed in connection with Eq. (6), it is easy
to recognize that if the volume conductor is homogeneous, the
difference (j j) in the second expression is zero, and it
drops out. Then the equation reduces to the equation of the
magnetic field due to the distribution of a volume source in a
homogeneous volume conductor.
Nature of Biomagnetic Sources
Equation (9) shows that the physiological phenomenon that
is the source of a biomagnetic signal is the electric activity of
the tissue J i. Thus, for instance, the source for the magnetocardiogram (MCG) or magnetoencephalogram (MEG) is the
electric activity of the cardiac muscle or nerve cells, respectively, as it is the source of the electrocardiogram (ECG) and
electroencephalogram (EEG).
The difference between biomagnetic and bioelectric signals
is seen from the form of their mathematical equations. When
comparing Eqs. (6) and (9), one can note that the magnetic
field arises from the curl and the electric field from the divergence of the source. This distinction holds for the first component on the right-hand side of these equations that arises
from the distribution of impressed current, and for the second
component that arises from the boundaries of the inhomogeneities of the volume source.
In the design of magnetic leads, one must keep in mind the
electric origin of the magnetic signal and the characteristic
form of the sensitivity distribution of the magnetic measurement. If the lead of a magnetic measurement is not carefully
designed, it is possible that the sensitivity distribution of a
magnetic lead will be similar to that of another electric lead.
In such a case the magnetic measurement does not provide
any new information about the source.
Note that the biomagnetic signal previously discussed is
not assumed to arise from magnetic material because such
material does not exist in these tissues. There are special circumstances, however, where biomagnetic fields are produced
by magnetic materials, for example, in the case of the signal
due to the magnetic material that contaminates the lungs of
welders or the iron accumulated in the liver in certain diseases. Such fields are not discussed in this article.
Instrumentation
Biomagnetic fields have very low amplitude compared with
the ambient noise fields and to the sensitivity of the detectors.
A summary of these fields is presented in Fig. 1 (7). The figure
indicates that it is possible to detect the MCG with induction
coil magnetometers, albeit with a reasonably poor signal-tonoise ratio. However, even the most sensitive induction coil
magnetometer built for biomagnetic purposes (8) is not sensitive enough to detect the MEG for clinical use. The Superconducting QUantum Interference Device (SQUID) is the only
396
BIOMAGNETISM
104
107
108
ume conductor is identical to the distribution of the sensitivity

of the lead.
Static field
of earth
Geomagnetic
noise
Magnetic flux density [T]
(nT)
10
Commercial
flux-gate
magnetometer
high
low
1010
1011
(pT)
NASA
fluxgate
magnetometer
Radiofrequency
noise
MOG
1012
1013
1015
MMG
MCG
MEG
Magnetic noise of brain
1014
(fT)
Differences in the Information Content of the

Bioelectric and Biomagnetic Recordings
Line-frequency
and harmonic
noise
Laboratory
noise
SQUIDmagnetometer
Thermal
noise field
in eddy-current shield
Induction-coil
magnetometer
(at Tampere)
1016
Thermal noise field of body
103 102 101 100 101 102 103
Frequency [Hz]
Biomagnetic
signals
Equivalent
input noise
Noise fields
104
105
106
Thermal
noise fields
Biomagnetic signals
MCG magnetocardiogram
MMG magnetomyogram
MEG magnetoencephalogram
MOG magneto-oculogram
Noise fields
Static field of the earth
Geomagnetic fluctuations
Laboratory noise
Line frequency noise
Radio frequency noise
Equivalent input noise

Commercial flux-gate magnetometer
Ring-core flux-gate (NASA)
Induction coil magnetometer
SQUID-magnetometer.
Thermal noise fields

Eddy current shield
The human body
The initially optimistic view of the new information content

of magnetic recordings was based on consideration of Helmholtzs theorem which states that: A general vector field,
which vanishes at infinity, can be represented as the sum of
two independent vector fields; one that is irrotational (zero curl)
and another which is solenoidal (zero divergence). These vector fields are often called the flow source and the vortex
source, respectively.
In an idealized case where the head or the thorax are modeled by concentric conducting spheres, it can be shown that
the electric field generated by the bioelectric current sources
arises from a flow source and the associated magnetic field
from a vortex source. At the beginning of biomagnetic research, it was believed that because of the Helmholtz theorem
these two fields are independent and that as much new information could be obtained from magnetic recordings as is already present in electric recordings. However, experimental
studies demonstrated that these signals look very much the
same and are not fully independent. Malmivuo resolved this
apparent contradiction in the following way (10).
The sensitivity of a lead system that detects the dipolar
term of the flow source consists of three orthogonal components, each of which is linear and homogeneous (Fig. 2) (10).
Orthogonality means that none of them can be obtained as a
linear combination of the other two. Thus, the three sensitivity distributions are fully independent. However, the electric
signals each lead records cannot be completely independent
because each represents a different aspect of the same volume source.
Similarly, the sensitivity distribution of a lead system for
detecting the dipolar term of the vortex source also has three
orthogonal components. Each component can be represented
by a set of concentric circles, so that the lead sensitivity is
always tangential to the axis of symmetry. The magnitude of
Figure 1. Magnetic signals produced by various sources.
instrument sensitive enough for high-quality biomagnetic

measurements. The instrumentation for measuring biomagnetic fields is not discussed further in this article, but a good
overview of instrumentation appears in Williamson et al. (9).
LEAD-FIELD THEORY
Concept of Lead Field
The bioelectromagnetic differences between bioelectric and
biomagnetic signals may be explained by the different sensitivity distributions of electric and magnetic measurement
methods (10). The lead field is an electric current field in the
volume conductor generated by feeding a unit current to the
lead. (Because the volume conductor is not superconducting,
in magnetic leads an alternating current that has a unit time
derivative is used.) According to Helmholtzs reciprocity theorem (11), the current field produced in this manner in the vol-
Figure 2. Sensitivity distribution of a detector that detects the electric dipole moment of a bioelectric volume source.
BIOMAGNETISM
397
Ability of a Lead to Concentrate Its Measurement Sensitivity
Although the geometric form of certain electric and magnetic

leads might be similar, if one of these had its measurement
sensitivity concentrated in a smaller region, that is, if it were
capable of measuring a source region with smaller dimensions
or of localizing an equivalent dipole with better accuracy, it
would be considered superior for brain research.
Localization of a source is not possible with a lead whose
sensitivity is homogeneously distributed. Such a lead can be
used to determine only the magnitude and orientation of the
source. Therefore, the electric and also magnetic leads used
for localizing the source have forms different from those described in the previous section.
Application of the Results to Electric and Magnetic Stimulation

y
x
Figure 3. Sensitivity distribution of a detector that detects the magnetic dipole moment of a bioelectric volume source.
the sensitivity is proportional to the radial distance from the

symmetry axis (Fig. 3) (10). Again, because the sensitivity distributions of these three components are orthogonal, none of
them can be constructed as linear combinations of the other
two. Thus, all three magnetic sensitivity distributions are
also fully independent. However, as before, the three signals
detected by the magnetic leads are not fully independent because each represents a different aspect of the same volume
source.
Now it is possible to resolve the paradox involving Helmholtzs theorem. What the Helmholtz theorem expresses is not
the independence of electric and magnetic signals, but the independence of the sensitivity distributions of the recordings of
the flow and vortex sources, that is, the electric and magnetic
lead fields. It indicates that the three electric lead fields are
orthogonal to the three magnetic lead fields. However, the six
signals, measured by the dipolar electric and magnetic leads,
cannot be completely independent because they all arise from
different aspects of the underlying current source where the
activation of the cells is strongly interconnected.
On the other hand, if the sensitivity distributions of two
detection methods, regardless of whether they are electric or
magnetic, are identical in the source region, the signals and
their information contents are also identical.
We may illustrate this principle with a mechanical analog.
Though we live in a three-dimensional world, the movement
of a body is not restricted to only three directions. In addition
to the linear movement in the directions of the three coordinate axes, a body may also rotate around these three coordinate axes. These six directions are mutually independent and
are analogous to the sensitivity distributions of the measurements of dipolar electric and magnetic sources.
Because of reciprocity, the sensitivity distributions of electric

and magnetic leads can be directly applied to electric and
magnetic stimulation. In that case the sensitivity distributions can be understood as stimulation energy distributions.
This is easy to understand because what is done when calculating the lead fields is actually feeding a unit current to the
lead which can be thought of as a stimulating current.
In practice the physical dimensions of the coils in magnetic
stimulation are much larger than those used in measuring
biomagnetic fields. Therefore the results of this article concerning the calculation of magnetic lead fields are not as directly applicable to stimulation problems as are those of the
electric lead fields.
EQUIPMENT AND EXPERIMENTS
Magnetocardiography
Selection of the Source Model for MCG. In ECG and MCG
the clinical problem is to solve the inverse problem, that is,
to find the source of the detected signal so as to obtain information about the anatomy and physiology of the source. Although the actual clinical diagnostic procedure is based on
measuring certain parameters, such as time intervals and
amplitudes from the detected signal, and actually not to display the components of the source, the selection of the source
model is very important from the viewpoint of available information.
In the clinical ECG, the source model is a dipole. This is
the model for both the 12-lead ECG and vectorcardiography
(VCG). In 12-lead ECG, the volume conductor (thorax) model
is not considered, which causes considerable distortion of the
leads. In VCG, only the form of the volume conductor is modeled. This decreases the distortion in the lead fields but does
not eliminate it completely. Note that today the display systems used in these ECG and VCG systems do not play any
role in the diagnostic procedure because the computerized diagnosis is always based on the signals, not on the display.
In selecting the source model for MCG, it is logical, at least
initially, to select the magnetic source model on the same theoretical level with the ECG. Only in this way is it possible to
compare the diagnostic performance of these methods. It is
clear, of course, that if the source model is more accurate,
that is, has more independent variables, the diagnostic performance is better. But when comparing ECG and MCG, the
comparison is relevant only if their complexity is similar (10).
398
BIOMAGNETISM
x
z
(a)
x
z
100 mm
170 mm
y
70 mm
140 mm
170 mm
30 mm
(b)
Figure 4. Various methods for detecting
the magnetic dipole moment of the heart.
(a) The basic principle, the XYZ-lead system. (b) Symmetrical XYZ-lead system. (c)
Symmetrical unipositional lead system.
Detection of the Equivalent Magnetic Dipole of the Heart. The

basic method for detecting the equivalent magnetic dipole moment of a volume source is to measure the magnetic field on
each coordinate axis in the direction of that axis [Fig. 4(a)].
To idealize the sensitivity distribution throughout the volume
source, the measurements must be made at a distance that is
large compared with the source dimensions. Of course, this
decreases the signal amplitude. The quality of the measurement increases considerably if bipolar measurements are
used, that is, measurements are made on both sides of the
source [Fig. 4(b)]. Measurement of the magnetic field on each
coordinate axis, however, is difficult to perform in MCG because the geometry of the human body. It would require ei-
(c)
ther six sequential measurements with one magnetometer

(dewar) or six simultaneous measurements using six dewars.
It has been shown (12) that all three components of the
magnetic dipole also can be measured from a single location.
By applying this unipositional method symmetrically so that
measurements are made on both the anterior and posterior
sides of the thorax at the same distance from the heart, only
two dewars are needed, and a very high quality of lead fields
is obtained [Fig. 4(c)] (10).
Diagnostic Performance of ECG and MCG. The diagnostic
performance of ECG as well as MCG was compared in an extensive study made at the Ragnar Granit Institute (13). The
BIOMAGNETISM
study was made using the asymmetrical unipositional lead

system, that is, making measurements only on the anterior
side of the thorax. The patients were selected, however, so
that myocardial changes were located dominantly on the anterior side.
This study consisted of 290 normal subjects and 259 patients with different myocardial disorders. It was found that
the diagnostic performance of ECG as well as MCG is about the
same (83%). Then diagnostic parameters were selected from
both ECG and MCG. With this combined method, called electromagnetocardiogram (EMCG), diagnostic performance of 90%
was obtained. This improvement in diagnostic performance was
obtained without increasing the number of parameters used in
the diagnostic procedure. Moreover, this improvement is significant because it means that the number of incorrectly diagnosed patients was reduced by approximately 50%.
This important result may be explained as follows. The
lead system recording the electric dipole moment of the volume source has three independent leads. (This is also the case
in the 12-lead ECG system.) Similarly, the lead system detecting the magnetic dipole moment of the volume source has
three independent leads. Therefore, the diagnostic performance of these methods is about the same. However, because
the sensitivity distributions of electric and magnetic leads are
different, the patient groups diagnosed correctly with both
methods are not identical.
As stated before, the electric leads are independent of the
magnetic leads. If the diagnostic procedure simultaneously
uses both the ECG and the MCG leads, we obtain 3 3 6
independent leads, and the correctly diagnosed patient groups
may be combined. Thus the diagnostic performance of the
combined method is better than that of either method alone.
This is the first large-scale statistically relevant study of the
clinical diagnostic performance of biomagnetism.
Technical Reasons to Use MCG. The technical differences
between ECG and MCG include the MCGs far better ability
to record static sources, sources on the posterior side of the
heart, to monitor the fetal heart, and to perform electrodeless
recording. As a technical drawback, it should be mentioned
that the MCG instrument costs two to three times more than
the ECG. An important feature of MCG is that, unlike the
MEG instrument, it does not need a magnetically shielded
room. This is very important because a shielded room is very
expensive and also limits application of the technique to a
specific laboratory space.
Theoretical Reasons to Use MCG. It has been shown that
MCG has clinical value and that it can be used alone or in
combination with ECG as a new technique called the electromagnetocardiogram (EMCG). The diagnostic performance of
the combined method is better than that of either ECG or
MCG alone. With the combined method, the number of incorrectly diagnosed patients may be reduced by approximately 50%.
399
The two main theoretical aspects in favor of MEG are that,

because the skull is transparent to magnetic fields, the MEG
should be able to concentrate its measurement sensitivity in
a smaller region than the EEG, and the sensitivity distributions of these methods are fundamentally different. These issues are discussed in the following: The analysis is made by
using the classic spherical head model introduced by Rush
and Driscoll (14). In this model, the head is represented by
three concentric spheres, where the outer radii of the scalp,
skull, and brain are 92, 85, and 80 mm, respectively. The resistivities of the scalp and the brain are 2.22 m and that
of the skull is 80 times higher, 177 m.
The two basic magnetometer constructions in use in MEG
are axial and planar gradiometers. In the former, both coils
are coaxial, and in the latter, they are coplanar. The minimum distance of the coil from the scalp in a superconducting
magnetometer is about 20 mm. The coil radius is usually
about 10 mm. It has been shown (10) that with this measurement distance, decreasing the coil radius does not change the
distribution of the sensitivity in the brain region. In the following the sensitivity distribution of these gradiometer constructions is discussed.
To indicate the magnetometers ability to concentrate its
sensitivity in a small region, the concept of half-sensitivity
volume has been defined (15). This concept means the region
in the source area (brain) where the detector sensitivity is
one-half or more of the maximum sensitivity in the source
region. The smaller the half-sensitivity volume, the better is
the detectors ability to focus its sensitivity in a small region.
In magnetocardiography, it is relevant to detect the magnetic dipole moment of the volume source of the heart and to
make the sensitivity distribution within the heart region as
independent of the position in the axial direction as possible.
In magnetoencephalography, however, the primary purpose is
to detect the electric activity of the cortex and to localize the
regions of certain activity.
Sensitivity Distribution of the Axial Magnetometer. In a cylindrically symmetrical volume conductor model, the lead field
flow lines are concentric circles and do not cut the discontinuity boundaries. Therefore, the sensitivity distribution in the
brain area of the spherical model equals that in an infinite,
homogeneous volume conductor.
Figure 5 illustrates the sensitivity distribution of an axial
magnetometer. The thin solid lines illustrate the lead-field
flow lines. The dashed lines join the points where the sensitivity has the same value, the so-called isosensitivity lines. The
half-sensitivity volume is represented by the shaded region.
Magnetoencephalography
Sensitivity Distribution of the Planar Gradiometer. Figure 6

illustrates the sensitivity distribution of a planar gradiometer. Again, the thin solid lines illustrate the lead-field flow
lines, and the dashed lines represent the isosensitivity lines.
The half-sensitivity volume is represented by the shaded region. The sensitivity of the planar gradiometer is concentrated under the center of the two coils and is mainly linearly
oriented. Further, two zero-sensitivity lines exist.
Similarly as in the cardiac application, in magnetic measurement of the electric activity of the brain, the benefits and
drawbacks of the MEG can be divided into theoretical and
technical categories. First, the theoretical aspects are discussed.
Half-Sensitivity Volumes of Electro- and Magnetoencephalography. The half-sensitivity volumes for different EEG and MEG
leads as a function of electrode distance and gradiometer
baselines are shown in Fig. 7(a). The minimum half-sensitiv-
400
BIOMAGNETISM
100
80
Magnetometer coil r = 10 mm
1 A/s
40
20
0
20
40
60
Coil distance
from scalp = 20 mm
10
20
10
Maximum
sensitivity
104
60
r[mm]
100
80
Half-sensitivity
volume
30
5000
4000
40
50
5000
4000
3000
3000
2000
2000
rb 76 mm
78 mm
80 mm
100
rc 80 mm
1000
1000
rs 85 mm
rt 92 mm
150
500
500
400
400
300
300
200
Figure 5. Sensitivity distribution of an

axial gradiometer in the inhomogeneous
spherical head model.
200
h [mm]
200
100
Zero sensitivity line
ity volume, of course, is achieved with the shortest distance/

baseline. For three- and two-electrode EEG leads, the halfsensitivity volumes at 1 of electrode distance are 0.2 and
1.2 cm3, respectively. For 10 mm radius planar and axial gradiometer MEG leads, these volumes at 1 of coil separation
(i.e., 1.6 mm baseline for axial gradiometer) are 3.4 and
21.8 cm3, respectively.
The 20 mm coil distance from scalp and 10 mm coil radii
are realistic for the helmet-like whole-head MEG detector.
However, MEG devices exist for recording in a limited region
where the coil distance and the coil radii are on the order of
1 mm. Therefore the half-sensitivity volumes for planar
gradiometers with a 1 mm coil radius at 0 mm to 20 mm
recording distances are also illustrated in Fig. 7(b). These
curves show that when the recording distance is about 12 mm
and the distance/baseline is 1 mm, such a planar gradiometer
has about the same half-sensitivity volume as the twoelectrode EEG.
Short separation, of course, also decreases the signal amplitude. An optimal value is about 10 of separation. Increasing the separation to 10 increases the EEG and MEG signal
amplitudes to approximately 70% to 80% of their maximum
values, but the half-sensitivity volumes do not increase considerably from their values at 1 of separation.
100
0
JLM [pA/m2]
Thus, contrary to general belief, the EEG can focus its sensitivity better on a small region in the brain than the wholehead MEG. At about 20 to 30 of separation, the two-electrode EEG lead needs slightly smaller separation to achieve
the same half-sensitivity volume as the planar gradiometer.
The sensitivity distributions of these leads, however, are similar. Note that if the sensitivity distributions of two electric or
magnetic different lead systems, are the same, they detect
exactly the same source and produce exactly the same signal.
Therefore, the planar gradiometer and two-electrode EEG
lead detect similar source distributions.
Sensitivity of EEG and MEG to Radial and Tangential
Sources. The three-electrode EEG has its maximum sensitivity under the electrode that forms the terminal alone. This
sensitivity is mainly directed radially to the spherical head
model. With short electrode distances, the sensitivity of the
two-electrode EEG is directed mainly tangentially to the
spherical head model. Thus with the EEG it is possible to
detect sources in all three orthogonal directions, that is, in
the radial and in the two tangential directions relative to the
In the axial gradiometer MEG lead, the sensitivity is directed tangentially to the gradiometer axis of symmetry and
BIOMAGNETISM
thus also tangentially to the spherical head model. In the planar gradiometer, the sensitivity has its maximum under the
center of the coils and is directed mainly linearly and tangentially to the spherical head model. The MEG lead fields are
oriented tangentially everywhere to the spherical head model.
This may be easily understood by recognizing that the lead
field current does not flow through the surface of the head
because no electrodes are used. Therefore, the MEG detects
only sources oriented in the two tangential directions relative
to the spherical head model.
The biomagnetic measurement technology is and will always

be more expensive than the bioelectric technology. Especially
this holds on MEG due to its very low signal amplitude.
Therefore the biomagnetic measurements must have verified
benefits over the bioelectric measurements to be worth to
apply.
Magnetocardiography
There exists one study comparing the clinical diagnostic performance of ECG and MCG (13). This study did demonstrate that the diagnostic performance of these methods is
10
80
about the same. But because the sensitivity distributions

of the methods, when applied correctly, are fundamentally
different and independent, the patient groups diagnosed
correctly with both methods are not identical. Therefore,
by combining the methods, we may combine the patient
groups and increase the diagnostic performance even so
much that the number of incorrectly diagnosed patients
may be decreased to half.
Magnetoencephalography
As discussed before:
EVALUATION
100
401
60
Coil distance from

scalp = 20 mm
40
Planar gradiometer
Coil radius = 10 mm
1 A/s
20
0
20
Maximum
sensitivity
40
The MEG is spatially not more accurate than the EEG.

The planar gradiometer MEG does not measure a source
complementary to the EEG and therefore does not provide information essentially different from that of the
EEG.
From the electric sources the MEG does not measure the
component radial to the head. The EEG measures separately all three orthogonal components of the electric
sources.
Though there is no need to fix electrodes with the MEG,
the dewar restricts the movement of the patient. At pres-
60
Half-sensitivity
volume
x[mm]
100
80
z
y
x
20
30
40
104
50
103
103
103
102
100
0
Zero sensitivity line
JLM [pA/m2]
150
102
200
h [mm]
102
Figure 6. Sensitivity distribution of a

planar gradiometer in the inhomogeneous
BIOMAGNETISM
60
50
e te
(Inner sphere volume = 683 cm3)

Vortex
r = 10 mm
h = 20 mm,
mete
Ax
ia
gr
om
di
h = 20
mm
r = 10
mm
EEG/MEG
Half-sensitivity volume
[cm3]
1 coil
Vortex
402
30
Plan
Vo
rte
ar gr
adio
40
Ta
1
10 Electrode
ng
en
tia
20
Radial
2 Electrodes
es
3 Electrod
30
0
0
50
d = 36mm
Radia
60
100
d = 72mm
(21 electrodes)
90
120
150
200
[ ] 180
150
250 [mm]
Electrode/coil separation
(a)
EEG/MEG
Half-sensitivity volume
[cm3]
(Inner sphere volume = 683 cm3)
El
ec
tro
de
5
Planar gradiometer
h = 20 mm r = 10 mm
r = 1 mm
h = 20 mm
15 mm
lec
3E
10 mm
5 mm
0 mm
Figure 7. (a) Half-sensitivity volumes of different EEG and MEG
leads as a function of electrode distance/magnetometer baseline; (b)
Lower left corner of the previous figure magnified.
ent there are available electrode caps which allow fixing

of over 100 electrodes to the head within some 10 min.
The MEG needs, at least at present, a magnetically
shielded room whose size due to the size of the MEG
dewar exceeds normal laboratory height. This restricts
the application of the MEG to certain locations. Instead,
15
10
0
10
trod
20
es
[ ] 20
[mm] 30
Electrode/coil separation
(b)
the EEG can be recorded at any location making it more

easily accessible for patients.
Even though there exist more than 40 MEG installations
in the world, there does not exist any clinical study with a
relevant number of patients where the diagnostic perfor-
BIOMEDICAL ENGINEERING
mances of the MEG and EEG were compared. Therefore, at

least at the moment, there does not exist any theoretical or
clinical evidence on the superiority of the MEG over the EEG
which would justify its use.
CONCLUSION
Biomagnetic measurements have been performed for 35
years. SQUID-technology, which is essential for measurement
of ultralow intensity biomagnetic fields, has existed for almost
30 years. At present there exist some 40 large-scale installations for MEG research in the world. In addition to those,
several smaller groups make biomagnetic studies. The number of active scientists in biomagnetism is several hundreds.
The groups working with MEG have been able to demonstrate that the technology can be used for detecting the electric activity of the brain and for obtaining new important research results in neurology. However, none of these groups
has demonstrated that the results have been obtained with
the MEG more accurately than with the EEG or are only
available with the MEG.
In MCG research most of the research activities are concentrated on localization of the arrhythmogenic foci of the
heart or finding indicators for the risk of sudden cardiac
death. Neither of these approaches have been more successful
than the ECG. The number of researches on utilizing the
unique sensitivity distribution of the MCG in relation to the
ECG is very limited. However, it is only just this application
which has been successful in biomagnetism.
Before biomagnetic methods can be accepted for clinical
use there must exist several clinical studies demonstrating
the superiority of these methods.
BIBLIOGRAPHY
1. R. Plonsey, Bioelectric Phenomena, New York: McGraw-Hill,
1969, p. 380.
2. D. B. Geselowitz, On bioelectric potentials in an inhomogeneous
volume conductor, Biophys. J., 7 (1): 111, 1967.
3. J. A. Stratton, Electromagnetic Theory, New York: McGraw-Hill,
1941.
4. W. R. Smyth, Static and Dynamic Electricity, 3rd ed., New York:
McGraw-Hill, 1968.
5. J. D. Jackson, Classical Electrodynamics, 2nd ed., New York: Wiley, 1975, p. 84.
6. D. B. Geselowitz, On the magnetic field generated outside an inhomogeneous volume conductor by internal current sources,
IEEE Trans. Magn., MAG-6: 346347, 1970.
7. J. A. Malmivuo et al., Improvement of the properties of an eddy
current magnetic shield with active compensation, J. Phys. E.:
Sci. Instrum., 20 (1): 151164, 1987.
8. K.-P. Estola and J. A. Malmivuo, Air-core induction coil magnetometer design, J. Phys. E.: Sci. Instrum., 15: 11101113, 1982.
9. S. J. Williamson et al., (eds.), Biomagnetism: An Interdisciplinary
Approach, New York: Plenum, 1983, p. 706.
10. J. Malmivuo and R. Plonsey, Bioelectromagnetism-Principles and
Applications of Bioelectric and Biomagnetic Fields, New York: Oxford Univ. Press, 1995, p. 480.
11. H. L. F. Helmholtz, Ueber einige Gesetze der Vertheilung elektrischer Strome in korperlichen Leitern mit Anwendung auf die
thierischelektrischen Versuche, Ann. Physik Chemie, 89: 211
233, 354377, 1853.
403
12. J. A. Malmivuo, On the detection of the magnetic heart vector-an

application of the reciprocity theorem, Helsinki Univ. Tech., Acta
Polytechn. Scand., Elec. Eng. Series, 39: 112, (Ph.D. Thesis),
1976.
13. O. S. Oja, Vector magnetocardiogram in myocardial disorders,
MD Thesis, University of Tampere, Medical Faculty, Tampere,
Finland, 1993.
14. S. Rush and D. A. Driscoll, EEG-electrode sensitivity-an application of reciprocity, IEEE Trans. Biomed. Eng., BME-16: 1522,
1969.
15. J. Malmivuo, V. Suihko, and H. Eskola, Sensitivity distributions
of EEG and MEG measurements, IEEE Trans. Biomed. Eng., 44
(3): 196208, 1997.
JAAKKO MALMIVUO
Tampere University of Technology
403
Biomedical engineering is the collective term for the disciplines that bring the concepts and principles of engineering to
the field of medicine. The integration of chemical, mechanical,
electrical, and computer engineering fundamentals with biology and medical science has a relatively recent history that
began in the mid-1900s. Technological and scientific advances
in the twentieth century created the opportunity for biomedical engineering innovations, such as physiological simulation
and modeling, designing of implants and drug delivery systems, and development of instrumentation and diagnostic
tools.
The emergence of biomedical engineering followed the
movement of primary medical care from the home to the hospital in the 1930s and 1940s. Until this time, hospitals were
used mainly for care of the poor. Home care by physicians,
midwives, and family was the predominant form of health
care. The hospital, however, became the center of modern
medical care after the discovery of X rays and antibiotics. By
the 1930s, the use of barium salts and radio-opaque materials
allowed X-ray visualization of practically all organ systems
(1). Because of its cost, the improved diagnostic capability
that radiation equipment provided was available only at hospitals. The advent of antibacterial agents and antibiotics, for
example, sulfanilamide in the mid-1930s and penicillin in the
early 1940s (1), helped prevent cross-infection among patients
and staff, a previous deterrent to hospital care.
Electronic innovations developed for the military effort in
World War II provided the basis for advances in medical electronics in the post-war era. These advances made it possible
to measure low-level biosignals, which lead to a better understanding of electrical impulses and the central nervous system. The biologists who had been recruited for radar work
during the war were prepared for these developments. However, the next generation of biologists was without this benefit, and now technology was advancing rapidly. The need for
a bridge between the gap of technical knowledge and biology
resulted in the emergence of the biomedical engineer (2).
The areas in which engineering blends with medicine are
abundant and diverse. Biomedical engineers design imaging
and diagnostic instrumentation, drug delivery systems, medical sensors, prostheses, rehabilitative devices, and artificial
organs. They develop biocompatible materials, model physioJ. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering. Copyright # 1999 John Wiley & Sons, Inc.
404
logical systems, or create patient information databases that

assist in making clinical decisions. Some biomedical engineers work in hospitals where they design clinical systems
and procedures and others work as consultants in the field of
rehabilitation engineering to help restore mobility and other
functions for disabled individuals. In recent years, some biomedical engineers have begun to work in the areas of genetic
and tissue engineering. Some of the disciplines within biomedical engineering include the following (3):
Biological effects of electromagnetic fields

Biomaterials
Biomechanics
Biomedical instrumentation
Biomedical sensors
Biomedical signal analysis
Biotechnology (including tissue engineering)
Clinical engineering
Medical imaging
Medical informatics
Physiological modeling, simulation, and control
Prostheses and artificial organs

Rehabilitation engineering
The challenges facing biomedical engineers do not necessarily fall solely into one category. For example, the design of
an artificial heart involves knowledge of design requirements
for the artificial organ and the appropriate biomaterials.
Given the broad range of problems studied in biomedical engineering, it is rare for a single person to have expertise in all
facets of the field.
BIOLOGICAL EFFECTS OF ELECTROMAGNETIC FIELDS

A biomedical engineer involved in the study of the biologic
effects of electromagnetic fields studies a variety of diagnostic
and therapeutic applications of electromagnetic fields and the
adverse effects of such fields. The diagnostic uses of extremely
low-frequency (ELF) magnetic fields include magnetic resonance imaging (MRI), which also uses a radiofrequency (RF)
field, and neural stimulation by strong magnetic pulses. Bone
and cartilage repair, soft-tissue healing, and nerve repair or
regeneration are among the therapeutic applications under
investigation for applications of low-frequency, pulsed electromagnetic fields (PEMF). There is considerable concern in todays society regarding the bioeffects of electromagnetic fields,
but no deleterious effects have been shown to be associated
with long-term exposure to low-level electromagnetic fields.
In cancer treatment, heat generated by radio-frequency energy (3 kHz to 300 GHz) or microwave energy (300 MHz to
300 GHz) kills radiation-resistant tumor cells. This therapeutic use of electromagnetic energy is termed hyperthermia. Another use of electromagnetic energy, called electroporation,
involves the use of an electrical pulse to disrupt the membranes of cells. This is a common DNA transformation technique used in biotechnology. There is additional interest in
electroporation that stems from its possible use in drug delivery systems.
BIOMATERIALS
Biomedical engineers working in this area are concerned with
researching and designing safe and reliable synthetic materials that can intimately contact living systems and tissues.
This contact makes it essential that these materials are physiologically acceptable and pharmacologically inert, that is,
nontoxic and noncarcinogenic. Additional requirements include (1) adequate mechanical strength, (2) adequate fatigue
life, (3) proper weight and density, and (4) usable in reproducible and cost effective large-scale fabrication (4). Examples of
biomaterials range from replacement parts to sutures, diagnostic aids, and tooth fillings. The three main classes of biomaterials are metals, ceramics, and polymers.
Metallic Biomaterials
Metals in the body can corrode and possibly cause damage to
an implant and harmful interactions with its corrosion products. Some metals, such at iron (Fe) and cobalt (Co), are required by the body for normal function but are still harmful
if available in more than minute quantities. Implants consisting of stainless steels, cobalt-chromium (CoCr) alloys, and
titanium (Ti) and titanium alloys are corrosion resistant and
biocompatible. Stainless steels with molybdenum (Mo), types
316 and 316L, have increased salt-water corrosion resistance
and are commonly found in temporary implants like fracture
plates or screws. Type 316L, which differs from type 316 only
in carbon content, is more widely recommended. Cobalt-chromium alloys are used in dentistry and artificial joints (castable CoCrMo) and in knee and hip prostheses (wrought CoNiCrMo). Titanium is a strong, lightweight metal that is ideal
for implants. However, its poor shear strength precludes it
from being used in bone screws and plates. Titanium-nickel
alloys (TiNi) have the uncommon property of shape memory
effect (SME). This is involved when, after deformation, a material returns to its previous shape when heat is applied. Attempts to take advantage of this property include research
into intracranial aneurysm clips, contractile artificial muscles
for an artificial heart, and orthopedic implants. There are
other specialized uses for metals, such as platinum alloys for
electrodes and tantalum for wire sutures. In dentistry, gold
provides durable, corrosion-resistant fillings, and gold alloys
are implemented in cast restorations, inlays, crowns, and
cusps. Dental amalgam for cavities is a mixture of liquid mercury with silver, tin, copper, and zinc (5).
Ceramic Biomaterials
Ceramics are primarily inorganic, polycrystalline compounds,
such as silicates and metal oxides. However, the covalently
bonded forms of carbon, such as graphite and diamonds, are
also considered ceramics. Improvements in ceramic formation
in the late twentieth century have resulted in materials without the characteristic brittleness and low impact and tensile
strengths that previously limited the use of ceramics. Ceramics are typically used in bone replacement and dental crowns.
The three general classifications of ceramic biomaterials are
(1) nonabsorbable, (2) bioactive or surface reactive, and (3)
biodegradable or resorbable. Ceramics are also designated as
relatively inert, semi-inert, and noninert.
Common nonabsorbable or relatively inert ceramics include alumina (Al2O3), zirconia (ZrO2), and carbonaceous ce-
ramics. Implant uses for nonabsorbable ceramics are generally for structural support, such as bone plates, bone screws,
and femoral heads. This class of ceramics has also been used
in ventilation tubes and in sterilization and drug delivery devices. Carbon ceramics are primarily for coating surfaces of
devices that are used to repair diseased heart valves and
blood vessels because of their high compatibility with tissue
and blood.
Surface-reactive ceramics are primarily used to coat metal
prostheses. When implanted, these ceramics actually form
strong bonds with surrounding tissue. Dense, nonporous
glass-ceramics, formed by controlled crystallization of glasses,
fall into this category.
Resorbable or noninert ceramics are used to make both implants and drug delivery devices. The implants, predominantly variations of calcium phosphate, typically act as a substitute for bone. After implantation, resorbable ceramics
degrade while endogenous tissue replaces it (6,7).
Polymeric Biomaterials
Various medical supplies, devices, and implants consist of
polymers. The biocompatibility, ease of processing into diverse shapes, and the relatively low cost make polymers ideal
biomaterials. Out of the multitude of polymers, only about
twenty are used as biomaterials. Polyvinylchloride (PVC) tubing, sheets and films form IVs, catheters, cannulae, blood and
solution bags, and surgical packaging. Pharmaceutical bottles, pouches and bags, and orthopedic implants are made
from polyethylene (PE) of varying densities. Artificial vascular grafts, suture, and packaging of devices are among the
medical uses of polypropylene (PP). Soft contact lenses, implantable ocular lenses, dentures, bone cement for joint prostheses fixation, blood pumps and reservoirs, membranes for
blood dialyzers, and IV systems contain polymethylmethacrylate (PMMA). Sutures, including resorbable sutures, and
artificial vascular graft applications involve polyesters. Nylon
thread is a common surgical material. Other polymeric biomaterials include polystyrene (PS) and polystyrene copolymers; rubbers, such as silicone rubber; polyurethane; polyacetal; polysulfone; polycarbonate; and fluorocarbon polymers
(primarily for coatings), such as Teflon (7).
Composite Biomaterials
Materials consisting of two distinct phases or components are
called composites. Various biological materials, such as bone
and skin, are naturally occurring composite materials. Research and development in the field of biomaterials includes
implants formed from composites. Composites offer a means
to manipulate properties, such as strength-to-weight ratios
and stiffness, in ways not possible with homogeneous materials.
The shape of the inclusion material of a composite can be
classified in three ways: (1) particulate, (2) fiber, and (3)
platelet or lamina. These consist of none, one, or two long
dimensions, respectively. Polymeric biomaterials can contain
particulates or fibers to improve stiffness. Examples include
inclusion of bone particles or metal fibers in PMMA to improve the stiffness and fatigue life of bone cement and silica
(SiO2) particles in rubber to strengthen catheters and rubber
gloves (8). Honeycombs and foams are composite materials
and voids that are filled with either air or a liquid. Some po-
405
rous composites have been used as a support for tissue

growth.
BIOMECHANICS
Materials can be classified as either solids or fluids based on
the response of the material to a constant force. A solid gives
a discrete, finite time-independent response whereas a fluid
responds with a continuous, time-independent response.
Whether a material behaves as a solid or fluid has much to
do with its thermodynamic state. At a given temperature and
pressure, some materials, called viscoelastic, display both responses depending on the rate at which the force is applied.
Biomechanics deals with determining the time and space
characteristics of biological solids, fluids, and viscoelastic materials in response to imposed systems of internal and external forces (9).
The knowledge gained by mechanical analysis of biological
materials has a significant impact on understanding the
mechanisms of failure and requirements for replacement devices, such as prostheses and artificial organs. Biomechanics
assists in designing orthopedic prosthetics by evaluating surface motion and lubrication of joints to provide information
about joint wear, stability, and degeneration. Models have
been developed which account for the viscoelastic behavior of
weight-bearing long bone and assist in evaluating design
specifications for lower limb prostheses. One active area of
research involves accurately modeling porous bone to study
the effects of aging, such as osteoporosis. Analysis of blood
vessels has led to an understanding of vascular mechanics,
which is useful when designing vascular grafts (10). This type
of information is necessary for accurate modeling of physiological systems, for example, determining the forces generated by skeletal muscles (11) and those that act on muscles
through tendons (12).
In addition to quantifying forces, biomechanics provides information about mechanisms of failure or injury needed to
modify the environment of individuals. Head and neck injuries are studied to help in designing better support systems,
for example air bags, that are important in the event of a car
accident. Ergonomics is a field in which work conditions are
analyzed to help determine methods for preventing back injuries or discomfort.
BIOMEDICAL INSTRUMENTATION
Biomedical instrumentation provides the necessary tools for
measuring physiological variables and parameters (1317).
Great advances in biomedical instrumentation have resulted
from developments in electronics and from the advent of the
computer age. Biomedical instrumentation includes equipment that is used to diagnose disease in a patient, devices
that are used to improve or maintain the health and wellbeing of a patient, and instruments that are used to continuously monitor the current physiological state of a patient.
While developments in electronics have contributed much to
the increased capabilities and sophistication of biomedical instrumentation, the proliferation of medical and nonmedical
electronic devices has also contributed to radio frequency interference (RFI), which can affect the performance of some
medical equipment.
406
The electrocardiograph (ECG), which first appeared in hospitals in 1910, measures the electrical activity of the heart
(18). Devices that measure the electrical activity in other
parts of the body also contribute to current diagnostic capabilities. In addition to the ECG, bioelectric phenomena that are
measured for research and diagnostic purposes include electroencephalography (EEG), electromyography (EMG), electroretinography (ERG), and electrogastrography (EGG), which
measure the electrical activity of brain, muscle, eye, and
stomach, respectively. The measurement of propagated neural impulses that result from electrical stimulation is used to
assess nerve damage.
Biomagnetic fields arise from the electrical activity of tissue. The magnetocardiogram (MCG), or magnetic measurement of the electric activity of the heart, has the highest amplitude of biomagnetic signals (50 pT) and was first detected
in 1963 by Baule and McFee. The MCG, unlike the other
lower amplitude biomagnetic signals, does not require a magnetically shielded room. Comparisons between the MCG and
the ECG have revealed similar capabilities for diagnosing
myocardial disorders with 50% improvement when combined
as an electromagnetocardiogram (EMCG) (19). The ECG is
still much more widely used than the MCG.
Other biomagnetic measurements, for example, the electrical activity of the brain which is called a magnetoencephalogram (MEG), are limited in location by the need for a room
with magnetic shielding because of the very low amplitude of
the signals. The development of the superconducting quantum interference device (SQUID) in 1970 made it possible to
record these low biomagnetic signals with good signal quality.
There are thought to be two advantages of MEG over the
EEG: (1) the ability to measure smaller regions of the brain
and (2) fundamental differences in the sensitivity distribution
between the two methods.
Implantable pacemakers help patients who cannot maintain a steady heartbeat by supplying a controlled, rhythmic
electric stimulus to the heart. This stimulus mimics the action of the sinoatrial node (SA node) of a healthy heart, the
hearts natural pacemaker. With modern implantable pacemakers, clinicians use telemetry to program and monitor
functions externally.
Ventricular fibrillation (VF) is a type of cardiac arrhythmia
that is lethal. Death occurs in minutes during VF if the condition is not corrected. Because self-correction is rarely possible,
defibrillation, typically by the application of an electrical
shock to the heart, resets the heart to normal beating. Defibrillators are used externally, as in emergency rooms or ambulances, or are implanted into patients who are at constant
risk of developing VF. Some commercial airlines are now
equipped with automatic defibrillators that will trigger a
shock if the device determines that the patient is having VF.
These devices do not have to be operated by clinically
trained personnel.
Bioelectric impedance analysis (BIA) of tissue provides information about the small pulsatile impedance changes that
occur during heart and respiratory action. BIA is used to determine body characteristics (e.g., percent body fat) or to reconstruct tomographical images of the body (20,21) by measuring conductivity and permittivity at different frequencies.
BIOMEDICAL SENSORS
Biomedical sensors, or biosensors, convert biologically significant signals into electrical signals (13,15,17,18,22,23).
These sensors have both diagnostic and therapeutic applications, and can be active or passive devices. Two major classes
of biomedical sensors, which are based on the variable measured, are physical and chemical sensors. Bioanalytic sensors
are a special class of chemical sensors that take advantage of
biochemical binding reactions to identify complex biomolecules with high specificity and selectivity. One of the earliest
and most clinically relevant biosensor applications was developed for measuring blood gases (O2, CO2) and pH. Measuring
blood gases and pH continues to be an important use of biomedical sensors. Other aspects of blood chemistry, for example, glucose and lactate, can now be measured. Another
method of classification, involving the method of application
of the sensor, is divided into four categories: (1) noncontacting, (2) skin surface, (3) indwelling (minimally invasive),
and (4) implantable. Indwelling and implantable devices involve carefully selecting inert biomaterials for the sensing interface and packaging. Implantable sensing devices need to
maintain long-term calibration and function.
BIOMEDICAL SIGNAL ANALYSIS

Biomedical signals, signals that contain information about a
biological system, often need processing so that the physiologically meaningful parts of the signal are extracted (2430).
Processing involves enhancement that reduces noise or transformation to obtain hidden information. Typical digital signal
processing techniques include filtering, averaging, and spectral estimations. Signal enhancement to remove noise in frequency domain signals is achieved through optimal and adaptive filtering. Optimal filtering is for stationary signals
whereas adaptive filtering adjusts to perform under changing
circumstances. In general, adaptive filters are more appropriate for biomedical signals.
Sources of biomedical signals include (1) bioelectrical signals generated by nerve cells and muscle cells; (2) bioimpedance signals from the impedance of tissue; (3) bioacoustic
signals from the flow of blood and air and sounds in the digestive tract, the joints, and contracting muscle; (4) biomagnetic
signals from various organs, such as the brain and heart; (5)
biomechanical signals resulting from mechanical function,
such as motion, displacement, pressure, tension, and flow; (6)
biochemical signals arising from chemical measurements; and
(7) biooptical signals by both natural and induced optical
functions.
For analysis, the main concern is the signal characteristics, not the origin of the signal. Therefore, another classification system involves identifying the signal as either continuous or discrete. Continuous signals are transformed into
discrete signals by sampling. Additionally, biomedical signals
are generally stochastic, which means they cannot be described exactly graphically or by an equation but rather in
terms of probability.
BIOTECHNOLOGY
Biotechnology is not considered a discipline but rather a collection of procedures and techniques by which a scientist or
engineer attempts to modify biological organisms for the benefit of humanity. These attempts include improving plants
and animals for agricultural and food production, genetic en-
gineering of organisms to produce therapeutic proteins, and

biological fuel generation (3136).
A predominant area of biotechnology is manipulating biological organisms to produce proteins, including industrial enzymes, therapeutic proteins, and animal feed supplements.
These proteins are found naturally in the organism (e.g., bacterial amylases used in food production and biological detergents) or are introduced by gene transfer techniques (e.g., insulin production in bacteria). Therapeutic proteins produced
through genetic engineering are termed recombinant therapeutic proteins. This comes from the term recombinant DNA,
which means a combination, not possible in nature, of DNA
from two organisms through genetic engineering. Examples
of recombinant therapeutic proteins include insulin (with
about 100 amino acids it is technically a polypeptide), the
growth hormone somatostatin, and immunity-enhancing lymphokines.
Human Genome Project
The human genome project (HGP) represents an area of great
possibilities for biotechnology in medicine (31,32,34,37). The
term genome refers to the entire genetic material of an organism. Begun in 1990, the HGP will sequence the approximately
100,000 genes on the 22 homologous chromosomes and the
two sex chromosomes of a human by the year 2005. The HGP
has already begun to provide valuable information on singlegene defects (diseases caused by a mutation in a single gene)
and to improve possibilities for gene therapy. Engineers have
contributed to the HGP by developing equipment that can
rapidly sequence large segments of DNA or produce large
quantities of a single DNA strand.
Tissue Engineering
Tissue engineering is separated into two main categories: (1)
in vitro and (2) in vivo. In vitro methods in tissue engineering
involve the use of bioartificial tissues, which are hybrids of
synthetic and living material. A typical use of in vitro tissue
engineering is in organ replacement in lieu of an organ transplant. In vivo applications attempt to alter the growth and
function of cells. A typical in vivo application would use implanted polymeric tubes to promote nerve regeneration by reconnecting damaged nerves in the peripheral nervous system
(38). Generally mammalian cells need a support or attachment surface (substrate) to proliferate. Extracellular protein
influences how cells interact with the surface and surrounding cells, especially cell adhesion. The seeding density of
cells on these supports is a primary concern for the necessary
interaction and communication between cells. Therefore, the
three main determining factors in the ultimate morphology of
the tissue are cellsubstrate adhesion, cellcell adhesion, and
the rigidity of the substrate. Cultures can be seeded in a
three-dimensional matrix, on single surfaces, or in a sandwich configuration. The seeding support is typically composed
of collagen. Other considerations in tissue engineering involve
types of cells selected (typically differentiated cells, such as
hepatocytes and pancreatic islets cells), metabolic requirements for the cells (oxygen tends to be limiting), and control
of tissue organization.
CLINICAL ENGINEERING
Biomedical engineers who work within hospitals or clinics are
called clinical engineers. Clinical engineers support and main-
407
tain all biomedical instrumentation within the hospital and

provide recommendations for and assessments of new instrumentation. This involves managing equipment inspections
and preventive maintenance schedules, modifying or repairing instrumentation, and overseeing medical technician
training on the safe and proper use of the equipment. Clinical
engineers play a role in the design of medical instrumentation
and new clinical or hospital facilities. Additionally, equipment
inventory and computer support fall within the scope of clinical engineering (39).
MEDICAL IMAGING
Medical imaging provides vital information about a bodys
structures and functions. Examples of medical imaging modalities include X rays, magnetic resonance imaging (MRI),
positron emission tomography (PET), single-photon emission
computed tomography (SPECT), ultrasound, and computed
tomography (CT). These areas have advanced rapidly with
the computer age. However, challenges still exist to reduce
the cost of common imaging equipment.
The discovery of X rays by Wilhelm Roentgen in 1895 provided the first technique for seeing inside the human body
(40). The theory behind the images involves the exposure of
the body to X rays which pass through to a detector or interact by being absorbed or scattered. When scattered, the X
rays may still reach the detector and cause a loss in image
quality. When there is not enough variation in the absorption
of X rays between the area of interest and the surrounding
tissues, contrast is provided by barium salts (strong X-ray absorbers). Radiopaque materials, such as iodine compounds,
provide the contrast in X-ray angiography (serial radiographs
of the circulatory system). Standard X-ray imaging is used to
detect disease or injury in bones or other body structures,
while mammograms are used to diagnose breast cancer
(41,42). The X-ray tube for mammograms is different from the
one used to detect changes in bony structures.
Computed tomography (CT), which was developed in the
1970s and is based on the same principles as X-rays, provided
the first cross-sectional images of internal body structures
(43). CT images are produced by reconstructing a large number of X-ray transmission measurements, called projection
data, into tomographic maps of the X-ray linear attenuation
coefficient. Now a standard procedure in most hospitals,
practically all parts of the body are imaged by CT technology. One of the problems associated with both CT and Xray imaging is that tissue damage can occur if single exposures or the accumulated life-time exposures to X rays exceed safe levels.
Magnetic resonance imaging (MRI) uses a strong magnetic
field to align the weak nuclear moments of materials with
atoms containing an odd number of protons or neutrons (e.g.,
1
H, 13C, and 31P) (44,45). Typically, MRI images the protons
(1H) of water because the body is two-thirds water. However,
it is not possible to directly measure the weak signals of the
protons that are aligned with the strong applied magnetic
field. Therefore, resonance techniques are employed to measure the collection of the nuclear moments, called spins. To
distinguish the locations of spins, the magnetic field that is
imposed in MRI has spatial variations. Primarily, MRI images provide diagnostic information. Recently, research efforts on blood flow and brain perfusions, termed functional
408
magnetic resonance imaging (fMRI) have been aided by using

MRI. In addition to diagnostic capabilities and research, an
interest has developed in using MRI in image-guided surgical
procedures. Future developments will include open MRI, realtime MRI, and continuous MRI during surgery.
A new form of microscopy, magnetic resonance microscopy,
allows studying biological samples noninvasively and nondestructively, unlike electron and optical microscopy. This extension of magnetic resonance imaging provides three-dimensional images with spatial resolution better than 100 m (46).
Magnetic resonance microscopy is used in histologic studies,
toxicologic studies, and developmental biology.
In nuclear medicine, the patient is given a small dose of
radionuclide either intravenously or by rebreathing or ingestion. The low doses of radioactivity are safe for the patient
and allow external imaging without interaction with the organ of interest. Single-photon emission computed tomography
(SPECT) is the result of combining nuclear medicine and computed tomography. SPECT uses radioactive pharmaceuticals,
which undergo differential distribution based on the type of
tissue or organ, in lieu of the conventional X rays used in CT
scans. SPECT is generally the primary imaging technique for
the brain (47).
In positron emission tomography (PET), tracers are injected into the patient (48). These tracers are metabolically
active biomolecules with positron-emitting isotopes, such as
11
C, 13N, 15O, and 18F. An array of detectors captures simultaneous recordings of two photons that travel in opposite directions. These pairs of photons result from the annihilation of
an electron and an emitted positron (positive electron) from
the tracer. Because the annihilation event occurs along the
line that connects the locations of the recorded pair of photons, the PET image is reconstructed from a collection of
these recordings. The use of tracers makes PET a powerful
research tool. Examples of its uses include the study of neurotransmitters, tumors, and Alzheimers disease.
In ultrasonic imaging, high-frequency mechanical waves
are reflected and scattered by the soft tissues of the body and
the echoes of backscattering are captured and displayed as
real-time moving images (49,50). Better resolution of images
is achieved at higher frequencies. However, the depth of penetration decreases. Therefore, the choice of frequency depends
on the application. Typical ultrasonic frequencies are 2 MHz
to 5 MHz for deep penetration, 20 MHz to 50 MHz for shallow
penetration with high resolution, and up to 200 MHz for examining subcellular structures with ultrasonic microscopy.
Some ultrasound applications are based on transmission of
the ultrasound waves rather than on backscattering. Advantages of ultrasonic imaging include lower cost of equipment,
portability of equipment, minimal use of expendables, and the
absence of harmful side effects.
Virtual reality is beginning to play a part in medicine that
will only expand as advances are made in computing techniques and less expensive equipment becomes available. Virtual surgeries, used as a teaching aid in lieu of cadavers or
animals and as a tool for practicing professionals to improve
skills or preplan procedures, can improve performance in surgery by providing a risk-free method for training surgeons to
deal with possible complications. Virtual reality can also play
an important role in medical informatics, including telemedicine and telesurgery (remote surgery) (51).
MEDICAL INFORMATICS
Biomedical engineers working in medical informatics develop
computer databases and networks that contain patient-related information (5255). This information facilitates healthcare delivery and assists in clinical decision making. Two
prime examples are the hospital information systems (HIS)
and computer-based patient medical records (CBPMR). The
HIS database encapsulates all of the information regarding
patients, not just a limited departmental or clinical view. A
modern HIS database includes (1) the entire clinical record of
a patient, including all inpatient and outpatient procedures;
(2) all patient charges and financial information; (3) admission, transfer, and release information (hospital bed control);
(4) patient management (prescribed therapy) information;
and (5) clinical decision making functions. The CBPMR is an
electronic form of a patients medical record that includes radiological and pathological images. It has advantages of accessibility and ease in information retrieval over the typical
paper medical record. The CBPMR database supports clinical
decision-making functions by assisting in patient treatment
with suggestions for diagnosis and further testing and by providing therapeutic protocols and alerts for possible drug interactions. Confidentiality of patient information is protected by
having different layers of access available to users with different privileges.
The technological advances in computers and telecommunications have brought about the field of telemedicine. The
CBPMR contains the entire patients record, including images, which can be transferred electronically to consulting
physicians in distant locations. The benefit of telemedicine becomes apparent when considering patients in areas without
major hospitals and medical universities who need the expertise of the medical profession to analyze digital images, such
as magnetic resonance images. One area of concern involves
the quality of digital representations, such as scanned X-rays.
PHYSIOLOGICAL MODELING, SIMULATION, AND CONTROL

Modeling attempts to find the most simplified method for accurately defining a system. Physiological modeling, or biomodeling, assists in (1) research by verifying hypotheses or
indicating areas needing further study, (2) teaching and
training in medical schools, and (3) clinical applications by
aiding in such areas as diagnosis, determination of drug regimens, or design of biomedical devices, including prostheses or
drug delivery systems (56). Typically, these models are continuous models and some use artificial intelligence and neural modeling.
Biomedical engineers who model physiological systems
must have (1) an in-depth understanding of the physiology,
anatomy, biochemistry and biophysics of the physiological
system being modeled; (2) knowledge of instrumentation,
methods of measurement, and sources of data for important
parametric and system variables; (3) a background in applied
mathematics, such as ordinary differential equations (ODEs),
partial differential equations (PDEs), and statistics; and (4)
experience with computer hardware and software, including
differential equation solving and compiler languages (56).
Models of physiological systems need to consider transport
phenomenon associated with the system under consideration.
Transport mechanisms in the body include momentum, mass,

energy, and information transport. Momentum transport is
considered when modeling blood flow. Mass transport deals
with the flow of various substances, such as oxygen, carbon
dioxide, and pharmaceuticals, that are carried in the blood,
air, food and digestive juices, and urine, and with the diffusion of these substances into and out of air, blood, and tissues.
Energy transport refers to the mechanisms the body uses to
deal with heat energy. Energy transformation and transport
need to be considered when models involve muscle tissue. The
transmission of information through nerves or hormones is
what is meant by information transport.
A typical modeling method for quantifying the kinetics of
materials in the body via production, distribution, transport,
utilization, or substratehormone control interactions involves compartmental analysis (57). One example of compartmental analysis is a model of the kinetics of a pharmaceutical
in the blood stream. These models treat any part of the physiological system which can be considered homogeneous, as a
compartment, and the system that is being modeled is segmented into a finite number of these compartments. The direction of flow of material between these compartments is determined and then modeled with differential equations.
Unlike modeling, simulation attempts to reproduce the experimental data without trying to identify the mechanisms responsible for the experimental observations (58).
Closed-loop drug delivery (CLDD) systems represent a
practical application of control (59). CLDD systems are used
for therapeutic and diagnostic purposes. For example, an infusion pump administers a drug to the patient, the patients
response is sent to a monitor, and the monitor feeds the information to a controller which determines the next infusion
rate for the patient and adjusts the pump accordingly. The
control laws typically applied to CLDDs are proportional-integration-derivative (PID), adaptive, and fuzzy control. Adaptive is the most prevalent. In the clinical use of these systems,
a supervisor is present to override control in case of unphysiological disturbances, such as a change in drug concentration.
PROSTHESES AND ARTIFICIAL ORGANS

A device that is an artificial substitute for a body part,
whether it is a limb or a heart valve, is called a prosthesis.
When the prosthesis replaces all or part of an organ, it is
called an artificial organ. Though replacement of organs from
donor transplants is a more straightforward and reliable
method, the supply of donor organs and thus their use is limited. Artificial organs have been designed because they can be
produced in sufficient quantities to meet demand and they
eliminate the possibility of transferring infections, for example, HIV and hepatitis, from the donor to the recipient. When
designing an artificial organ, function is of primary concern
and can result in a device that bears little resemblance to its
natural counterpart. Typically, artificial organs are made
from synthetic materials not found in nature and use mechanisms different from those of the natural organ to achieve the
same function. Disadvantages of artificial organs include the
relative inability to adapt to growth, which limits their use in
children, the mechanical and chemical wear due to use, and
the bodys environment, which can limit the life of the device.
Recently the design of artificial organs has included combin-
409
ing biological material, such as organelles, cells, or tissues,

with synthetic, engineered devices. These hybrids are called
bioartificial organs (60).
Artificial hearts are primarily used as a bridge-to-transplant, that is, a temporary replacement used until a donor
organ donor is transplanted. Research continues in developing long-term, completely implanted heart replacements.
The heart-lung machine is a short-term artificial organ used
for patients undergoing transplant operations. It allows the
patient to survive the removal of the heart until the replacement organ is surgically implanted. Common prostheses for
the circulatory system are cardiac valve prostheses and vascular grafts. Concerns with these prosthetics include the formation of fibrous blood clots inside the circulatory system
(thrombi), tissue overgrowth, hemorrhage from anticoagulants, and infection (61).
The artificial lung must provide a mechanism for the uptake of O2 by the blood and the removal of CO2. It can be used
to completely replace the function of the lung temporarily
during surgery or to assist with gas exchange temporarily until the lung can heal. Artificial lungs also replace or assist
lung function permanently, if necessary. Typically, artificial
lungs are not placed where the natural lung is located so the
blood in the pulmonary system must be diverted to the artificial lung and pumped to return it to the heart and systemic
circulation. Gas is commonly exchanged by using membrane
oxygenators. Difficulties in design include developing membranes as thin as the walls of the alveoli and finding a blood
distribution method that mimics the branching achieved in a
short distance by the lung (62).
One kidney can sustain function for a lifetime which
makes live kidney donation possible: however, donors are typically cadavers. The artificial kidney provides a common intermittent treatment for renal failure during diminishing
function of the kidneys or for patients who are waiting for a
donor kidney. Dialysis, the mechanism of the artificial kidney,
performs the necessary functions of the kidneys. These involve regulating (1) the volume of the blood plasma (contributing significantly to the regulation of blood pressure), (2) the
concentration of waste products in the blood, (3) the concentration of electrolytes (Na, K, HCO3, and other ions) in the
plasma, and (4) the pH of plasma (63). More aggressive dialysis of the peritoneum, the membrane surrounding the body
cavity and covering some of the digestive organs, is a recently
developed treatment for irreversible end-stage kidney failure (64).
The main concern with the loss of liver function is loss of
the ability to detoxify the blood. Therefore, devices which augment liver function focus on methods of detoxification. Some
procedures currently in practice or under investigation involve dialysis, filtration, absorbent materials, and immobilized enzymes to convert specific toxins to less harmful substances. Currently, temporary replacement of the liver
involves systems with mammalian hepatocytes (liver parenchymal cells which remove most of the carbohydrates, amino
acids, and fat from the digestive products absorbed from
the intestines by the blood) attached to a synthetic support, where input from the host is separated from the device
by a semipermeable membrane. Bioartificial livers using
functional hepatocytes in a device immersed in body fluids
are being investigated as an alternative to organ replacement (65).
410
Partial or complete removal of the pancreas can occur due

to polycystic disease, trauma, or tumors. The replacement artificial pancreas focuses on the hormonal or endocrinal activity of the pancreas (i.e., insulin and glucagon secretion),
which regulates the uptake and release of glucose. Devices
have not yet been developed that can replace the exocrine
function of the pancreas, namely, the secretion of proteolytic
and lypolytic enzymes in the gastrointestinal tract. Other artificial organs for the digestive system include trachea replacements, electrical and pneumatic larynxes, which replace
only the phonation function of the larynx because a complete
artificial organ that restores respiration and protection of the
lower airway during swallowing has yet to be designed, and
extracorporeal and intraesophageal stents (66).
Skin replacement following loss from events, such as a fire
or mechanical accident, or through conditions, such as skin
ulcers, is achieved by using autographs of the patients skin,
allographs from cadavers, xenographs from animals, or artificial skin. The risk of viral infection and rejection are concerns when using allographs and xenographs. Artificial skin
is a bilayer membrane whose top layer is a silicone film that
controls moisture and prevents infection and whose bottom
layer consists of a porous, degradable copolymer. The top
layer is removed and replaced by an autograph after about
two weeks, and the bottom layer is removed by complete degradation after it induces the synthesis of new dermis. Clinical
studies have shown that autographs take better than artificial skin, but donor sites in which the top layer has artificial
skin instead of silicone film heal faster and appear more like
the patients skin than donor sites that used autographs (67).
REHABILITATION ENGINEERING
A rehabilitation engineer designs and develops technologies
that augment or replace impaired sensory, communication, or
motor systems. A device that augments an impaired function
is called an orthosis, and a replacement device is called a
prosthesis. Rehabilitation engineering is concerned with restoring the ability to perform activities of daily living (ADL),
such as (1) eating, brushing teeth, and reading; (2) public
transportation and building accessibility; (3) personal mobility; (4) sensory disabilities, such as impaired sight or hearing;
and (4) communications. In addition to the biology, physiology, and engineering involved in design, a rehabilitation engineer needs to consider the social, financial, and psychological
impacts of a device or technology. One particular difficulty in
rehabilitation engineering is the variation in retained ability
and needs of patients. Even if a standard device exists, individual modifications are expected because of differences
among patients (68).
Traditional orthoses for sensory impairments are eyeglasses or contacts and hearing aids. The retention of some
function in the sensory system is required for these devices to
work. If vision has been completely impaired through damage
to the retina, optic nerve, or cerebral cortex, other methods
for restoring ADL have been developed. An example of this is
the development of Braille to allow the visually impaired to
read. Advances in computing make scanning text and conversion into either voice or Braille (by the movement of a matrix
of pins) available as other possible reading methods. Many
deaf individuals use their vision and sign language as a sub-
stitute for speech. If deafness results from damage only to the

cochlea, implants are available which do not usually completely restore hearing but give the ability to sense environmental sounds. Advances in computing with translation aids
that are marketed to travelers could assist the deaf by capturing and displaying spoken phrases in a language known to
the individual.
The wheelchair provides a primary replacement for loss of
motor skills in the lower body. Although wheelchairs provide
great mobility, they also require special access, such as ramps
and elevators in lieu of stairs, and terrain amenable to rolling. Increased mobility has resulted from development of
hand controls for most major methods of transportation (cars,
vans, airplanes, etc.). Artificial limbs commonly replace lost
limbs. The design of these orthopedic prostheses involves selecting materials that provide weight and strength similar to
those of the limb that is being replaced along with a slow
yielding mode of failure, determining the method of attachment to the body for stability and appropriate load distribution, including the appropriate mobility or motion for the
limb, and making it cosmetically acceptable to the recipient.
A prime example of an orthopedic orthotic is the brace for
such areas as the neck, limbs, and feet. The use of external
power and control has led to improvements in orthotics and
prosthetics by restoring hand functions and providing active
limbs which assist in ease and speed of locomotion.
Communication disorders that result from damage to the
larynx are currently relieved only through the use of an external artificial larynx or through a device which converts typing
to speech. An implantable artificial larynx was recently used
to restore speech to a man who had lost his larynx due to
injury. For individuals with impaired motor skills, words and
concepts can be communicated through the use of symbol or
letter boards. There is active research to determine if information may be obtained from the speech of individuals who
produce sound that is difficult to understand.
CONCLUSION
A very brief overview of many of the areas that are currently
important in biomedical engineering has been presented.
However, this is a very diverse field that is constantly expanding. Future developments will occur in nanofabrication,
microelectromechanical (MEM) technology, sensory replacements (e.g., the artificial retina), engineered tissues, molecular electronics, low-cost medical devices that will help improve health care without increasing health care costs, and
other emerging areas.
BIBLIOGRAPHY
1. J. D. Bronzino, Biomedical engineering, in Encyclopedia of Applied Physics, New York: VCH, 1991, Vol. 2.
2. H. S. Wolff, Biomedical Engineering, New York: McGraw-Hill,
1970.
3. J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca
Raton, FL: CRC Press, 1995.
4. J. B. Park and R. S. Lakes, BiomaterialsAn Introduction, 2nd
ed., New York: Plenum, 1992.
5. J. B. Park, Metallic biomaterials, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton, FL: CRC Press,
1995.
6. P. K. Bajbai and W. G. Billotte, Ceramic biomaterials, in J. D.
Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton, FL: CRC Press, 1995.
7. H. B. Lee, S. S. Kim, and G. Khang, Polymeric biomaterials, in J.
D. Bronzino (eds.), The Biomedical Engineering Handbook, Boca
8. C.-C. Chu, Biodegradable polymeric biomaterials: An overview,
in J. D. Bronzino (ed.), The Biomedical Engineering Handbook,
Boca Raton, FL: CRC Press, 1995.
9. A. A. Biewener (ed.), Biomechanics (Structures and Systems): A
Practical Approach, Oxford: Oxford Univ. Press, 1992.
10. T. R. Canfield and P. B. Dobrin, Mechanics of blood vessels, in J.
D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca
11. R. B. Davis III, Musculoskeletal biomechanics: Fundamental
measurements and analysis, in J. D. Bronzino (ed.), Biomedical
Engineering and Instrumentation-Basic Concepts and Applications, Boston: PWS Engineering, 1986.
12. R. L. Lieber and T. J. Burkholder, Musculoskeletal soft tissue
mechanics, in J. D. Bronzino (ed.), The Biomedical Engineering
Handbook, Boca Raton, FL: CRC Press, 1995.
13. R. Aston, Principles of Biomedical Instrumenation and Measurement, New York: Macmillan, 1990.
14. J. J. Carr and J. M. Brown, Introduction to Biomedical Equipment
Technology, 3rd ed., Upper Saddle River, NJ: Prentice-Hall, 1998.
15. J. G. Webster (ed.), Medical Instrumentation: Application and Design, 3rd ed., New York: Wiley, 1998.
16. W. Welkowitz, S. Deutsch, and M. Akay, Biomedical Instruments:
Theory and Design, 2nd ed., San Diego, CA: Academic Press,
1992.
17. D. L. Wise, Bioinstrumentation: Research, Developments and Applications, Boston, MA: Butterworth, 1990.
18. L. Cromwell, F. J. Weibel, and E. A. Pfeiffer, Biomedical Instrumentation and Measurements, 2nd ed., Englewood Cliffs, NJ:
Prentice-Hall, 1980.
19. J. Malmiuvo, Biomagnetism, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton, FL: CRC Press, 1995.
20. R. Patterson, Bioelectric impedance measurements, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton,
FL: CRC Press, 1995.
21. J.-P. Morucci et al., Biomedical impedance techniques in medicine, Crit. Rev. Biomed. Eng., 24: 223679, 1996.
22. E. A. H. Hall, Biosensors, Englewood Cliffs, NJ: Prentice-Hall,
1991.
23. M. R. Neuman, Physical measurements, in J. D. Bronzino (ed.),
The Biomedical Engineering Handbook, Boca Raton, FL: CRC
Press, 1995.
24. M. Akay, Biomedical Signal Processing, San Diego, CA: Academic
Press, 1994.
25. M. Akay (ed.), Time Frequency and Wavelets in Biomedical Signal
Processing, New York: IEEE Press, 1998.
26. J. Dempster, Computer Analysis of Electrophysiological Signals,
San Diego, CA: Academic Press, 1993.
27. W. J. Tompkins, Biomedical Digital Signal Processing, Englewood
Cliffs, NJ: Prentice-Hall, 1993.
28. R. E. Ziemer, W. H. Tranter, and D. R. Fannin, Signals and Systems: Continuous and Discrete, 3rd ed., New York: Macmillan,
1993.
29. A. Cohen, Biomedical signals: Origin and dynamic characteristics; frequency-domain analysis, in J. D. Bronzino (ed.), The Bio-
411
medical Engineering Handbook, Boca Raton, FL: CRC Press,

1995.
30. L. T. Mainardi, A. M. Bianchi, and S. Cerutti, Digital biomedical
signal acquisition and processing, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton, FL: CRC Press,
1995.
31. J. E. Smith, Biotechnology, 3rd ed., Cambridge, UK: Cambridge
Univ. Press, 1996.
32. F. B. Rudolph and L. V. McIntire (eds.), BiotechnologyScience,
Engineering, and Ethical Challenges for the Twenty-First Century,
Washington, DC: Joseph Henry Press, 1996.
33. P. M. Doran, Bioprocess Engineering Principles, London: Academic Press, 1995.
34. E. S. Lander and M. S. Waterman (eds.), Calculating the Secrets
of Life, Washington, DC: National Academy Press, 1995.
35. A. J. Russel and C. Vierheller, Protein engineering, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton,
36. S. Sundaram and D. M. Yarmush, Monoclonal antibodies and
their engineered fragments, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton, FL: CRC Press, 1995.
37. C. B. Lawrence (ed.), Managing data for the Human Genome
Project, IEEE Eng. Med. Biol. Mag., 14 (6): 688761, 1995.
38. F. Berthiaume and M. L. Yarmush, Tissue engineering, in J. D.
39. J. D. Bronzino, Clinical engineering: Evolution of a discipline, in
J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca
40. D. O. Cooney, Biomedical Engineering Principles: An Introduction
to Fluid, Heat and Mass Transport Process, New York: Marcel
Dekker, 1976.
41. R. E. Shroy, Jr., M. S. Van Lysel, and M. J. Yaffe, X-ray, in J.
42. K. K. Shung, X-ray, in K. K. Shung, M. B. Smith, and B. M. W.
Tsui (eds.), Principles of Medical Imaging, San Diego, CA: Academic Press, 1992.
43. I. A. Cunningham and P. F. Judy, Computed tomography, in J.
44. S. Conolly et al., Magnetic resonance imaging, in J. D. Bronzino
(ed.), The Biomedical Engineering Handbook, Boca Raton, FL:
CRC Press, 1995.
45. M. B. Smith, K. K. Shung, and T. J. Mosher, Magnetic resonance
imaging, in K. K. Shung, M. B. Smith, and B. M. W. Tsui (eds.),
Principles of Medical Imaging, San Diego, CA: Academic Press,
1992.
46. X. Zhou and G. A. Johnson, Magnetic resonance microscopy, in J.
47. B. Y. Croft and B. M. W. Tsui, Nuclear medicine, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton,
48. T. F. Budinger and H. F. VanBrocklin, Positron-emission tomography (PET), in J. D. Bronzino (ed.), The Biomedical Engineering
Handbook, Boca Raton, FL: CRC Press, 1995.
49. K. K. Shung, Ultrasound, in K. K. Shung, M. B. Smith, and B.
M. W. Tsui (eds.), Principles of Medical Imaging, San Diego, CA:
Academic Press, 1992.
50. R. L. Goldberg et al., Ultrasound, in J. D. Bronzino (ed.), The
Biomedical Engineering Handbook, Boca Raton, FL: CRC Press,
1995.
412
BIOMEDICAL NMR
51. W. J. Greenleaf, Medical applications of virtual reality technology, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton, FL: CRC Press, 1995.
52. T. A. Pryor, Hospital information systems: Their function and
state, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton, FL: CRC Press, 1995.
53. J. M. Fitzmaurice, Computer-based patient records, in J. D.
54. M. H. Loew, Informatics and clinical imaging, in J. D. Bronzino
CRC Press, 1995.
55. S. Sengupta, Computer networks in health care, in J. D. Bronzino
CRC Press, 1995.
56. V. C. Rideout, Mathematical and Computer Modeling of Physiological Systems, Englewood Cliffs, NJ: Prentice-Hall, 1991.
57. C. Cobelli and M. P. Saccomani, Compartmental models of physiological systems, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton, FL: CRC Press, 1995.
58. J. L. Palladino, G. M. Drzewiecki, and A. Noordergraaf, Modeling
strategies in physiology, in J. D. Bronzino (ed.), The Biomedical
Engineering Handbook, Boca Raton, FL: CRC Press, 1995.
59. E. A. Woodruff, Clinical care of patients with closed-loop drug
delivery systems, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton, FL: CRC Press, 1995.
60. P. M. Galletti, Prostheses and artificial organs, in J. D. Bronzino
CRC Press, 1995.
61. A. P. Yoganathan, Cardiac valve prostheses, in J. D. Bronzino
CRC Press, 1995.
62. P. M. Galletti and C. K. Colton, Artificial lungs and blood-gas
exchange devices, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton, FL: CRC Press, 1995.
63. S. I. Fox, Human Physiology, 5th ed., Dubuque, IA: Wm. C.
Brown, 1996.
64. P. M. Galletti, C. K. Colton, and M. J. Lysaght, Artificial kidney,
in J. D. Bronzino (ed.), The Biomedical Engineering Handbook,
65. P. M. Galletti and H. O. Jauregui, Liver support systems, in J.
66. P. M. Galletti et al., Artificial pancreas, in J. D. Bronzino (ed.),
The Biomedical Engineering Handbook, Boca Raton, FL: CRC
Press, 1995.
67. I. V. Yannas, Artificial skin and dermal equivalents, in J. D.
68. C. J. Robinson, Rehabilitation engineering, science, and technology, in J. D. Bronzino (ed.), The Biomedical Engineering Handbook, Boca Raton, FL: CRC Press, 1995.
TAMARA L. TATROE
SUSAN M. BLANCHARD
North Carolina State University
BIOMEDICAL TELEMETRY
BIOTELEMETRY
TELEMETRY IN MEDICINE
MEDICAL TELEMETRY
The word biotelemetry is actually an abbreviation of
biomedical telemetry, which means that the measurement
of different electrical and nonelectrical values measured
on human or animal subjects are transmitted contactless
to a stationary place. When we refer to measurements on
human subjects, we are generally talking about patients
in hospitals or as follow-up in the rehabilitation process,
where freedom of movement is required. Biotelemetry also
can be applied to athletes. The most measured physiological variables are heart and brain potentials known as electrocardiogram (ECG) and electroencephalogram (EEG)
and, for athletes, muscle potentials [i.e., electromyography
(EMG) and accelerations or forces]. In hospitals, in intensive care units or in wards where patient monitoring is performed, the radio frequency (RF) telemetry often is applied
between the free-moving patient and the bedside monitor
or central unit. The RF telemetry allows the patient more
freedom and eliminates the need for nurses to be present
when the patient needs to leave the bed. The nonelectrical values that are most often measured are ECG blood
pressure, temperature, oxygen saturation, myocardial ischemia (by ST-level), and respiration. Biologists and human ecologists have great interest in measuring similar
values for animals. Biotelemetry allows us to receive quantitative data from freely moving animals in their normal
unconned environment without disturbing the transmitter, which must be small and lightweight in relation to the
animals size and should not exceed 1% of the body weight.
For the ecologists and biologists, the study of homing and
migration of birds and terrestrial and aquatic animals is
of great interest.
Besides telemetry over great distances, biotelemetry
over very short distances is also realized when a transmitter is implanted under the skin or swallowed by the patient
or animal. In the latter case, a receiver loop antenna is located over the patient close to the place where the transmitter is implanted. The different processes in the patients
intestines such as pressure, temperature, or pH-value can
be investigated and measured. Swallowed transmitters are
usually called endoradiosondes. This sonde passes through
the stomach and all the intestines transmitting data until
it reaches the exit.
BIOTELEMETRY SYSTEMS
A biotelemetry system consists of a transmitter and a receiver with a transmission link between, as shown in Fig.
1. Transmitted information can be a biopotential or a nonelectric value like arterial pressure, respiration, or body
temperature and pH value. The transducers convert nonelectrical values into electrical signals. However, for biopotential measurements, adequate electrodes must be used.
The voltage at the output of the transducer or on the electrodes is very low (in the range of a few microvolts to about
10 mV), which must be amplied by an amplier. This
measured signal must change the amplitude or frequency
of the carrier [i.e., high-frequency (HF) signal that carries information and enables transmission]. This process
is called modulation. The modulated HF signal on the receiver side usually is low, depending on the distance between the transmitter and the receiver and is inuenced
by outside noise. The receiver must rst amplify this low
signal directly or by the heterodyne principle and then demodulate it by a demodulator to obtain the same signal
waveshape as on the transmitter input. The same process
occurs between a radio station and a radio receiver, except
that the microphone is replaced by a transducer and the
loudspeaker is replaced by a penrecorder or some other display. For HF transmission, two antennas are required on
the transmission and receiving side, which must be in resonance with the carrier frequency. The length of the antenna
is usually one-quarter wave length /4 for a whip antenna
and /2 for a dipole antenna. In biotelemetry, loop antennas or coils for implants are also in use. This is the wireless
transmission realized by RF electromagnetic waves.
Wireless transmission can also be accomplished by using infrared radiation or ultrasound. Biotelemetry over
wires is used less frequently because the patient cannot
move about freely, being limited to short distances. However, if the patient is stationary, telephone lines can be
used for long distances. Telephone lines are used also in
telemedicine where pictures can also be transmitted over
the Internet. The wire and wireless systems can be combined in monitoring a patient at home, with the short-range
wireless system connected to the telephone system. However, telemetry over wires, while trivial, will not be considered here.
MODULATION MODES
Between amplitude modulation (AM) and frequency modulation (FM), FM has the explicit advantage. AM is sensitive
to the patients movement, which changes the signal amplitude as well as the reections from the walls inside the
room. AM is also very sensitive to noise. FM, in contrast,
does not have these drawbacks because the information is
included in the frequency changes of the carrier frequency
f0 , and the amplitude changes of the signal do not inuence the transmitter information. The deviation of the frequency f from the carrier frequency f0 is proportional to
the signal voltage. In addition, FM electronic circuits can
be very simple if only short distances are to be spanned
(Fig. 2).
For greater energy saving pulse modulation (PM) is
more appropriate because of low duty cycle. In this case
the consuming up power is much lower and is reduced
mostly between 0.05 to 0.005 times in comparison to continuous power supply. This means that the pulse current
can be 20 to 200 times stronger than average current con-
J. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering. Copyright 2007 John Wiley & Sons, Inc.
Medical Telemetry
Figure 1. Biotelemetry system: transmitter and receiver (heterodyne and xed-frequency).
Figure 2. Amplitude, frequency, and pulse modulation.
sumption. The information or measured value can change

the pulse amplitude, which is called pulse-amplitude modulation (PAM), or the pulse frequency, which is called
pulse frequency modulation (PFM). When the pulse width
changes but the frequency and interval between pulses do
not change, this modulation is called pulse width modulation (PWM). However, when the time interval t between
pulses changes and the pulse width does not change, this
is called pulse position modulation (PPM). The third possibility is pulse code modulation (PCM), wherein one amplitude (8 or 12 pulses) is used, depending on the resolution required. The n-bit resolution has 2n quantum levels.
The pulse width is equal in this case to the interval between pulses. PPM is the most energy efcient because
short pulses with the lowest duty-cycle can be created, and
only two pulses are required for each sample of the signal.
Those types of modulation are shown in Fig. 2.
It is quite evident from the previous discussion that the

transmitter is a crucial part of biotelemetry systems, because it is carried by the patient or the animal and, consequently, should be small and lightweight. The transmitter
should not get in the animals way, especially for an animal,
which is not aware of carrying such an object. It is obvious
that the transmitter can be much bigger when carried by
an elephant, bear, or horse than by a small bird or insect
like a grasshopper or honey bee, which may weigh only
2 g. Moreover, Winter (1) reports that an ECG transmitter made in surface-mount SMD technology, which weighs
0.45 g, consumes 15 A and has a range of 2 m to 5 m. On
the other hand, the receiver has no size or weight restrictions because it is, in most cases, stationary. The reverse
situation occurs during biostimulation when the receiver is
implanted inside the body and must be as small as possible,
whereas the transmitter has no size or weight restrictions.
Medical Telemetry
Figure 3. (a) Frequency, (b) time division modulation, and (c) amplitude shift keying.
If the signal must be transmitted from a small animal or

if a transmitter has to be implanted, in which case small
size is required, only one signal or channel is often measured and transmitted. However, in a clinical application,
often more variables must be measured, and a multichannel system is required (1). When more measured data need
to be transmitted over only one link or medium, each separately measured type of information is called a channel.
To transmit more data simultaneously, each channel can
have its own carrier frequency, which is called frequency
division multiplexing. However, frequencies from one channel should not enter into the frequency domain of another
channel [Fig. 3(a)]. The channels can be transmitted in
time sequences which is called time division multiplexing
and are represented in Fig. 3(b). To know when to start a
series of channels, at least one synchronization pulse is required. In biotelemetry, time multiplexing is mostly used.
When an RF electromagnetic wave is transmitted, each
pulse must modulate an RF frequency oscillator in on-off
key mode. The same is true for ultrasonic waves (2, 3). In
each case, a pulse of duration tp must span a few periods
n of the transmitted signal [i.e., the fosc n/tp as shown in
Fig. 3(c)]. In infrared telemetry, such an oscillator is not required, and only a light-emitting diode emits during pulse
duration, which is simpler (4).
A simple three-channel frequency multiplex system is
shown in Fig. 4(a), which can be extended by adding
more channels. Two channels are provided for nonelectrical quantities like temperature and pressure, and one for
electrical values. Nonelectrical quantities are converted by
transducers into electrical quantities (voltage). After necessary amplication, these signals modulate, a crystal os-
cillator by means of a varicap diode. The series resonant

frequency of quartz crystal can be changed slightly (no
more than f/f0 = 0.5 103 ) by changing the capacity of the
varicap diode by the measured voltage as shown in Fig.
4(b). Each oscillator oscillates on a different frequency f01 ,
f02 , f03 so that multiplexing by frequency division is realized (5). The oscillators output over the antenna coils are
summed and transmitted over a common antenna. Instead
of using varicap diodes, a voltage-controlled multivibrator
can be used as the subcarrier, as shown in Fig. 4(c).
Multichannel transmission is often used in electroencephalography, but different physical values like heart potentials, temperature, respiration, and miography can also
be measured simultaneously (5). Multichannel telemetry is
also applied in telemetry by gait analysis (i.e., in each case
even more different values must be transmitted simultaneously). With respect to energy saving, time multiplexing is
preferable especially if the low-duty cycle can be obtained
(6).
In this regard pulse position modulation is most promising, because only two pulses are required to restore an amplitude (Sync. pulse is not included). Using PCM for 8-bit
quantization, four pulses are required on average, which
means that twice as much energy is consumed and even
more is used for 10- or 12-bit quantization. PWM is less
accurate because of the possibility of pulse shape distortion, which may occur in long-distance transmission.
In Fig. 5(a) a transmitter using PFM is shown. The
pulses at the output drive IR-emitting diodes if infrared
(IR) biotelemetry is used; however, an HF oscillator can
also be modulated in on-off key mode. This telemetry system is used for force measurement in legs and crutches
Medical Telemetry
Figure 4. (a) Frequency division three-channel multiplex, (b) three-channel FM transmitter, and
(c) voltage-to-frequency converter with multivibrator.
and has six force transducers in the input (6). The signals
from the transducers are amplied and fed into the multiplexer (MUX). A clock circuit controls a series of binarycoded pulses which are sent to pins A, B, C through a
counter. These inputs to the multiplexer sequentially select the channel (1, 2, 3, 4, 5, 6) to be connected to the
comparator. The central unit also sends the pulses from
the clock to a monostable multivibrator MM1, which sends

the start pulse to the transmitter output stage and also
starts the ramp voltage. When the voltage for a selected
channel is equivalent to the ramp voltage, the comparator
sends the second pulse. We dene t as the time interval
between proportional to the measured voltage. After this,
the multiplexer chooses the next channel, and the proce-
Medical Telemetry
Figure 5. (a) Six-channel transmitter with PPM and (b) time diagram of PPM.
dure is repeated. In this way, pulse position modulation

is realized. Each sequence of six channels is followed by a
synchronization series consisting of seven pulses. The time
diagram is shown in Fig. 5(b). The transmitted pulses can
be infrared or HF signals. On the receiver side the signal
is demultiplexed.
If frequency multiplexing is employed, the signal is
transmitted continuously, as a sum of different frequencies, and in this case energy consumption is much greater.
POWER SOURCES
Primary Batteries
The power supply is a vital part of a biotelemetry system.
It must have a high enough voltage so that one cell can
power silicon transistor circuits and it must also have a
high energy capacity, which is expressed in ampere-hours
(Ah). A high ratio between cell ampere-hours and weight is

desirable. In the early days of biotelemetry, mercury cells,
with an open circuit voltage of 1.35 V which remained almost constant to the end of the battery life, were mostly
used. They had an energy density in the range of 100 Wh/kg
to 125 Wh/kg, sustained voltage under the load, and low
constant internal resistance. During their use, gas does
not evolve. The chemical reaction of mercury cells is Zn +
HgO ZnO + Hg. The mercury oxide (HgO) is the positive
terminal, zinc (Zn) is the negative terminal, and potassium
hydroxide is used as the electrolyte. There are also some
other cells like zinc-manganese dioxide cells (or carbonzinc), but they are very rarely used because of their large
size and a low energy density of 55 Wh/kg to 77 Wh/kg.
Previously mentioned power sources cannot compete
with lithium cells, which have gravity energy density up
to 330 Wh/kg, nearly three times that of mercury and silver cells. They are composed of lithium (Li) as the neg-
Medical Telemetry
ative terminal and iodine (I) as the positive terminal.

Lithium iodine (LiI) with polyvinil-piridin (P2VP) is used
as a solid-state electrolyte, which provides some advantages with respect to battery life and high-temperature
endurance, as well as a much higher voltage (2.8 V). Also,
lithiumiodine cells have the highest volumetric energy
density (600 Wh/dm3 ). Their chemical reaction is
During the discharge process, lithium and iodine are

turned into lithium iodine (LiI). This is the simplest reaction, but others are used in improved batteries with lithium
compounds like lithium-manganese dioxide with 3.0 V
open circuit voltage or lithium-thionyl chloride (SOCl2 )
with 3.5 V. Also successful is the lithium-sulfur dioxide cell
with 2.7 V open circuit voltage. Lithium batteries are currently the most used power sources in biotelemetry and
pacemakers (7).
Nickel-Cadmium Secondary Batteries
Nickel-cadmium batteries are the most commonly used
batteries among secondary batteries (i.e., those that can
be recharged repeatedlyapproximately 500 cycles with
100% depth of discharge and 2000 cycles with 50% depth
of discharge). Nickel-cadmium batteries have substantial
advantages when compared with lead-acid secondary batteries because they can be produced in the sealed form
since they do not generate gas during discharge. This make
them appropriate for electronic equipment, including implanted biotelemetry circuits. They can be overcharged and
have low internal resistance. However, when compared to
the primary batteries, their energy density is low (only 40
Wh/kg) as is their voltage (1.25 V). Their chemical reaction goes in both directions, depending on whether they
are charged or discharged
Cd is the negative terminal, and Ni(OH)3 is the positive terminal. NiCd batteries are manufactured as button cells
or cylindrical cells. Button cells are more appropriate for
small transmitters in biotelemetry. Cylindrical cells are adequate for large biotelemetry transmitters carried by patients where size is not restricted and multiple channels
with higher energy consumption are required. Rechargeable batteries can be implanted inside an animal or human
subject (some of the rst pacemakers had rechargeable batteries) and can be recharged during the night, when the
subject is sleeping, over the inductive coils (7).
Solar cells can also be applied or mounted on an animals back, but rechargeable batteries are usually used to
avoid voltage changes when the light changes intensity. Solar cells can also be placed beneath the skin, greater light
intensities are available (2).
BIOTELEMETRY APPLICATIONS
As already pointed out, the transmitter is the most important part of a biotelemetry system and usually requires
the most sophisticated design. A transmitter including its
power supply is always limited in size and weight and must

be related to the size of the subject carrying it. If the transmitter is carried by a human, it might weigh up to 400
g, which permits multichannel signal transmission, as in
the case of electroencephalography, when up to 16 channels can be transmitted, or in that of cardiology, when with
three channels the patient can be tested under different activities during cardiac exercise tests. Multichannel telemetry is applied in gait analysis measuring forces in the legs
(8), fetal heart energy examination, and in neurological
cases where the patients with epilepsy can be followed up.
In clinical multichannel telemetry, arterial blood pressure
is often measured, as is respiration. Among the interesting data is body temperature during physical activities.
Biotelemetry also enables measurements from relatively
inaccessible regions like the stomach, small intestine and
colon, where temperature, pressure, acidity (pH), and oxygen partial pressure are of interest (9). Of course, the patient should be unaware of the transmitters presence. In
this case, transmitters must be particularly small. For example, when investigating the intestines, the transmitters
are usually swallowed and are called endoradiosondes or
radio pills.
Endosondes can also be used for colon investigation and
pressure measurement in the bladder, when introduced
through the urethra. The similar endoradiosonde can be
applied into the same places in animals as in humans.
In some cases with humans, it is necessary to measure some internal variables like pressure, temperature,
or acidity in intestines, stomach, or colon when endosondes are swallowed or implanted after a surgical procedure.
A dolphin can swallow a whole sh in which a temperature transmitter is positioned in the gill (2). For measurement purposes in humans, implantation is avoided except
for electrostimulation. For animals, surgery is often preferable because the animal cannot pull or pick the transmitter
from the location. A few days after surgery, the animal is
recovered and does not feel discomfort. At this point, measurements can commence. Implanted telemetry can also be
applied on animals in zoo exhibits (2).
In animal tracking, the same values are usually measured as in humans; however, some special values depend
on the biologists particular interest. Here, not only must
the size of the transmitter be related to the size of the animal, but the transmitter must also be adjusted to the environment. For terrestrial animals, birds, and marine mammals, which rise to the sea surface to take a breath of air, RF
electromagnetic waves are used exclusively. Consequently,
among marine mammals, dolphins are tracked most often,
but penguins, sea lions, and whales, among others, are also
tracked. The transmitter is switched on only when they are
on the surface. Biotelemetry is also applied to birds to examine the signicance of their voice and singing (10), acceleration, and the power applied during the ight by measuring oxygen O2 consumption. Because the birds can soar,
distances over 50 km can be spanned. However, animalecologists and zoographers are most interested in studying
bird homing and migration phenomena and tracking the
birds in their daily life (1, 2). Biologists are also interested
in tracking terrestrial animals like bears, wolves, and deer
as well as different sorts of turtles in order to trace their mi-
Medical Telemetry
gratory routes. Tracking also can be performed on rabbits,

cow elk, and horses, to name a few. This kind of tracking
is called wildlife tracking. To study sh, which never come
up to the surface of the sea, ultrasonic transmitters must
be applied. The ultrasonic receiver is connected to an RF
transmitter oating on the sea surface. It receives sonic
signals and transmits RF signal farther away as a relay.
Sonic signals may be transmitted over distances of 10 m to
20 m to more than 2 or 3 km if the animal is large enough
to carry a bigger transmitter, which is described later in
more detail in the wildlife tracking section.
ther a dipole or a whip antenna (l = /4). The transmitting

whip antenna is usually shorter than l = /4, in order not to
be clumsy, which greatly diminishes the effective antenna
length hef , which is proportional to l/, see Eq. (3b). The
gain G, which is dened as the product G = D(), where
is the antenna efciency and D() the radiated power in
the direction determined by , drops faster than the l/ ratio according to the diagram in Fig. 6(c). The electric eld
E on the receiving side depends on the input power Pi and
gain G of the transmitting antenna and decreases with the
distance R according to
TRANSMISSION LINKS AND PROPAGATION

In biotelemetry, most applicable transmission links are
realized by electromagnetic waves in the radiofrequency
range or in so-called RF biotelemetry. Also, electromagnetic
waves in the infrared region with wavelengths from 0.8 m
to 1 m are used in infrared (IR) biotelemetry. Ultrasonic
transmission links are used least often. Ultrasonic transmission usually is applied to the aquatic animals because
they live under the water. However, it is also used in a
ranging system for gait analysis.
Electromagnetic Transmission
Electromagnetic transmission covers almost all areas of
biotelemetry applications. Biotelemetry practiced on humans has mostly clinical applications, but it has been used
in sports by athletes and also in some vocations where
strong physical activities are required, especially in hazardous environments (5).
In a biotelemetry system, most attention should be paid
to the transmitter, because it is carried by the subject and
should be made as small as possible. The receiver size is
not limited.
The modulation modes were already described. Frequency modulation is usually applied on small, onechannel devices because it can be realized with simple circuits. Pulse (on-off key) modulation is applied mostly in all
other cases, including multichannel systems.
The distance that can be spanned between the transmitter and the receiver depends on the strength of the electric
eld at the receiving antenna location, because the induced
voltage V in the receiving antenna is proportional to the receiving antenna effective length hef and the electric eld
strength E around the antenna
When the antenna is a dipole with l = /4, the effective antenna length hef is /, where is the wavelength of the
transmitted electromagnetic wave, as shown in Fig. 6(a,b).
This situation occurs when the transmitting antenna is either a dipole or a whip antenna (l = /4). The transmitting
whip antenna is usually shorter than l = /4, in order not
to be clumsy, which greatly diminishes the effective antenna length hef , which is proportional to l/, see Fig. 6(a,b).
This situation occurs when the transmitting antenna is ei-
This equation is valid if R is longer than the wavelength for

the far eld. For the near eld (distance R < 0.2 ), the electric eld decreases by R3 . A whip antenna is usually used
for the transmitter, whereas a dipole or long whip antenna
can be used on the receiver where the antenna dimensions
are not critical. Dipole antennas always have a span about
5% less than /2. To obtain better directivity, a dipole antenna has a number of metal rods, the so-called directors,
which are 5% to 7% shorter than the dipole and located in
front of the antenna and one or more longer rods (by about
5%) behind the antenna to act as reectors. This type of
antenna is called a Yagi antenna or array, [Fig. 6(d)]. In addition to the dipole, the loop antenna also has a directivity
feature and may have one or more turns depending on the
required inductivity. The hef for the loop antenna is given
as hef = 2 n S/, where n is the number of turns and S is
the surface of the loop (Fig. 13). With high permeability, the
directivity and hef can be greatly improved if enough space
is available. The loop antenna is also used exclusively for
implanted transmitters or swallowed endoradiosondes because it can be made small. Usually the loop antenna or
coil is the resonant circuit inductor of the endoradiosonde
oscillator.
The transmitting frequency is usually determined by
the size of the transmitter. If the size of the transmitter
must be small, the transmitting frequency must be high.
The oscillator resonant circuit of a small transmitter has
small inductance L and capacitance C, which determines
the oscillating frequency
The range of radiofrequencies used in biotelemetry lies

usually between 27 MHz and 450 MHz. Because the transmitters are small, higher frequencies are preferred. Nevertheless, low frequency is required for underwater measurements, especially in well-conducting ocean water. The
greatest possible in this case is given by an empirical formula
where d is measured in meters and f is measured in hertz.

We can easily see that when 50 kHz is applied, a distance
of only 1.07 m can be spanned. The same is true of the implanted transmitter or swallowed endoradiosonde, because
body tissue is just as highly conductive as ocean water. In
Medical Telemetry
Figure 6. (a) Whip antenna, (b) dipole antenna, (c) antenna gain as function of l/, and (d) Yagi
antenna.
this particular case, a compromise must be found between

the size and the depth of body penetration. For telemetry
inside the body, frequency modulation is used primarily in
a range between 80 MHz and 400 MHz.
To obtain longer distances especially in free space,
higher transmitting power is required. Consequently, HF
power ampliers must be employed.
Infrared Diffuse Transmission
Even though radiofrequency telemetry is exclusively used
in open spaces, infrared telemetry has some advantages in
enclosed spaces (4). Infrared biotelemetry is always working in an on-off key mode (i.e., mostly PPM or PCM in order to save energy). In this case, transmitters and photoreceivers are much simpler and do not require cumbersome
antennas and coils for tuning. The mobility of a patient carrying a transmitter is complete inside the room, and there
are no remarkable changes in the amplitude of the received
signal. IR transmission shows almost no interference from
other RF stations, power lines, or appliances. However, IR
radiation is limited to the inside of an enclosed space. The
space may be quite large when multiple receives are used
and placed in the locations where the IR radiation is the
weakest (8). The IR transmitter carried by a patient can
be moved to other rooms and used in stairwells, halls, and
wherever IR receivers are located.
To enable the users unrestricted movement, the IR radiation must ll the whole room where the IR transmitter
and receiver or receivers are located because of reection

by the walls. Because the wall roughness is greater than
the IR wavelength, each ray I0 from the transmitter is reected at the wall in all directions according to the cosine
law (Lamberts law) and Fig. 7(a),
This type of IR biotelemetry based on reection by the walls

is called IR diffuse biotelemetry (4).
In Fig. 7(b,c) an IR transmitter and receiver is presented. The receiver is working in a current mode, which
is most effective, but it can also work in a voltage mode.
To improve the signal-to-noise ratio, more photodiodes in
the receiver must be used, but a large number of photodiodes increase the capacitance, which limits the IR-receiver
frequency range. However, an optimum exists between six
and eight photodiodes (8).
The IR-diffuse telemetry range can be extended by using
multiple IR receivers in locations where IR radiation is
weakest. The output signals from all these IR receivers
are connected to a summing unit where they are added.
The summing unit with the amplier is designed in such
a way that if the signal from some receiver is too small, or
does not exist at all, that signal and the noisy output of its
receiver is eliminated from the summing unit.
IR telemetry is mostly used for patients in hospitals
transmitting ECG or EEG signals. There are also applications in gait analysis for force measurement in legs and
Medical Telemetry
Figure 7. (a) IR diffuse reection, (b) IR transmitter, and (c) IR receiver.
crutches (6, 8). IR telemetry is employed in the measurement of myographic potentials, temperature, and heart
rate for sportsmen and athletes but only inside rooms
where IR rays can be reected.
voltage V is induced according to the relation q = C V,
or
Ultrasonic Transmission
We have already shown in Eq. (6) that RF transmission is
strongly attenuated in water, so that measuring aquatic
animals in their normal habitat, where they enjoy free
movement, is not possible. In this case, only sound signals
at frequencies above the audible threshold can be used.
Sound signals transmit very well in water. The transmitting frequencies are usually low, between 50 kHz and 200
kHz, and the distances that can be spanned by them do
not exceed 200 m. The emitted power is usually not higher
than 1 W/cm2 (2, 3).
The conversion of electric energy into mechanical energy is done by piezoelectric ceramics like lead-zirconatetitanate. This material has a dual behavior. When a force F
is applied on a disc of piezoelectric material, a charge q is
induced on its electrodes according to q = F (where is a
constant), as shown in Fig. 8(a). Because there is a capacitance C = 0 (S/d) between the electrodes, where S is the
area of the electrodes and d the thickness of the ceramic, a
Also, from the Eq. (8b), where a stress = F/S is induced by

the electric eld E = V/d
In this case, the stress in ceramic material changes the

thickness d in relation to the applied voltage V. This feature is used by the ultrasonic transmitter shown in Fig.
8(b). An oscillator oscillating on frequency f0 can power
a piezoelectric transducer that emits ultrasound. The frequency of the oscillator is usually adjusted to the resonant
frequency of the piezoelectric ceramic to obtain the best efciency. The oscillations are interrupted by a multivibrator
or the oscillator that is working in the squegging mode (see
the discussion in the next section). The frequency of these
interruptions is controlled by a measured parameter like
temperature or by an electrocardiogram. In Fig. 8(c), an
ultrasonic receiver with a piezoelectric transducer and an
10
Medical Telemetry
Figure 8. (a) Piezoelectric crystal, (b) ultrasonic transmitter, and (c) ultrasonic receiver.
HF transmitter as a relay station is shown. The ultrasonic

signal, received from a transmitter on a sh, converted to
voltage, controls the frequency (VCO) of an astable multivibrator (A.M.). Now this FM signal is amplied by an HF
power amplier and emitted by an antenna as an electromagnetic wave. This relay station is oating on the ocean
surface. If the transducer is a disc, the ultrasound is transmitted in only one direction. To transmit ultrasound in all
directions, a small hollow cylinder is used [Fig. 8(a)] (diameter of 0.5 cm and about 1 cm long). With this cylindrical transducer, spherical wave spreading is obtained, and
sound intensity falls off proportionally to 1/R2 . In shallow
water where sound cannot spread in all directions, the intensity falls off by the 1/R law (3). Some dolphins are sensitive to frequencies up to 150 kHz, therefore the transmitted
ultrasound frequency must be two or three times higher so
as not to disturb them (2). Acoustic telemetry is also useful to divers for measuring heart and respiration rate and
diving depth (11).
ENDORADIOSONDES
Endoradiosondes are transmitting devices for measuring different variables within the body. They are either
swallowed or noninvasively introduced into the bladder
through the urinary tract or into the colon. Endosondes
are mainly used for measuring pressure, temperature, or
pH value (acidity) in the stomach and intestines. They also
can measure biopotential with special sensors, especially
oxygen O2 and, rarely, chlorine Cl. Endosondes can be active or passive. Active endosondes have their own power
source. Today, lithium batteries are used mostly. Passive
endoradiosondes or implants do not have their own power
source and are powered from an outside source by coils. Endoradiosondes can also be implanted after a surgical procedure inside the human body to follow up some healing
process. This type of endoradiosonde is passive, because it
must dwell inside the body for a long time and small batteries do not store sufcient energy for such a long-term
operation.
Endoradiosondes that are swallowed should be as small
as possibletheir largest parts are the batteries and coils.
They are usually very simple oscillator circuits whose frequency f0 is altered by the measured signal. This frequency modulation has low deviation f/f0 , not higher than
10%, but sometimes lower than 1%. In Fig. 9, transformercoupled, Collpits, and Hartley oscillators are shown. These
oscillators are usually applied in endoradiosondes. FM is
the simplest modulation mode in one-channel biotelemetry. The transmitted signals are received by a loop antenna
(up to 35 cm) located over the patient in a supine position.
With smaller loop antennas that are only a few centimeters in diameter, endosonde tracking can be provided because of better directivity. Endosondes are primarily used
for pressure measurement in the small intestine or bladder with inductive transducers. This inductance L is at the
same time part of the oscillators resonant circuit that determines the oscillation frequency f0 according to Eq. (5).
Medical Telemetry
11
Figure 9. Oscillators: (a) transformer-coupled, (b) Colpitts, and (c) Hartley.
The change of the oscillator frequency f0 , which is caused

by a change in inductance L or capacitance C, can be expressed by partial derivatives,
According to this formula, 20% change in inductance L

or capacitance C yields a 10% deviation in oscillation frequency. For pressure measurement, the simplest solution
is to change the inductance L of the coil by moving a ferromagnetic core inside the coil. The core is connected to a
membrane whose displacement is proportional to the outside pressure. This is depicted in Fig. 10. The inductance
of the coil can be changed by air gap displacement as well.
For temperature measurement, a temperature sensitive capacitor C(t) that changes the oscillating frequency is
used. Temperature can be measured using a thermistor (a
very sensitive temperature resistor), which determines the
frequency of the intermittent oscillations of a squegging
oscillator. Squegging oscillators have strong positive feedback and a large time constant RC in the base circuit of the
oscillator transistor. When the oscillations start increasing
and reach saturation, the oscillator feedback factor (A) becomes low, and the oscillations start to decrease to the end.
The time constant RC1 denes the time when the next oscillations will start. In this way, the period between two
subsegment oscillations t is proportional to the RC1 time
constant. The squegging oscillator is shown in Fig. 11. This
type of oscillator is also used to determine the temperature
of the human brain on the durameter following a surgical procedure, where resistor R is a temperature-sensitive
sensor (thermistor) (2). The thermistor R controls the frequency of an astable multivibrator, which keys the HF oscillator. In this way, we can measure the temperature in
women during their menstrual cycles and neonatal mesal
air ow for apnea monitoring (12).
The acidity which is often measured in the intestines
and stomach is dened by the pH value, which for acids is
between 1 and 7. For hydroxides the pH value is between
7 and 13. The pH value is dened as pH = log [H+ ]. Using
special electrodes, we can measure acidity as the potential
difference VH between them. One of the electrodes, the

so-called glass electrode, is made of a thin glass membrane
that permits the passage of only hydrogen ions, and the
obtained voltage difference VH on the glass membrane is
proportional to the pH value given by the Nernst equation:
Here R is the gas constant, T is the absolute temperature in

degrees Kelvin, F is the Faraday constant, n is the valence,
[H+ 1 ] is the measured concentration of hydrogen ions outside the glass membrane, and [H+ 2 ] is the concentration of
hydrogen ions inside the membrane, which is close to 1 (or
pH = 1). Substituting all these constants into Eq. (12),
where t is the temperature in C of the solution and VH is

measured in millivolts. The second electrode is a reference
electrode with a constant potential Vr = 0.2424 V, which is
not inuenced by the outside H ion concentration. The reference electrode is usually made of calomel (Hg2 Cl2 ), so
that the measured voltage differences VH in Fig. 12(a)
(13, 14). The electrodes are placed on opposite sides of the
endosonde, and the voltage VH between them is applied
to the voltage-sensitive capacitor or varactor (p-n diode),
which is also the capacitive arm of the resonant circuit. In
this way, it changes the oscillator frequency. The transmitter circuit with amplier and an astable multivibrator as
oscillator is shown in Fig. 12(b).
PASSIVE IMPLANTED TELEMETRY DEVICES
The largest components in all implanted electronic devices
are the battery and inductive coil. If the device must remain inside the body for a long time, the size of the battery
is the main obstacle to implantation. Passive devices have
no battery, and energy is supplied to them externally from
a battery source through the skin and other tissue. The
magnetic eld created in an external coil L1 , usually called
the energizing coil, induces a voltage in the implanted coil
12
Medical Telemetry
Figure 10. Endoradiosonde for pressure measurement.
Figure 11. Endoradiosonde for temperature measurement with temperature-sensitive capacitor

C(t) or thermistor.
L2 as shown in Fig. 13. To avoid interference with the power

supply frequency fp , the external coil L1 takes energy from
a power amplier and oscillator at a frequency that differs
from the implanted transmitters oscillator frequency fs ,
which transmits the measured data. However, the power
amplier and transmitter usually work at different times.
The power oscillator charges the capacitor Cp for a period
of time and then stops emitting energy at which point the
transmitter sends data on the frequency fs (15). The external coil can also be used to control an implanted stimulator.
The HF voltage induced in the receiving coil is rectied
to obtain direct current, which can power the implanted
data transmitter. The rectier can be connected as a voltage doubler [Fig. 14(a)] or a full-wave rectier, as shown
in Fig. 14(b). In Fig. 14(a), a temperature-measuring FM
transmitter is powered by the rectier with a temperature sensitive capacitor C in a resonant circuit. The energy
can be transmitted continuously or with interruptions. The
maximum energy will be transmitted when the coupling
factor k is high; it will also transmit Q1 and receive Q2
resonant circuit quality factors. Here L1 and L2 are inductivities of the energizing and receiving coils (Fig. 13). The
efciency factor = P2 /P1 between received power P2 and
transmitted power P1 can be calculated by
If the value k2 Q1 Q2 is much greater then 1, the efciency

is approaching 100%. However, in practice, this value cannot be realized; consequently, the optimal efciency is not
better than 20%. The value k2 Q1 Q2 can be written as
Here the resistances R1 and R2 include losses in the transmitting and receiving coils as well as the capacitors C1 and
C2 . Mutual inductivity M depend on the distance between
coils L1 and L2 . Both coils, L1 and L2 , and capacitors C1
and C2 are resonant circuits tuned to the same frequency
fp (5). The implanted circuit can last for more than one
year. For example, after a hip operation, tissue transplant
rejection or tumor development can be monitored by measuring its temperature. Also, after brain surgery of the hydrocephalus, measurement of intracranial pressure might
warn of impending danger. Besides taking measurements,
electromyographic potential could be transmitted to control an articial limb or strain gages could be used to monitor the forces in massive orthopedic implants (16, 17). Pas-
Medical Telemetry
sive implanted devices are applied for relative strain measurements with strain gages hermetically sealed in the
bone cavity for monitoring fracture healing. Also, the temperature increase in an articial joint at the hip or knee
can be monitored (17).
Some stimulators, like the cochlear stimulator (an implant for the deaf) use telemetry for measurement of stimulus amplitude and pulsewidth of any of its 16 control leads.
This implant receives energy via an external coil (18). Figure 15 shows a demodulated signal taken by an endoradiosonde for acidity measurement. It shows the pH value in
a healthy man, which changes after taking 0.5 g NaHCO3
before and after breakfast.
Note that some implanted devices (even nonpassive)
also have small rechargeable cells (mostly NiCd), which
can be recharged with magnetic coils when necessary. In
this way, the transmitter can work continuously.
ACTIVE IMPLANTED TELEMETRY DEVICES

Active implanted telemetry devices are powered by the battery. The telemetry transmitter is made up by a frequency
(FM) or pulse coded (PCM) oscillator modulated by measured signals and a power-amplier with a resonant circuit tuned to the oscillating frequency. The inductance of
resonant circuit is made as a coil which serves as a loop antenna in this case (Fig. 9). In most cases the telemetry circuits are a small part of a composed implanted device such
as a pacemaker or stimulator. Programmable pacemakers
or stimulators are mostly controlled by RF-pulses via inductive coils outside and inside the body as is shown in Fig.
13 for energy transfer. Inside the body is the receiving coil
as a resonant circuit, which receives control pulses, which
is not telemetry. Some pacemakers have also telemetry for
sending measured data in an opposite direction from the
control pulses over the same transmitter coil used for receiving the control pulses, but in a different time interval.
That information can be stimulating the pulse amplitude
and duration or the batterys internal resistance to predict its end of life. Also, stored data of heart beat rate or
an electrocardiogram can be transmitted. The modulation
type most used is the amplitude on-off pulse modulation
or frequency modulation with carrier frequency between
30 kHz and 175 kHz. Pacemaker telemetry transmitter is
switched on only to send required data and can be activated
over a receiving coil by coded RF-pulses or by a reed-relay
built into the pacemaker which is responding via magnetic
pulses or permanent magnet.
For fetal studies before and during birth, a transmitter with coil for multichannel transmission of fetal ECG
and heart rate can be used. In some cases also brain voltage (EEG) can be transmitted by telemetry. This contactless telemetry measurement is more convenient than if
the electrodes are placed on the maternal abdomen where
interference between maternal and fetal ECG usually occurs. In this case frequency modulation (FM) is applied, or
FMFM modulation for multichannel transmission. These
measurements are performed on the fetuses of dogs, sheep,
or monkeys, and even dolphins (2).
13
Biotelemetry is also used for monitoring the rejection

of heart transplants based on analysis of intramyocardial electrograms. This method is noninvasive in comparison to the frequent endomyocardial biopsies. Acute heart
transplant rejection can be estimated from an intramyocardial electrogram and ventricular evoked response with
the same electrodes employed for heart stimulation and for
sensing this biopotential (19).
TRANSPONDERS
A transponder is a transmitter that sends a high-frequency
pulse, when it is triggered by a pulse from a sensitive continuously active receiver associated with the transmitter.
When the information from a distant measurement transmitter is required, a pulse from the researchers location
is sent. The receiver triggers the transmitter to send the
required data over a monostable multivibration. The receivers circuits are similar to those of the passive implanted transmitters, except that the receiver is continuously powered by the batteries and the transmitter only
when it is required by the pulse from the receiver. A block
diagram of a transponder is shown in Fig. 16. The transponder is usually located on the animal from which data are
being collected. The transmitter uses a whip antenna. The
animal can either carry the transponder on a collar around
the neck or have it implanted for long-term temperature
measurement, as in pigs (20). In this case the transponder
is made as an active implanted device.
The transponder can also be used for distance measurement between the animal carrying the transponder and the
transmitter at the researchers location. The transmitter
sends an HF pulse that is received by the transponders receiver and triggers a monostable multivibrator over which
the transmitter returns the HF pulse, but on a different frequency ft than the received frequency fr at the researchers
location. The time interval t between leading edges of
the sending and receiving pulse is proportional to twice
the distance 2d between the animal and the researcher,
which when multiplied by the electromagnetic wave velocity c, gives distance d = (c/2)t. The transponder can be
adjusted so that as soon as the receiver receives the pulse
from the transmitter of the transponder, it sends the new
pulse. The whole system then becomes self-oscillating, and
the frequency of this oscillation is inversely proportional
to the distance (2).
Booster transmitters are relay stations that extend the
range of transmission. If a weak signal from an endosonde
cannot be received far from the body it is monitoring, the
receiver of the booster transmitter receives this weaker signal, amplies it, and transmits it by a more powerful transmitter to reach over a greater distance. The frequency of
a transmitted signal must be different from the frequency
on which the receiver is tuned to avoid positive feedback.
PACKAGING
Packaging is of great signicance for implantable electronics. Body uids and moisture ingress can short-circuit implanted electronic circuits. The polymers most used for
14
Medical Telemetry
Figure 12. Endoradiosonde for pH-value measurement with (a) varactor and (b) astable multivibrator as voltage-to-frequency converter.
Figure 13. Energy and data transmission by electromagnetic

coupling.
packaging are epoxy or silicon rubber. Epoxy is convenient

for coil encapsulation because, as a nonconducting material, it does not signicantly attenuate the power transmitted by the coil. However, metallic packing, mostly made
from titanium and its alloys, enables hermetic sealing of
built-in electronic circuits, but as a conducting material
it absorbs the RF power of the transmitting coil, which is
therefore usually placed outside the titanium package. In
this regard, ceramic material recently developed, enables
hermetic sealing of the transmitting coil and, as a nonconducting material, it does not permit creation of eddy currents and power attenuation in the transmitting coil. Due
to the high packing density, it is possible to integrate all
analog and digital sophisticated functions of a pacemaker
onto about a 4 mm 4 mm silicon chip. In the telemetry
transmitting devices this technology can be applied using hybrid or surface-mounting technology (SMT) to make
connections between monolithic and passive components.
Usually electronic circuitry is surrounded with the trans-
mitter coil which with the battery, if it is used, takes up

most of the available space in the implant.
WILDLIFE TRACKING
Of great interest to biological scientists is the study of homing and migrating birds and terrestrial and marine animals. In this regard, telemetry offers unbeatable advantages in conveying the position of freely moving animals in
their normal unrestricted environment without any disturbance. At the same time, additional data can be collected
from the animal (temperature, heart beat) as well as the
environment. To full this requirement, the scientist must
know the exact location of the animal at any given time
including its direction of travel, so that he can plot its position on a map.
To accomplish this task, the animal must be tagged with
a transmitter that emits a coded signal or that operates on
a particular frequency that identies the animal. The ra-
Medical Telemetry
15
Figure 14. (a) Rectier with voltage doubling and (b) full-wave
rectier.
dio transmitter carried by the animal has usually a vertical whip antenna that radiates electromagnetic waves
in all directions. The receiving antenna must be a directional antenna that has maximal gain in the direction in
which the transmitter is located and that is determined by
azimuth angle. This can be a multielement (three to ve
directors and a reector) Yagi antenna or a loop antenna
(see section entitled Electromagnetic Transmission). The
transmitter frequency for a Yagi antenna with reasonably
long elements must be about 150 MHz. By rotating a directional antenna at the researchers location A, a maximum
receiving signal must be found, as must the azimuth angle in which direction the animal is located. The maximum
receiving signal can be found with earphones or a pointerscale indicator. To nd the accurate position, a second receiver with a directional antenna must be located at some
other place B. When a maximum is found at this place and
the azimuth angle is determined, two straight lines from A
and B are plotted in the direction dened by the azimuths.
Their intersection is the location of the animal, as shown
in Fig. 17.
However, using this method to determine this location
is not very accurate, because the error in the azimuth angle may be as large as 5 . It is also possible to receive reected signals on some hilly terrain. In this case, another
location C must be used, and the systematic error needs to
be analyzed by testing the triangulation system (21). See
Fig. 17. It is easier, but less accurate, to nd the location
of the animal with only one receiver, which can nd the
direction, and a transponder as the range-measuring instrument. Refer to Fig. 16 for more details. The distance to
be spanned depends on the transmitters output power carried by the animal. Also, the spanned distance is dependent
upon the altitude at which the animal is located. Tracking
birds that are ying high or sitting on a tree requires that
much longer distances be spanned. For example, a trans-
Figure 15. The pH value changes during the breakfast and

NaHCO3 taking.
Figure 16. Transponder.
mitter carried by an animal and having a 0.1 W output

power on the ground can achieve a range of 200 m. When
the animal climbs 3 m up into a tree, the measured distance is nearly 4 km. Distances that can be spanned are
usually in a range of from a few hundred meters to about
50 km (19).
Much shorter distances can be spanned when permanent magnets or radioactive markers are used instead
of battery-powered transmitters. For detecting these tags,
sensitive magnetometers or radiation meters must be used.
16
Medical Telemetry
Figure 17. Animal tracking by directional antennas.
Magnetometers mostly employ Hall-effect sensors, or inductance detectors. Radioactive iodine can be used as the
radioactive marker. These tags are mostly intended for
short distances and small animals.
Today Global Positioning System (GPS) is employed
more and more for animal tracking. It allows scientists
to obtain the location of the animal with high accuracy
(within 25 m). GPS uses 24 satellites 20,183 km above
the earths surface and provides 24 h coverage of the entire planet. The satellites work on two L-band frequencies
(1575.42 MHz and 1227.6 MHz). The GPS users on earth
have access to at least four of the satellites. The GPS tracking system consists of an animal-carried GPS receiver and
transmitter, which is produced by many manufacturers
(22, 23). The built-in microcontroller periodically (usually
every 3 s) turns on the GPS receiver to get a position from
the satellite. The GPS receiver computes its position and
puts this information into RAM in the control unit. Then
the GPS receiver is switched off and the RF transmitter
is turned on in order to transmit these data via a digital
telemetry link to the data-logging computer system at the
remote researchers site, as shown in Fig. 18. Automatic
data recording is possible by computer, and a mapping program displays the information on the monitor. GPS gives
the latitude, longitude, and elevation of the animals location. The accuracy within 25 and 40 m can be signicantly
improved using a differential technique, the so-called dif-
ferential GPS (DGPS). DGPS can obtain accuracies within

0.5 and 5 m, but it requires a stationary GPS receiver at
a known location (reference station) to send corrections to
the mobile receiver. Neither the DGPS technique nor the
real-time kinetic (RTK) GPS system, which is even more
accurate in the centimeter range, is in wide use.
The Argos System is older and is less accurate than GPS.
The Argos system, however, employs two satellites that circle the globe at a lower altitude (850 km) seven to eight
times a day. The transmitter carried by the animal sends
the coded signal to the satellite, which responds to it giving its coordinates. The transmitter can be made small, but
its accuracy is lowbetween 150 m and 1 km. For example, the Argos system application is used in the migratory
study of dolphins. The transmitter works only when dolphins stay near the sea surface (0.75s). It has a 2 W power
output on 400 MHz and weighs 1.4 kg. The battery life span
is 1200 h (24).
Satellite communication is, except in wildlife telemetry, applied in emergency medicine from a moving vehicle to a satellite using mobile satellite communications
and then from the satellite to a xed station, in a hospital and to a doctor. After a diagnosis is established it is
sent back the same way via satellite. There are now more
than 1500 satellites in orbit which can be divided with regard to their altitude and orbits as: far geosynchronous
orbit (GEO), medium (MEO), and low (LEO) altitude orbit
satellites. Similar to GPS, the Russian space program has
launched 24 satellites (GLONASS) with an 11 h 15 min orbital period, which is considered to be far orbit. There is also
the Global Navigation Satellite System, and many others.
On satellites, there are transponders (passive) or repeaters
(active), which send received information back to the earth.
Active repeaters have onboard processors connected with
other communication electronics. In mobile satellite communication (MSC), measured physiological signals, particularly ECG and blood pressure and also a color image and
audio signal obtained from a patient, are mostly transmitted from emergency vehicles (25).
Figure 18. GPS tracking.
Medical Telemetry
BIBLIOGRAPHY
1. Y. Winter, Techniques for the laboratory construction of miniature hybrid-circuits: A 350 mg ECG-transmitter, Proc. 14th
ISOB, Biotelemetry XIV, Marburg: 1998, pp. 6570.
2. R. S. Mackay, Biomedical Telemetry, Sensing and Transmitting Biological Information from Animals and Man, 2nd ed.,
Piscataway, NJ: IEEE Press, 1993.
3. C. A. Caceres (ed.), Biomedical Telemetry, New York: Academic
Press, 1965.
4. A. Santi
c,Theory and application of diffuse infrared biotelemetry, Crit. Rev. Biomed. Eng., 18 (4): 289309, 1991.
5. V. V. Parin (ed.), Biomedical Telemetry, Moskow: Medicina,
1971.
6. A. Santi
c, M. Saban,
V. Bilas, A multichannel telemetry system
for force measurement in the legs and crutches, Proc. 10th Int.
Symp. Biomed. Eng., Zagreb: 1994, pp. 2833.
7. T. R. Crompton, Battery Reference Book, London: Butterworth,
1990.
8. A. Santi
c, M. R. Neuman, V. Bilas, Enhancement of infrared
biotelemetry for monitoring ambulatory patients over wide
areas, Proc. 14th ISOB, Biotelemetry XIV, Marburg: 1998, pp.
9196.
9. J. W. Hines, et al. Advanced biotelemetry systems for space life
sciences: pH telemetry, Proc. 13th ISOB, Biotelemetry XIII,
Williamsburg, VA: 1995.
10. T. Feuerabendt et al., Radiotelemetry on song-birds, Proc. 14th
ISOB, Biotelemetry XIV, Marburg: 1998.
11. B. Woodwared, R. Sh. H. Istepanian, Acoustic biotelemetry of
data from divers, Proc. 15th Annu. Int. IEEE Eng. Med. Biol.
Soc. Conf., Paris: 1992, pp. 10001001.
12. M. R. Neuman, P. T. Schnatz, R. J. Martin, Telemetry of basal
temperature in women and respiration in neonatales, Proc.
8th ISOB, Biotelemetry VIII, Dubrovnik: 1984, pp. 137140.
13. J. G. Webster (ed.), Medical Instrumentation, Application and
Design, 3rd ed., New York: Wiley, 1998.
14. K. Saito, et al. Telemetry capsule for measuring pH of gastric
juice, Proc. 12th ISOB, Biotelemetry XII, Ancona: 1992, pp.
415419.
15. K. Van Schuylenbergh, R. Puers, A computer assisted methodology for inductive link design for implant applications, Proc.
12th ISOB, Biotelemetry XII, Ancona: 1992, pp. 394400.
16. G. Bergmann, F. Graichen, Telemetry for long term orthopaedic implants, Proc. 9th ISOB, Biotelemetry IX,
Dubrovnik: 1987, pp. 295302.
17. S. Taylor, A telemetry system for measurement of forces in
massive orthopaedic implants in vivo, CD-ROM Proc. 18th
Annu. Int. IEEE Eng. Med. Biol. Soc. Conf., Amsterdam: 1996.
18. P. M. Meadows, P. Strojnik, Powering Sensors
with an Implanted FES System (online). Available
http://www.cdd.sc.edu/resweb/resna45.htm
19. G. Schreier, et al. A Non-invasive Rejection Monitoring System
Based on Remote Analysis of Intramyocardial Electrograms
from Heart Transplants, Proc. 17th Annu. Int. IEEE Eng. Med.
Biol. Soc. Conf., Montreal: 1995.
20. P. Wouters, M. De Cooman, B. Puers, Injectable biotelemetry
transponder for identication and temperature measurement,
Proc. 12th ISOB, Biotelemetry XII, Ancona: 1992, pp. 383391.
21. M. Taillade, Trends in satellite-based animal tracking, Proc.
12th ISOB, Biotelemetry XII, Ancona: 1992, pp. 291297.
17
22. M. Soma, A. Nakamura, M. Tsutsumi, The design of satellitelinked transmitter for migratory study of dolphins, Proc. 9th
ISOB, Biotelemetry IX, Dubrovnik: 1987, pp. 319322.
23. J. J. Cupal, L. J. Lacy, F. G. Lindzey, A GPS animal tracking,
Proc. 12th ISOB, Biotelemetry XII, Ancona: 1992, pp. 298304.
24. Anonymous, Leaets from Trimble Navigation Limited,San
Dimas, CA:Sunnyvale, CA, and Magellan Systems Corporation.
25. H. Murakami, et al. Telemedicine using mobile satellite communication, IEEE Trans. Biomed. Eng., 41: 488496, 1994.
ANTE S ANTIC
University of Zagreb, Zagreb,
Croatia
426
BIOMEDICAL SENSORS
Physiologic
system
Sensor
Processor
Display
Observer
Figure 1. Block diagram of a general biomedical instrument.
BIOMEDICAL SENSORS
A biomedical instrumentation system (Fig. 1) consists of three
main components: the sensor, the processor, and the recorder
and/or display. This article is concerned with the sensor portion of the instrumentation system. As seen in Fig. 1, the sensor is the interface between the biological system and the
electronic signal processing portion of the instrument. When
we consider a biomedical sensor, we must be concerned about
both the biological and the electronic aspects of sensor performance. Biological concerns involve the response of the biological system to the presence of the foreign sensor and the response of the sensor to the biological environment. Electronic
concerns relate to the type of signal that the sensor provides
and how this signal is interfaced to the processor portion of
the instrument. Thus, in considering biomedical sensors we
must look at both the biological and electronic performance of
this component.
Biomedical sensors are classified in many different ways,
as summarized in Table 1. Classifications are determined by
the type of biological variable measured by the sensor, the
technology used for sensing, the approach to obtaining the
output signal from the sensor, and the interface that the sensor establishes with the physiological system. All of these concerns are important in classifying sensors, but depending on
the sensor and the application, it may not be necessary to
use all of the descriptors in the columns of Table 1 for sensor
characterization. It is important to note, however, that the
ways that biomedical sensors differ from sensors used in nonbiomedical instrumentation systems are found in these classifications. Although any sensor can be described by the variable measured and the sensing technology used, their
interaction with a physiological system represents a special
characteristic of biomedical sensors that is not generally of
concern with similar conventional sensors. There are some
biomedical sensors, for example, temperature sensors, that

are identical to sensors of the same variable used for nonbiomedical applications. The application of these sensors to biological systems, not the technology, makes them unique. For
example, a conventional temperature sensor, such as a thermistor, becomes a biomedical sensor when it is incorporated in
a rapidly responding electronic oral thermometer with disposable protective sheaths.
Biomedical sensors sense physical quantities, such as displacements or pressures, and they sense chemical quantities,
such as the activity of hydrogen ions or partial pressure of
oxygen. There is a special subcategory of chemical sensors of
sufficient importance to be often listed as a separate category.
This is the bioanalytical sensor that incorporates biological
recognition reactions in sensing biochemical analytes.
There are various mechanisms whereby a sensor measures
a specific variable. It may be possible to convert that variable
directly into an electrical signal or the variable can produce
an optical or mechanical representation. Chemical or biological responses are also obtained from sensors, and this is the
case in natures own sensors, such as in the nervous system
of biological organisms.
Sensors are classified according to the way they convert a
physiological variable to the output signal. In a single-step
conversion process, the physiological variable, such as temperature, is directly converted to the output variable, such as
an electrical signal. An example of this is a thermistor that
directly converts its temperature into an electrical resistance
functionally related to that temperature. In a multistep sensor, there are intermediate variables. For example, a glucose
sensor does not directly convert the concentration of glucose
to an electrical current, but rather an intermediate step occurs where the glucose determines the concentration of another substance, such as hydrogen peroxide, which in turn is
converted into an electrical signal. Knowledge of whether a
sensor produces its output by a single- or multiple-step conversion is often useful in determining what signals interfere
with that sensors response.
The previous classifications are based on the sensor technology used to measure a variable. Biomedical sensors are
also classified in terms of how they are applied in making a
measurement. Noncontacting sensors are located remotely
from the biological signal source and do not actually touch the
biological material being measured. A radiation thermometer
is an example of such a device. Sensors are considered nonin-
Table 1. Classification of Sensors

Type of
Variable Sensed
Technology Used
for Sensing
Physical
Chemical
Bioanalytical
Electronic
Optical
Electrochemical
Mechanical
Biologic
Sensing Mechanism
Single step
Multistep
Application to
Biologic System
Noncontacting
Noninvasive
Minimally invasive
Invasive
BIOMEDICAL SENSORS
427
scribe sensors with their primary application in biomedical

measurement.
Table 2. Examples of Physical Sensors

Used in Biomedical Measurements
Sensor
Variable Sensed
Variable resistor
Strain gage
Linear variable differential
transformer
Velocimeter (laser or ultrasound)
Accelerometer
Thermistor
Thermocouple
Strain gage pressure
transducer
Load cell
Electromagnetic flow meter
Linear or angular displacement

Strain (small linear displacement)
Linear or angular displacement
Velocity
Acceleration
Temperature
Differential temperature
Static and dynamic pressure
Liquid Metal Strain Gages. The liquid metal strain gage was
described by Whitney (1) as a simple means of estimating
changes in limb volume by measuring changes in the limbs
circumference. The sensor shown in Fig. 2 consists of a thin,
compliant silicone elastomer tube filled with mercury. Electrical contacts seal the mercury within the tube at each end and
are connected to lead wires. If the tube is arranged in a
straight line as shown in Fig. 2, the electrical resistance of
the mercury column between the electrical contacts R, is
given by
Force
Flow of electrolytic liquids
vasive if they touch the body but do not enter its cavities or
tissues. Sensors placed on the skin surface, such as a transcutaneous carbon dioxide sensor, are considered noninvasive.
Minimally invasive sensors enter the body but only through
normal orifices, such as the mouth or urethra. These sensors
are often called indwelling sensors. A miniature pH sensor for
measuring gastric pH might seem very invasive to the individual on whom the measurement is made, but in fact it is
only minimally invasive because it enters a natural body cavity. Invasive sensors, on the other hand, must be placed surgically. Tissue must be incised or penetrated to position such
sensors. Sensors located within the cardiovascular system,
such as miniature intraarterial pressure transducers enter
the arterial system only by a surgical cut-down or a skin-penetrating needle. The biomedical environment is extremely
hostile especially for implanted sensors. Thus, special precautions must be made in packaging the sensors to minimize
problems resulting from this environment.
In the following sections we look at examples of biomedical
sensors based upon the variable sensed. We consider the operating principle of each sensor type and look at examples of
biomedical applications.
PHYSICAL SENSORS
A physical sensor is one that measures quantities, such as
geometrical, kinematic, electrical, force, pressure, and flow in
biological systems. Table 2 lists examples of important physical sensors for biomedical measurements. Although similar
sensing devices are used in biomedical and nonbiomedical applications, the realization of these devices as practical components is quite different depending on the application.
L
A
(1)
where is the resistivity of mercury, L is the length of the

mercury column, and A is its cross section.
As the tube is stretched, the mercury column becomes
longer but because its volume V must remain constant, its
cross sectional area must decrease according to

L0
A1 = A0
(2)
L1
where A0 and A1 are the initial and final cross-sectional areas
and L0 and L1 are the initial and final lengths of the mercury column.
Thus the change in resistance is given by
R = R1 R0 =
2
(L L20 )
V 1
(3)
where R0 is the initial resistance and R1 is the final resistance.

A characteristic of a strain gage that describes its sensitivity is the gage factor which is defined as
=
R/R0
L/L0
(4)
where L0 is the initial length of the strain gage, L is the

change in its length, R0 is the initial resistance of the strain
gage, and R is the change in the strain gage resistance when
its length is increased by L. Using (Eqs. (14), the gage factor for a liquid metal strain gage is given by
=
L1 + L0
L0
(5)
which for small displacements is approximated as 2.
;
;
;
;;;
;
Sensors of Linear and Angular Displacement
R=
A physical measurement frequently used in biomedical instrumentation is the determination of linear or angular displacement between two points. In biomedical measurements,
such displacements are frequently determined dynamically to
determine the function of an organ or organism. There are
numerous sensors applied for this measurement. Some are
applied in biomedical measurements, but others are useful
only for nonbiomedical applications. In this section, we de-
Metal
end-plug
Lead
wire
Elastomer tube
Mercury
Metal
end-plug
Lead
wire
Figure 2. Cross-sectional view of a liquid metal strain gage (1).
428
BIOMEDICAL SENSORS
The principal difference between the liquid metal strain

gage and conventional foil or wire strain gages is the compliance of the structure. Conventional strain gages can be
stretched only a very small fraction of their length and remain elastic. Liquid metal strain gages can be stretched as
much as twice their length and still return to their original
length when released.
Liquid metal strain gages are used for many applications
in biomedical research and in some cases even in clinical medicine. The main concern about these devices is that the silicone tube is fragile and breaks, thereby exposing the subject
to elemental mercury. Furthermore, oxygen diffuses through
the silicone tube and oxidizes the mercury making the electrical signals from this sensor noisy after repeated use. Nevertheless, there are several common applications for this sensor.
If one models a limb as a circular cylinder of fixed length, the
cross-sectional area and, hence, the volume of the cylinder are
determined by measuring its circumference. A liquid metal
strain gage wrapped around a limb is used to measure
changes in its circumference which then are used to estimate
changes in the limb volume (1). This is used in clinical devices
for determining whether any deep venous thromboses (blood
clots) are in patients legs. If a strain gage is wrapped around
an individuals leg and a sphygmomanometer cuff is placed
on the upper leg, inflating the cuff to a pressure between arterial and venous pressure for the patient occludes the venous
outflow from the leg but still allows blood to enter the limb
through the arteries. This causes the limb volume and, hence,
its circumference to increase until a new equilibrium point is
reached. Releasing the cuff allows the blood to leave the leg
via its veins, and the circumference returns to normal. By
plotting the length of the strain gage as a function of time, a
clinician determines the rate at which the blood leaves the
leg and whether an obstruction is present. This technique is
especially useful in evaluating patients following surgery because undetected venous thrombi can cause a pulmonary embolism which is life threatening.
Another application of the liquid metal strain gage is measuring breathing movements. By slightly stretching a strain
gage and taping it to the chest or abdomen, it is possible to
measure differences in displacement as the subject breathes.
This simple sensor provides reliable breathing signals in infants studied in the hospital (2).
Inductance Plethysmography. The inductance of a coil of
wire is approximately proportional to the cross-sectional area
within the coil:
L
=

16r
r
7.353 log10
6.386
2540
d
(6)
where L is the inductance in microhenries, r is the coils radius in millimeters, and d is the diameter of the wire in the
coil in millimeters. By placing a compliant coil around a limb
or the chest or abdomen, its inductance is proportional to the
cross-sectional area of the structure it circumscribes, and this
is used to determine volume or changes in volume. The problem with this arrangement is that a coil of wire is not as compliant as the liquid metal strain gage, and so it must be modified to become a compliant sensor. A simple way of doing this
is to form a wire in a sinusoidal pattern and attach it to an
elastic band so that the wavelength of the sinusoidal wire
pattern increases as the band is stretched. The fact that the

wire is in a sinusoidal pattern has no effect on the measurement of cross-sectional area if the sinusoid is in the plane of
the surface of the structure being measured.
This device is used for measuring breathing effort, and,
when appropriately calibrated, it provides a signal proportional to tidal volume (3,4). This makes it possible to use this
device to detect airway obstruction, when breathing motion
is still present (obstructive apnea), because the total volume
change in thorax and abdomen is unchanged.
Sonomicrometer. Sound waves propagate in soft tissue at
a velocity of approximately 1.1 mm/s. One can measure the
distance between two points in tissue by determining the time
it takes an ultrasonic pulse to propagate from one transducer
to another. The displacement is given by
d = ct
(7)
where c is the velocity of ultrasound in tissue, d is the distance between the two sensors, and t is the time it takes the
pulse to propagate from one sensor to the other. Miniature
ultrasonic transducers are made from piezoelectric ceramic or
crystal materials and are as small as a 1 mm to 2 mm cube.
These sensors are used to measure myocardial segment
length by implanting them at different points in the myocardium (5), and they also have been used to dynamically monitor the dilatation of the uterine cervix during labor (6).
Measurement of Force
The measurement of forces in biology and medicine is important in understanding the biomechanics of organisms. As
with displacement sensors, many of the force sensors used for
this are the same as force sensors used in other applications.
These force measurements are based on a load cell structure.
Two variations of this fundamental device, however, are
found frequently applied in biomechanical measurements.
These are the force-sensitive resistor and the compliant dielectric-capacitance sensor.
Force-Sensitive Resistor. One of the simplest and therefore,
least expensive force sensors consists of a carbon-loaded elastomer and metallic contact structure as shown in Fig. 3. The
carbon-filled elastomer is electrically conductive and has a
textured surface that contacts the metallic conductor. This
has been exaggerated in Fig. 3 to illustrate the operating
principle. When small normal forces are applied, the metallic
conductor contacts only the tips of the loaded elastomer layer,
but, as the force increases, the elastomer is compressed and
more of the textured surface makes contact with the metallic
electrode. This causes the electrical resistance between the
electrode and the metallic contact at the base of the conductive elastomer to decrease.
Sensors of this type are frequently used for measuring
forces between body surfaces and the external world. For example, with this type of device, it is possible to measure
grasping, sitting, and standing forces and their distributions.
By patterning the conducting contact, it is possible to have a
force sensor array to measure the distribution of forces over
an area. This is especially useful in studying seating pressures and the reduction of decubitous ulcers. The advantage
of this type of sensor is that it is very thin and relatively
BIOMEDICAL SENSORS
Upper substrate
429
Large applied force
Small applied force

Metal
film
Lower substrate
Lower
substrate
Conductive compliant
elastomer
inexpensive. Its limitation lies in the fact that it is not highly

reproducible, and, because of the use of the carbon-filled elastomer, it has high hysteresis.
Compliant Dielectric Force Sensors. One can also measure
biomedical forces with a structure similar to that of the forcesensitive resistor but in this case it is a capacitor. A compliant
dielectric material is placed between the parallel plates of a
capacitor as shown in Fig. 4, and, as a force is applied normal
to the plane of this structure, the plates move closer together.
Because the capacitance is given by
C=
A
d
(8)
where C is the capacitance, is the dielectric constant, A is

the area of the capacitor plates, and d is the separation between these plates. As the force increases, the plate separation decreases as
d =
F
AE
(9)
where d is the change in separation with an applied force F

and E is the elastic constant (Youngs modulus) of the dielectric material. Although the capacitance of such a sensor varies hyperbolically with applied force, the characteristics of the
sensor are linearized by measuring the capacitance from depositing a known charge on the structure and measuring the
voltage across the plates. This yields the relationship
A2 E
A
=
d0 d
AEd0 F

d0
1
Q
=Q
F
V =
C
A
A2 E
C=
(10)
(11)
where d0 is the initial thickness of the compliant dielectric

before the force F is applied. As with the force-sensitive resis-
Upper substrate
Figure 3. Cross-section of a force-sensitive resistor with small applied force (left)

and a relatively large applied force (right).
tor, it is possible to make capacitive sensors in arrays. These

devices are applied similarly to the force-sensitive resistors,
and in many cases their characteristics are more precise and
are linear. An example of an application is a shoe pad containing several capacitive force sensors to measure forces at
various points on the foot of a standing subject (7).
Pressure Sensors
The general design of a pressure sensor is illustrated in Fig.
5(a). It consists of a chamber coupled to the pressure being
measured and a diaphragm as part of the chamber wall. The
outside wall of the diaphragm is usually in contact with air
at atmospheric pressure, but in some cases, such as miniature
implantable sensors, a vacuum is on the other side of the diaphragm. Increasing the pressure within the chamber deflects
the diaphragm, and the extent of this deflection is proportional to the pressure. Thus, one can produce a two-step pressure sensor by adding a displacement sensor to the diaphragm. In most cases a strain gage is used as the
displacement sensor. Reusable strain gage pressure sensors
coupled to the biological fluid by a fluid-filled catheter have
been the mainstay of physiology laboratories for many years
(8). These devices, however, are seldom used clinically anymore because it is necessary to sterilize them between applications, and they are not entirely robust. They have been replaced by disposable pressure sensor cartridges [Fig. 5(b)]
with a semiconductor strain gage pressure sensor. In this case
the strain gages that detect diaphragm deflection are integrated into a silicon diaphragm within the sensor cartridge.
These diaphragms are manufactured by microfabrication
technology, and they and the completed sensor are much
smaller than the sensors previously described (9,10). These
devices are produced inexpensively, so they are used on a single patient and then discarded rather than being cleaned and
sterilized before use on the next patient.
Miniature semiconductor pressure sensors are also fabricated to allow introducing them into the vasculature or body
Large applied force
Small applied force

Metal
film
Lower substrate
Dielectric elastomer
Lower
substrate
Figure 4. Cross-sectional view of a compliant dielectric force sensor with low (left)
and high (right) applied force.
430
BIOMEDICAL SENSORS
Port
Fluid-filled
Chamber
Dome
;
;
;
;
;
;
;
;;;;;;;;
;
;
;
;;;;;;;
Diaphragm
Plastic housing
Lead wires
Port
Displacement
sensor
Vent
Vent
Silicon diaphragm
with integrated
strain gauges
Housing
(a)
(b)
Figure 5. Fundamental pressure sensor structure (a) and disposable pressure sensor (b).
cavities. These so-called catheter-tip sensors are generally

considerably more expensive than the external devices and
must therefore, be reusable. There are methods to liquid or
gas sterilize these devices to avoid cross-contamination from
one patient to the next.
One of the advantages of the miniature, semiconductorbased pressure sensors is their small size. This means that
the diaphragm and displacement sensors undergo considerably smaller displacement for a given pressure and have a
much higher resonant frequency. Thus, these devices have
much better frequency response characteristics than the older
separate strain gage pressure sensors. The catheter coupling
the pressure from the body to an external sensor also distorts
dynamic pressure waveforms caused by the fluid characteristics of the cathetersensor system. This results in further reduced high-frequency response, ringing, or motion artifact
due to movement of the catheter. Catheter tip sensors avoid
these problems.
Flow Measurement
A fundamental physiological mechanism is that of fluid transport through various vessels, ducts, and other anatomic structures to carry nutrients and waste, exchange materials with
cells or tissue, and provide a conduit for transport of chemical
signals. It is, therefore, understandable that one needs to
measure fluid flow to describe and understand physiological
mechanisms. Although there are many different types of sensors used to measure flow in pipes, only a few of these are
appropriate for application to physiological systems. Some of
these are described in the following paragraphs.
Pneumotachograph. One of the fundamental principles of
fluid mechanics is Poiseuilles law which states that the pres-
sure drop along the length of a fluid flowing in a tube is proportional to the volume flow through that tube. Thus, if one
measures the pressure difference along a known resistance,
such as a rigid tube, this pressure drop is proportional to the
flow. Although it is not practical to make such a measurement
in a blood vessel whose geometry changes according to physiological and fluid dynamic conditions, this principle is used for
measuring gas flow.
The pneumotachograph is used for measuring the flow of
air into and out of the airway, and hence, the lungs. By placing a known resistance, such as a metal screen or a corrugated foil in a tube through which the breathing air flows and
measuring the differential pressure across this resistance, it
is possible to obtain a signal proportional to the flow of gas
through this system. This signal can then be electronically
integrated to determine volume.
Pneumotachographs directly measure air flow into the respiratory tract because the actual gas entering the body must
pass through the sensing system. They, therefore, are used
only when there is a direct connection to the airway, such as
when a patient is intubated or a nasaloral face mask is used.
Thus, this device is primarily used for diagnostic studies,
such as in a pulmonary function laboratory.
Electromagnetic Flow Meter. It is known from electromagnetic field theory that charged particles moving in a plane
transverse to a magnetic field experience a force mutually
perpendicular to the direction of their velocity and that of the
magnetic field. If blood or some other fluid containing positively and negatively charged ions flows with a velocity u in
a direction perpendicular to a magnetic field B positive ions
are deflected transverse to the field and the direction of flow,
and negative ions are deflected in the opposite direction. This
BIOMEDICAL SENSORS
results in establishing an electrical potential e across the

flowing fluid that is given by Faradays law of induction.
Thus, it is possible to measure a voltage across the fluid column that is proportional to the magnetic field strength and
the velocity of the flowing blood:
e=
u B dL
(12)
where L is the separation of the electrodes. For the ideal case

of a uniform velocity profile u and a uniform magnetic field B
Eq. (12) reduces to
e = BLu
(13)
This principle is the basis of the electromagnetic flow meter

illustrated in Fig. 6. A magnetic field is set up transverse to
the axis of a blood vessel by a permanent magnet or an electromagnet. Electrodes at right angles to the magnetic field
direction through the blood vessel make contact with its outer
surface and detect the voltage resulting from the flowing
blood. For a fixed magnetic field, this voltage is a steadystate, dc potential when the flow is constant. This voltage,
however, can be in error due to offset potentials on the electrodes which induce errors in the flow measurement. This
problem is overcome by using an electromagnet driven by an
ac current to generate a time-varying magnetic field. The resulting voltage seen at the electrodes due to the flow is ac,
whereas the offset potentials continue to be dc. Thus the two
signals can be separated. Although one problem is overcome
by using ac magnetic excitation, another is introduced. There
is an induced voltage, the transformer effect, if time-varying
flux lines cross the area bounded by the electrodes, the vessel
diameter between them, and the lead wires. Many practical
approaches to minimize the transformer effect have been described (11,12).
Electromagnetic flow meters available commercially in
various sizes and shapes are used for blood flow and cardiac
output studies in animal models. Experimental versions of
electromagnetic flow meters that can be introduced into the
lumen of a blood vessel are reported (13), and these devices
are introduced through a peripheral vein or artery without a
major surgical procedure. It is important to note that electro-
Vessel
Electrode
B
e
Electrode
Magnet
u
Figure 6. Schematic view of an electromagnetic flowmeter.
431
magnetic flow meters measure flow velocity rather than volumetric flow. It is possible, however, to obtain volume flow information from them because placing the flow probe around a
blood vessel requires a snug fit so that the electrodes make
good contact with the vessel, and this fixes the inner diameter
of the vessel where the flow measurement is made. The inner
diameter is used to determine the cross-sectional area of the
blood vessel and multiplied by the flow velocity, gives the volumetric flow.
Ultrasonic Flow Measurement. The Doppler effect states
that the frequency of a sound or ultrasound signal from a
moving reflector is shifted according to the velocity of the reflector and the angle between the direction of the incident and
reflected sound and that of the reflector:
f =
2 f0 u
cos
c
(14)
where f is the reflected sound frequency, f 0 is the incident

sound frequency, u is the velocity of the reflector, c is the velocity of sound and is the angle between the direction of the
sound and that of the reflector velocity. This principle is applied to the direct and the indirect measurement of blood flow.
In the former case, miniature transducers made of piezoelectric materials that produce and detect ultrasound are placed
near or directly on a blood vessel so that they direct a beam
of ultrasound into the vessel at an oblique angle. One transducer is used to generate the ultrasonic signal, whereas the
other detects the diffusely scattered ultrasound waves from
the moving blood cells. The reflected signal is shifted in frequency according to Eq. (14) and is used to determine the flow
velocity of the blood. As with the electromagnetic flow meter,
this method cannot directly determine the volume flow, but
by constraining the diameter of the blood vessel it is possible
to get a signal proportional to volumetric flow.
It is also possible to measure flow velocity of blood in vessels noninvasively using an ultrasonic imaging system. In
this case a B-scan image of the vessel and its surrounding
tissue is obtained, and small pixels of this image within the
vessel are analyzed for frequency shifts in the reflected ultrasonic signal. In some ultrasonic scanners, the color of the image from within a vessel is varied according to this frequency
shift and hence the flow velocity. Such images help clinicians
to understand conditions of abnormal flow in major vessels
caused by plaque formation or other anatomical anomalies. In
the case of this noninvasive measurement of flow by imaging,
it is not possible to determine volumetric flow because one
does not know the vessel cross-sectional area or the angle between the incident ultrasound and the flow direction.
This latter problem is overcome with a new technique
known as ultrasonic speckle tracking (14). In this technique,
one considers the ultrasound image of the vessel and its contents as a motion picture taken one frame at a time. The vessel should lie within the image plane, and ideally the axis of
the vessel should lie in the image. The pattern within the
vessel lumen has a texture, the speckle, in one frame, and, if
one looks at the next frame of the image taken shortly after
the initial frame, the same speckle pattern appears as though
it was shifted along the vessel. By recognizing this shift and
knowing the time elapsed between the two frames, it is possi-
432
BIOMEDICAL SENSORS
ble to calculate the flow velocity regardless of the angle between the ultrasound beam and the flow direction.
Rs
Measurement of Temperature
Sensors for measuring biomedical temperature are the same
as those applied in other fields. Those most frequently used
include the thermistor, thermocouple, and temperature-sensitive metallic wire or film resistors. The thermistor is by far
the most common because of its relatively high sensitivity and
small size. The latter is important in many biomedical measurements so that the instrument has rapid response time.
Another area of temperature measurement becoming important for clinical and home applications is radiation measurement of temperature. Inexpensive devices that measure
the infrared radiation from the auditory canal are commercially available, and these respond almost instantaneously.
Tympanic membrane (ear drum) direct temperature measurement using a miniature thermistor or thermocouple has been
recognized as a minimally invasive method of determining
core temperature (15), and the infrared radiation devices take
advantage of this and the rapid response time of infrared detectors for making the measurement (16). Skin temperature
over a portion of the body, such as the breast or abdomen is
also measured by infrared radiation. The technique of thermography is useful in locating subcutaneous or deeper areas
of inflammation such as in the case of some tumors or localized infection.
In addition to the applications previously mentioned, the
most common medical application of temperature sensors is
determining body temperature. This, along with blood pressure is one of the fundamental vital signs used to evaluate
patients, and rapidly responding minimally or noninvasive
sensors are desirable. The most common approach to this
measurement is an electronic thermometer utilizing a lowmass thermistor placed orally. Because a nurse carries this
device on patient rounds, an important aspect of its design is
a protective disposable sheath that is placed over the temperature sensor before it is placed in a patients mouth or elsewhere. This minimizes cross-contamination from one patient
to the next, but it also increases the response time of the sensor because of the series thermal resistance and increased
mass. Thus, an important aspect of this design is to minimize
response time so that temperature is rapidly obtained and
documented, thereby allowing the nurse more time for other
patient interactions.
BIOPOTENTIAL ELECTRODES
The body produces many electrical signals that are useful in
diagnosing and monitoring normal function and disease. The
most frequently measured of these is the electrocardiogram
(ECG) from the heart, the electroencephalogram (EEG) from
the brain and the electromyogram (EMG) from muscle. Biopotential electrodes are sensors placed on or within the body to
pick up these signals for processing and display by an instrumentation system (17). Thus, electrodes serve as the sensor
for these instruments.
The basic operating principle of biopotential electrodes is
converting an ionic current within the body to an electronic
current in the electrode material and associated electrical cir-
Rp
Cp
Ehc
Figure 7. Equivalent electrical circuit for a biopotential electrode.
cuits. For a charge to move from ions to electrons, chemical

reactions known as redox reactions must occur.
The following is a general form of these reactions:
M
M
n+
A + pe
+ ne
A
(15)
where M is a metal atom, Mn is its cation, A represents a

neutral atom of a material, and Ap is an anion of that material. The numbers n and p are the valences of the atoms M
and A, respectively, and are integers representing the number
of excess positive or negative charges on the ions. Because
these reactions occur on a metal surface in contact with an
electrolytic solution, an excess positive or negative charge
builds up at the interface between the metallic conductor of
electrons and the solution of ions. This buildup of charge,
known as polarization, results in electrical potential changes
that are much larger when electrodes are moved than bioelectrical signals being measured. Thus, in many applications it
is desirable to have electrodes that do not show or at least
minimize this polarization. Although it is not possible to
achieve an ideal nonpolarizable electrode in practice, silver/
silver chloride (Ag/AgCl) electrodes come close (18).
Figure 7 shows the equivalent electrical circuit for a biopotential electrode. Capacitor Cp and resistor Rp result from the
polarization at the electrode/electrolytic solution interface.
The more polarizable the electrode is, the higher the values
of these components become. Series resistance Rs represents
the resistive component of the interface not associated with
polarization, and the half cell potential Ehc arises from the
redox reactions occurring at the electrode/solution interface.
This potential is different for different materials entering into
the redox reactions and hence is a function of the electrode
material, the ions in solution, and the condition of the electrode surface. Typical values for these components for a disposable adult skin electrode are: Rs 200 , Rp 200 k,
Cp 16 nF, and Ehi 220 mV.
It is possible to determine the source impedance of an electrode from the equivalent circuit, and this impedance, in general, is nonlinear. Individual component values are determined by electrode materials, surface area, and frequency of
the signal measured. Very small electrodes, such as microelectrodes used for intracellular measurements, have very
high source impedances because of their small effective surface area. Electrode source impedance is also affected by electrical current crossing the electrode/solution interface. Such
current drives the reactions of Eq. (15) resulting in increased
BIOMEDICAL SENSORS
polarization due to the current. Thus it is important that electrode current is as small as possible. Ideally, it should be zero.
One way to minimize electrode current is to have amplifiers
with very high input impedance and low bias current connected to the electrodes.
Silver/Silver Chloride Electrode
Although electrochemists know of several different electrode
systems that approach the behavior of a nonpolarizable electrode, only the Ag/AgCl electrode is used as a biomedical sensor. This use is generally limited to applications on the skin
surface because the silver ion is toxic in the body. The Ag/
AgCl electrode minimizes polarization because of the low solubility of silver chloride, resulting from oxidation of silver
atoms on the electrode surface in the presence of chloride, the
principal anion of the body (17). There are many ways to realize Ag/AgCl electrodes in practice (19). One of the most robust
forms is a sintered electrode with a silver wire placed along
the axis of a cylindrical mixture of finely powdered silver and
silver chloride compressed to form a pellet. A layer of silver
chloride is formed on a silver electrode surface by electrochemical oxidation in a chloride-containing solution. Exposing
the silver metal surface to chlorine gas, such as in sodium
hypochlorite, ordinary household bleach, also produces a thin
layer of silver chloride. With the silver chloride surface on the
electrode, electrical motion artifact and noise are of much
lower amplitude than with unchloridized electrodes (20).
Examples of Electrodes and Applications
Figure 8 shows some of the common forms of biopotential
electrodes. Skin electrodes are made from Ag/AgCl disks
formed by any of the methods described in the previous section [Fig. 8(a)]. Often a silver foil or silver plated surface is
used as the basis of these electrodes. It is possible to make
electrodes in the form of a needle, as shown in Fig. 8(b), that
is injected into a muscle to pick up EMG signals. Single or
multipolar coaxial electrodes are formed by running an insu-
;
;
;
;
; ;;
Lead wire
AgCl
lated wire through the lumen of a hypodermic needle and

grinding it off at the needles bevel.
The microelectrode is a very small miniature injectable
electrode consisting of a glass micropipette drawn to a very
fine point as small as 1 m in diameter [Fig. 8(c)]. This point
impales the body of a single cell, such as a nerve cell, to measure the electrical potential within the cell with respect to an
extracellular electrode, thereby measuring the voltage across
the cell membrane. An electrolytic solution within the micropipette couples the intracellular cytoplasm to an Ag/AgCl
electrode in the shank of the micropipette.
Microelectronic technology has been used to fabricate miniature electrodes for various biomedical applications. Although these structures are not as small as glass or metal
microelectrodes, they do allow fabrication in one- and twodimensional arrays. The use of microelectronic technology
makes it possible to batch fabricate the electrodes in highly
reproducible forms and to distribute the production costs over
a large batch of devices, thereby making individual devices
relatively inexpensive. Sensors consisting of silicon probes
with electrode arrays (21,22), miniature chambers that can
contain the electrode chemistry (23), sieves with electrodes
through which structures such as nerves can grow (24,25),
and plaques of two-dimensional arrays have been fabricated
and used for cardiac and neural recordings.
Intracardiac electrodes are used with pacemakers to pick
up cardiac electrograms to control whether or not the pacemaker generates a stimulus. Such electrodes are placed on a
flexible probe, such as shown in Fig. 8(d) and introduced into
the heart through a vessel. Often these electrodes are incorporated into the probe containing the stimulating electrodes for
the pacemaker. There are many other examples of biopotential electrodes described in several general references
(17,19,20,26,27).
CHEMICAL SENSORS
Biological organisms involve many complex chemical reactions, and so it is important to measure concentrations and
Exposed wire
at tip
Lead wire
Wire
Ag
(a)
Lead wire
Micropipette tip
Internal electrolytic Ag/AgCl electrode

solution
(c)
433
Insulation
Needle shank
(b)
Flexible catheter
Metal electrodes
(d)
Figure 8. Common forms of biopotential electrodes: (a) Ag/AgCl electrode; (b) coaxial needle
electrode; (c) microelectrode for intracellular measurement; and (d) intracardiac electrode for
sensing and pacing.
434
BIOMEDICAL SENSORS
ions of the analyte pass through the membrane. When this

membrane is in contact with an unknown solution of the analyte ions, an electrical potential difference is generated across
the membrane that is related to the log of the concentration
difference across the membrane according to the Nernst equation.
Table 3. Classification of Chemical Sensors

Electrochemical
Amperometric
Potentiometric
Coulometric
Optical
Colorimetric
Fluorescent
Scattered light
Emission and absorption spectroscopy
Thermal
Calorimetry
Thermal conductivity
E = E0
activities of various substances to understand biological function. Chemical sensors are devices that convert concentrations or activities of chemicals into electrical or optical signals
related to these quantities. The major classes of chemical sensors are listed in Table 3. Electrochemical sensors convert the
chemical substance being measured into an electrical quantity, such as voltage, current, or charge. Optical sensors have
their optical properties changed by the chemical being measured or by light of a specific wavelength produced by the
chemical. There are also thermal methods for detecting concentrations of substances and major analytical techniques,
such as spectroscopy and nuclear magnetic resonance, that
involve complete instrumentation systems and are beyond the
scope of this article.
Electrochemical Sensors
(16)
where E is the potential measured, E0 is a constant potential

associated with the membrane, R is the universal gas constant, T is the absolute temperature, n is the valence of the
analyte ion, and a1 and a2 are the activity of the analyte ions
on each side of the membrane. By placing a reference electrode (such as a Ag/AgCl electrode) on either side of the membrane and connecting these electrodes to a very high input
impedance voltmeter, the potential across the membrane is
determined. Because the activity of the analyte ion in the solution within the chamber is known, it is possible to determine the activity of the analyte ion in the external solution
by measuring this potential.
Ion-selective electrodes have been developed for several inorganic ions of interest in physiology and medicine. These include the common pH glass electrode and sensors for other
ions, such as potassium, sodium, chloride, calcium, lithium,
and ammonium. Some of the ion-selective membranes are
based on special glass formulations whereas others have a
polymeric matrix, such as poly(vinylchloride), that is highly
plasticized and contains ion-carrier molecules. In all cases the
electrical source impedance of such sensors is very high, and
measurements must be taken with a high input impedance
voltmeter, such as an electrometer.
Amperometric sensors measure electron currents associated with redox reactions that involve the analyte being measured. The most common amperometric sensor used in biomedical applications measures the partial pressure of oxygen
in solution. The redox reaction for oxygen at the cathode of
;
;
;
;
;
;
;
;
;
;;;;;;;;;;
;
;
;
;;;;;
;
;
;; ;;;;;; ;
;
;
;
;;;;
Potentiometric sensors produce a voltage proportional to the

activity of the chemical being sensed. Ion-selective electrodes
are a common form of potentiometric chemical sensors. Their
structure, shown in Fig. 9(a), consists of a chamber containing
a solution of the analyte at a known activity. A portion of the
chamber wall is a membrane specially formulated so that only
a
RT
ln 1
nF
a2
Lead wires
Insulator
Ag/AgCl
electrode
Internal
electrolytic
solution
Figure 9. Examples of chemical sensors

used in biomedical instrumentation: (a) an
ion-selective electrode; (b) a Clark oxygen
sensor (note: the electrolytic solution layer in
this illustration is thicker than in practice).
Platinum cathode
Oxygen-permeable
membrane
Ion-selective membrane
(a)
(b)
BIOMEDICAL SENSORS
an amperometric cell is as follows:

O2 + H2 O + 4e
4OH
(17)
A 600 mV bias is required for this reaction. An example of a

practical realization of an amperometric oxygen sensor is the
Clark electrode shown in Fig. 9(b). An oxygen-permeable
membrane separates the measurement chamber from the external environment. Oxygen diffusing through this membrane
into the inner electrolytic solution eventually makes its way
to the cathode where it is reduced according to Eq. (17) to
form hydroxyl ions and requiring electrons from the cathode
to complete the process. As the partial pressure of oxygen increases outside of the membrane, more oxygen diffuses
through the membrane into the inner solution and to the
cathode where it is reduced. Thus, as the oxygen level increases in the environment of the sensor, the cathode current
increases because of the greater availability of oxygen at its
surface.
Coulometric sensors measure the actual amount of charge
transferred in a redox reaction to determine the amount of a
specific reactant in a solution. Although this is an important
technique in the analytical laboratory, it is not used very frequently for biomedical measurements, and it is difficult to
make this kind of measurement in vivo.
Optical Sensors
Optical transduction techniques are used for physical and
chemical sensors. Most chemical sensor applications of optical
devices are similar to applications involving colorimetric or
fluorometric measurements with analytical laboratory instruments. These measurements involve looking at the color
change or optical absorption (or reflection) of a chemical dye
whose properties change as a function of the concentration or
activity of a particular analyte that reacts with the dye. A
simple example of this is litmus paper that changes its color
in contact with a solution as the pH of the solution changes.
Laboratory instruments use chemically sensitive dyes and
measure their color changes photometrically when they are
added to the solution being measured. Optical sensors carry
out a similar function by immobilizing the dye at the tip of a
probe in contact with the solution being measured. The probe
illustrated in Fig. 10 consists of a fiber optic bundle with some
of the fibers transmitting light from an external source to the
dye and the remaining fibers returning the light transmitted
through or reflected from the dye. This signal is processed at
the proximal end of the probe photometrically and electronically to determine differences in the returned light compared
with the transmitted light, which are related to the chemical
substance being measured. Such devices are described for intravascular pH measurements where the probe is a part of a
catheter introduced into a peripheral artery or vein (28).
The second general form of optical sensor based on analytical laboratory instrumentation consists of a similar fiber optic
probe with a cuvette containing a different type of dye at its
tip. The fluorescent properties of this dye are determined by
the concentration of an analyte in a solution in contact with
that dye. For example, it is known that oxygen quenches the
fluorescence of certain dyes at the tip of a probe in contact
with an oxygen-containing solution, such as blood (29). This
principle has been used to build an optical oxygen sensor.
435
There is another way that oxygen in blood is measured

optically. The red blood cells contain hemoglobin that is an
oxygen carrier. The optical spectrum of oxidized hemoglobin
is different from that of the hemoglobin molecule without oxygen. This is why that well-oxygenated blood is bright red, and
blood with low oxygen content is a deep maroon color. By
measuring the amount of light reflected from blood at various
wavelengths, the fraction of oxygenated hemoglobin in the
blood is determined. This is known as the percent hemoglobin
oxygen saturation. By constructing an intravascular fiber optic probe that transmits light at two different wavelengths
and returns the reflected light so that its intensity can be
measured, the oxygen saturation of blood is determined. This
process, known as oximetry, looks at the ratio of the intensity
of light at two different wavelengths reflected through or from
a blood specimen. One of these wavelengths should be an
isosbestic point, such as 805 nm, a wavelength where the
spectra of oxygenated and reduced hemoglobin are the same,
and the second wavelength should be in the visible red portion of the spectrum, such as 660 nm, where there are major
differences between the optical properties of oxygenated and
reduced hemoglobin. The hemoglobin oxygen saturation is determined from the ratio by the equation
oxygen saturation = a b(IR/R)
(18)
where a and b are constants based on the measurement conditions. It is important to note that oximetry gives the hemoglobin oxygen saturation, but it does not give the total oxygen
concentration in the blood, because the hemoglobin content is
unknown. If the hemoglobin is independently measured, however, it is possible to determine the total oxygen concentration. This is different from the oxygen tension (partial pressure of oxygen) in the blood. The partial pressure of oxygen
in well-saturated hemoglobin varies over a wide range of values even though the saturation is close to 100%. Oxygen tension is determined only by an electrochemical sensor, such as
the amperometric oxygen sensor, or by analytical laboratory
methods, such as Van Slyke analysis (30).
Although optical oximetry has been a technique for blood
analysis for over 60 years, only in recent years has it become
a major measurement for critical care medicine because of the
development of the noninvasive pulse oximeter (31,32). This
optical method is based on the transillumination of tissue at
the two wavelengths previously described. This is done by
placing light emitting diodes (LED) of the desired wavelengths on one side of a finger, toe, or earlobe and using a
light detector, such as a photodiode or phototransistor, on the
other side opposite the emitters. Now the tissue between the
light sources and detector is the cuvette that holds the blood,
but it differs from the laboratory instrument or invasive oximeter case in that the blood volume being measured is variable because the tissue is not made up entirely of blood. As a
matter of fact, the blood volume varies with time over the
cardiac cycle because of the compliance of the vasculature. At
systole a fresh bolus of arterial blood enters and distends the
vasculature of the tissue thereby increasing the percentage of
blood in that tissue with arterial blood. At diastole the pressure is lower, and so new blood does not enter the tissue.
Blood continues to exit the tissue through the venules and
veins, so that the total blood volume in the tissue decreases
during diastole. These changes in blood volume result in simi-
436
BIOMEDICAL SENSORS
Sample
Transmission
cuvette
Light
source
Optical fibers
Cuvette
Light
detector
Sample
Figure 10. Fiber optic chemical sensor
probe with transmission or reflectance
sample cuvette that contains the sample
itself or a dye in contact with the sample.
Reflectance
cuvette
lar but inverted changes in the transmitted light through the

tissue. The pulse oximeter measures these changes in transmitted light intensity and looks at the pulsatile component of
this transmitted light. Because this pulsatile component results from the entry of arterial blood into the tissue, looking
at the ratio of pulse amplitudes at the two different wavelengths using Eq. (18) gives a result related to the hemoglobin
oxygen saturation of the arterial blood. Although a theoretical
relationship can be calculated by light scattering theory, practical pulse oximeters use an experimentally derived relationship between the ratio of pulse amplitudes and hemoglobin
oxygen saturation. In addition most pulse oximeters normalize the measured pulse amplitude with respect to the steadystate light transmission and average their results over several cardiac cycles to determine the oxygen saturation. This
helps to minimize noise but also reduces effective response
time.
The device is very easy to use clinically. All one needs to
do is to tape or clamp the light emitting diodes and sensors
to the tissue being measured. The instrument takes care of
setup and calibration. Once the sensors are in place, information can be obtained. It is important to note that transmitted
light is affected by the arterial pulse and also changes in venous blood volume and movement of the patient produce variations in transmitted light intensity. This results in errors in
determining hemoglobin oxygen saturation, and these errors
frequently are due to motion artifact. Because of this artifact,
pulse oximeters are generally not useful for measuring actively moving subjects during exercise testing. Even with this
limitation, pulse oximeters are widely used in clinical care
because of their operating simplicity and ability to provide
critical information.
Bioanalytical Sensors
Analytes of biomedical interest are often biochemicals with
rather complex structures. Conventional electrochemical sensors often lack specificity in detecting these substances, and
so other approaches are needed. One approach is to utilize
biological recognition processes as part of the sensing mecha-
nism. These sensors, known as bioanalytical sensors, take advantage of one of the following general types of biochemical
reactions: (1) enzymesubstrate, (2) antigenantibody, or (3)
ligandreceptor. When these reactions are used, a sensor
highly specific for a particular biological molecule can be developed. This sensor is usually has two or more stages. The
first stage involves the biological sensing reaction, and this
part of the sensor contains one of the components of the reaction, such as an enzyme or an antibody. The second stage of
the sensor determines if, and to what extent, the biological
reaction has taken place. This portion of the sensor consists
of a physical or chemical sensor that senses the biological reaction based on changes in mass, electrical capacitance, electrical charge transfer, temperature, or optical properties. This
section of the sensor may also consist of a chemical sensor
that detects the product of a reaction or the depletion of one
of the reactants. Bioanalytical sensors are described for many
biological analytes. These sensors are often specific for a particular application (3336). The most common example of a
bioanalytical sensor senses glucose by using the enzyme glucose oxidase. The fundamental reaction involved is
Glucose + O2
glucose oxidase
Glucuronic acid + H2 O2
(19)
This reaction occurs at the first stage of the sensor and is

sensed in several ways: (1) by measuring the oxygen consumed, (2) by measuring the hydrogen peroxide produced,
and (3) by measuring the heat produced by the reaction. Each
technique has its advantages and limitations. For example,
the oxygen consumed can be determined by measuring the
partial pressure of oxygen in the vicinity of the glucose oxidase with an amperometric oxygen sensor. The problem with
such a technique lies in the fact that it depends on the partial
pressure of oxygen in the environment, and it is the change
in oxygen tension that is really necessary to determine. This
can be done with a differential sensor made up of two physically identical glucose sensor structures located adjacent to
one another. The only difference between the two is that one
contains the enzyme whereas the other does not. Thus, by
BIOMEDICAL SENSORS
measuring the difference in oxygen tension seen by these two

sensors, one determines the amount of glucose present (37).
Bioanalytical sensors present special problems because
they utilize biological substances. Often the stability of these
chemicals depends on environmental conditions. Exposure to
extremes of temperature or hydration lead to conformational
changes of the sensing molecules that change their biological
activity and hence the sensitivity of the sensor. Often bioanalytical sensors have limited lifetimes and must be stored using special conditions, such as in a dark, cool, humid environment, to remain functional. It is also important to note that
not all biochemical reactions are entirely reversible, and so
the bioanalytical sensors based on these reactions are also not
reversible. Thus, the sensor can be used only for a single measurement.
Microelectronic technology has been applied to chemical
and bioanalytical sensors as it has to physical sensors
(23,36,37). This technology makes small, reproducible structures possible, which gives the sensors characteristics that
are more similar than those from microsensors. Often the
need for repeated calibrations can be reduced using this technology.
APPLICATIONS OF BIOMEDICAL SENSORS
The many ways in which bioanalytical sensors are used continue to grow as biomedical instrumentation becomes increasingly important as an adjunct to diagnosis and care. Sensors
are finding their way into routine clinical care, the physicians
office and clinic, nonhospital care facilities such as nursing
homes, the patients home itself and intensive care units
where they are routinely used. The measurement is often carried out with noninvasive or minimally invasive sensors, but
it is important that these are applied properly for greatest
efficacy. Invasive sensors reliable and suitable for chronic applications for the most part still need to be developed to overcome major problems. These devices are desired for chronic
applications, such as feedback sensors for rate responsive
pacemakers, and they are expected to function for the remainder of the patients life with little or no external intervention.
Wireless telemetry devices are sometimes used with these
sensors to communicate with instrumentation outside of the
body. This avoids the necessity of having wires pass through
the skin, a source of infection in chronic applications. Often it
is not possible to calibrate implanted sensors once they have
been placed in vivo, and they must have stable characteristics
not only under in vitro laboratory testing but also in vivo.
Issues related to the sensor-biological system interface still
need to be understood and controlled. All foreign bodies implanted in tissue elicit a response from that tissue to reject or
at least wall off the invading substance. Implanted sensors
are usually surrounded by a fibrous capsule after 10 to 14
days implantation as a result of the organisms foreign body
response. This capsule is a barrier to transport of the analyte
to the sensor thereby changing the sensor characteristics. Understanding the process of forming this capsule and its transport properties is needed to characterize sensors in vivo.
Much work remains to be done in this area.
437
optical) instrument and the biological system being measured. The validity of the data provided by an instrumentation system is often linked to processes at this interface and
to the functionality of the sensor itself. Although electronic
signal processing compensates for some problems, in general
the quality of a measurement is determined by the quality
of the sensor making that measurement. Understanding the
physics, chemistry, engineering, biology, and applications of
sensors will lead to the development of better devices and
their meaningful application to biomedical problems.
BIBLIOGRAPHY
1. R. J. Whitney, The measurement of changes in human limb volume by means of a mercury-in-rubber strain gage. J. Physiol.,
109: 5, 1949.
2. R. S. Mendenhall and M. R. Neuman, Efficacy of five noninvasive
infant respiration sensors. Proc. IEEE Frontiers Eng. Med. Biol.,
New York: IEEE, 1983.
3. J. A. Adams et al., Measurement of breath amplitudes: comparison of three noninvasive respiratory monitors to integrated pnsumotachograph. Pediatr. Pulmonol., 16 (4): 254258, 1993.
4. L. J. Brooks, J. M. DiFiore, and R. J. Martin, and the CHIME
Study Group, Assessment of tidal volume in preterm infants using respiratory inductance plethysmography. Pediatr. Pulmonol.,
23: 429433, 1997.
5. H. F. Stegall et al., A portable, simple sonomicrometer. J. Appl.
Physiol., 23 (2): 289293, 1967.
6. I. Zador, M. R. Neuman, and R. N. Wolfson, Continuous monitoring of cervical dilatation during labor by ultrasound transit time
measurement, Med. Biol. Eng., 14: 299305, 1976.
7. S. Miyazaki and A. Ishida, Capacitive transducer for continuous
measurement of vertical foot force. Med. Biol. Eng. Comput., 22
(4): 309316, 1984.
8. E. H. Lambert and E. H. Wood, The use of a resistance wire
strain gage manometer to measure intraarterial pressure, Proc.
Soc. Exper. Biol. Med., 64: 186190, 1947.
9. M. Habibi et al., A surface micromachined capacitive absolute
pressure sensor array on a glass substrate, Sens. Actuators A:
Physical, 46: 125128, 1995.
10. J. Mandle, O. Lefort, and A. Migeon, A new micromachined silicon high-accuracy pressure sensor, Sens. Actuators A: Physical,
46: 129132, 1995.
11. J. A. Shercliff, The Theory of Electromagnetic Flow Measurement,
Cambridge: Cambridge University Press, 1962.
12. D. H. Bergel and U. Gessner, The electromagnetic flow meter, in
R. F. Rushmer (ed.), Methods in Medical Research, Chicago: Year
Book, 1966, Vol. XI.
13. C. J. Mills and J. P. Shillingford, A catheter tip electromagnetic
velocity probe and its evaluation, Cardiovasc. Res., 1: 263-273,
1967.
14. L. N. Bohs, B. H. Friemel, and G. E. Trahey, Experimental velocity profiles and volumetric flow via two-dimensional speckle
tracking, Ultrasound Med. Biol., 21 (7): 885898, 1995.
15. J. M. Dabbs, Jr. and M. R. Neuman, Telemetry of human cerebral
temperature, Psychobiology, 15 (6): 599603, 1978.
SUMMARY
16. J. Fraden, Noncontact temperature measurements in medicine,

in D. L. Wise (ed.), Bioinstrumentation and Biosensors, New York:
Dekker, 1991, pp. 511550.
Sensors serve an important function in biomedical instrumentation in that they are the interface between the electronic (or
17. M. R. Neuman, Biopotential electrodes, in J. G. Webster (ed.),

Medical Instrumentation: Application and Design, 3rd ed., New
York: Wiley, 1998, pp. 183232.
438
BIOMEDICAL TELEMETRY
18. J. G. Webster, Reducing motion artifacts and interference in biopotential recording, IEEE Trans. Biomed. Eng., 31: 823826,
1984.
19. G. J. Janz and D. J. G. Ives, Silver-silver chloride electrodes,
Ann. N.Y. Acad. Sci., 148: 210221, 1968.
20. L. A. Geddes, L. E. Baker, and A. G. Moore, Optimum electrolytic
chloriding of silver electrodes, Med. Biol. Eng., 7: 4956, 1969.
21. K. D. Wise, J. B. Angell, and A. Starr, An integrated circuit approach to extracellular microelectrodes, IEEE Trans. Biomed.
Eng. 17: 238246, 1970.
22. J. J. Mastrototaro et al., Rigid and flexible thin-film microelectrode arrays for transmural cardiac recordings, IEEE Trans. Biomed. Eng. 39: 271279, 1992.
23. O. J. Prohaska et al., Thin-film multiple electrode probes: Possibilities and limitations, IEEE Trans. Biomed. Eng., 33: 223
229, 1986.
24. G. T. A. Kovacs et al., Silicon-substrate microelectrode arrays for
parallel recording of neural activity in peripheral and cranial
nerves, IEEE Trans. Biomed. Eng., 41: 567577, 1994.
25. J. Pine et al., Silicon cultured-neuron prosthetic devices for in
vivo and in vitro studies, Sens. Actuators B: Chemical, 43: 105
109, 1997.
26. L. A. Geddes, Electrodes and the Measurement of Bioelectric
Events, New York: Wiley, 1972.
27. C. D. Ferris, Introduction to Bioelectrodes, New York: Plenum,
1974.
28. J. I. Peterson, S. R. Goldstein, and R. V. Fitzgerald, Fiber optic
pH probe for physiological use, Anal. Chem., 52: 864869, 1980.
29. J. I. Peterson, R. V. Fitzgerald, and D. K. Buckhold, Fiberoptic
probe for in vivo measurement of oxygen partial pressure, Anal.
Chem., 56 (1): 6267, 1984.
30. J. P. Peters and D. D. van Slyke, Quantitative Clinical Chemistry,
Baltimore: Williams & Wilkins, 193132.
31. T. Aoyagi, M. Kishi, and K. Yamaguchi, Improvement of the earpiece oximeter, Jpn. Soc. Med. Electron. Biomed. Eng., 9091,
1974.
32. I. Yoshiya, Y. Shimada, and K. Tanaka, Spectrophotometric monitoring of arterial oxygen saturation in the fingertip. Med. Biol.
Eng. Comput., 18: 2732, 1980.
33. J. S. Schultz, Biosensors, Sci. Am., 265 (2): 6469, 1991.
34. E. A. H. Hall, Biosensors, Englewood Cliffs, NJ: Prentice-Hall,
1991.
35. J. Janata, Principles of Chemical Sensors, New York: Plenum,
1989.
36. M. Lambrechts and W. Sansen, Biosensors: Microelectrochemical
devices, Bristol, UK: IOP Publishing, 1992.
37. Ph. Arquint et al., Integrated blood-gas sensor for pO2, pCO2 and
pH, Sens. Actuators B, B13: 340344, 1993.
38. W. Mokwa et al., Silicon thin film sensor for measurement of dissolved oxygen, Sens. Actuators B: Chemical, 43: 4044, 1997.
MICHAEL R. NEUMAN
Case Western Reserve University
114
DEFIBRILLATORS
SA node
LA
RA
AV node
RV
LV
HisPurkinji System
Figure 1. Diagram of the hearts electrical system. An electrical signal begins at the sino-atrial node (SA node) and travels to the left
atrium (LA) and right atrium (RA). The signal also travels to the
atrioventricular (AV) node. From the AV node, the electrical signal
travels through the HisPurkinje system to the right ventricle (RV)
and left ventricle (LV). The electrical signal causes the heart to contract in a coordinated fashion.
DEFIBRILLATORS
The function of the heart is to pump blood. The heart has two
sides, each consisting of an atrium and a ventricle. Deoxygenated blood is collected in the right atrium, passed to the
right ventricle, and pumped to the lungs where it is oxygenated. Blood is then collected in the left atrium, passed to the
left ventricle, and pumped to the rest of the body. Controlling
this pumping of blood is an electrical signal which passes
through the heart and triggers a coordinated mechanical contraction of the heart. During a normal or sinus beat, this
electrical activity starts in the sinus node, near the junction
of the superior vena cava and right atrium, and passes
through the right atrium to the atrioventricular node (Fig. 1).
From there, the electrical signal spreads down the His bundle
and throughout the left and right ventricles via a series of
specialized conducting cells called Purkinje fibers. Mechanical
action of the heart follows the same pattern with the atria
contracting first and then the ventricles contracting second.
The role of the Purkinje fibers is to rapidly spread the activation signal from the base to the apex of the heart so that
contraction can proceed from apex to base, pushing blood
out of the heart and into the aorta.
There are several disorders of this electrical system. Sometimes the heart beats too slowly, either because the sinus
node does not fire rapidly enough, or the signal is not able to
pass through the atrioventricular node to the ventricle. These
problems are best treated with an implanted pacemaker.
Sometimes, though, the heart beats too fast. A fast heart

rhythm of any sort is called a tachyarrhythmia. If the heart
is beating too fast but in an organized repeating fashion, the
rhythm is called tachycardia. If the heart is beating too fast
and in a disorganized fashion, the rhythm is called fibrillation. Either the atria or the ventricles can develop tachycardia or fibrillation. The danger of either tachycardia or fibrillation is that the heart is unable to pump enough blood to
support the body. Tachycardias, both atrial and ventricular,
as well as atrial fibrillation tend to be tolerated by patients
while ventricular fibrillation is fatal in 5 min to 20 min unless
corrected by defibrillation.
To date, the most effective treatment for atrial fibrillation
and the only effective treatment for ventricular fibrillation is
to apply a large electrical shock to the heart. Much research
has been done to understand how this large electrical shock
interacts with the heart to halt fibrillation. The first part of
this article will discuss practical aspects of defibrillators and
defibrillation: the most appropriate ways to measure defibrillation efficacy, the effect of electrode size and location, and
the effect of waveform shape. The second part will discuss
why an electrical shock is able to stop fibrillation and allow a
regular rhythm to resume.
TYPES OF DEFIBRILLATORS
There are two main types of defibrillators used today, the automatic internal defibrillator and the external defibrillator.
DEFIBRILLATORS
(a)
Figure 2. Diagram of an implantable cardiovertor defibrillator. The pulse generator is implanted in the pectoral
region. A transvenous catheter electrode is threaded from
the subclavian vein to the superior vena cava and into the
right ventricle of the heart. This catheter also contains a
pace/sense electrode on the tip. Implantation of this system only requires sedation of the patient and a local anesthetic.
(b)
The automatic internal cardiovertor-defibrillator (ICD) is a 40

mL to 100 mL box with electrodes attached to it that extend
either onto the epicardial (outside) surface of the heart or into
the chambers of the heart (Fig. 2). This device monitors the
cardiac rhythm and if ventricular fibrillation is detected delivers a strong electrical shock, usually 10 J to 30 J. Currently,
implantation of an ICD is the treatment of choice for patients
who have survived an initial episode of sudden cardiac death
and do not have a treatable cause for their arrhythmia. (1,2).
The external defibrillator is a device distributed throughout the prehospital and hospital setting (the paddles popularized in television hospital dramas). These devices deliver a
large electric shock, usually 100 J to 360 J, to the chest wall
of a patient via either hand-held paddle electrodes or selfadhesive patch electrodes (Fig. 3). The external defibrillator
is used to stop both ventricular and atrial tachyarrhythmias.
Traditionally, the external defibrillator operator has analyzed
the patients heart rhythm and decided whether or not to deliver the electrical shock. Newer devices, intended to be used
by minimally trained lay persons, such as police, firefighters,
flight attendants, or even passers-by, will analyze the pa-
tients heart rhythm and determine whether or not a shock

should be delivered without intervention from the operator.
Perhaps someday defibrillators will be as common as fire extinguishers. Quick action is vital to the survival of ventricular
fibrillation. The rate of survival following an episode of ventricular tachycardia/ventricular fibrillation is inversely related to the time that the patients heart has been in that
rhythm before a shock is delivered (Fig. 4). Therefore, it is
hypothesized that more patients would be saved if defibrillation occurred as soon as possible by individuals likely to be
near the person when his/her heart fibrillates (3).
MEASURING DEFIBRILLATION EFFICACY
Whenever a defibrillation shock is delivered to a patient, the
goal is to stop the tachyarrhythmia by stopping the heart
completely, thereby allowing the patients heart to resume
beating with a slow regular rhythm. Therefore, to compare
whether one defibrillator is better than another, it is necessary to be able to define the efficacy of a defibrillator. Although people describe defibrillation in terms of a threshold,
it is more appropriately described as a probability function.
Within a certain range, as the strength of the defibrillation
Probability of survival
.60
.40
5
.20
10
0
Figure 3. Diagram of electrode patch placement for external defibrillation. One electrode is placed over the right border of the sternum. The second electrode is placed on the left axillary line overlying
the apex of the heart.
115
15
5
10
15
Collapse to defibrillation interval
20
Figure 4. Probability of survival as a function of time in minutes

from collapse to the beginning of cardiopulmonary resuscitation and
time of defibrillation. Each contour represents a different time interval from collapse to the beginning of cardiopulmonary resuscitation.
Note that the probability of survival from cardiac arrest drops as the
time from arrest to the beginning of cardiopulmonary resuscitation
increases and as the time to defibrillation increases. Reproduced with
permission from the American Heart Association (98).
I0 = 50 A
80
60
= 30 ms 660 J
40
20
0
15
= 10 ms
660 J
30
45
= 20 ms 440 J
8000
1.0
6000
0.8
4000
0.6
0.4
2000
0.2
0
0.0
200 250 300 350 400 450
Defibrillation test shock leading edge voltage
Final current in amperes
(a)
Figure 5. Relationship between percent success of ventricular defibrillation and final current for exponential waveforms having an initial current (I0) of 50 A and time constants of decay (t) of 10, 20, and
30 ms. Energy is shown in joules for each time constant of decay (t).
Reproduced with permission from The Institute of Electrical and
Electronics Engineers (99).
1700
(V)
1360
680
340
160
320
480
(V)
640
800
(b)
90
Simulated rms error (V)
shock increases, the probability increases that the shock will

defibrillate the heart (Fig. 5) (4). This increase is plotted as a
probability of success curve. Defining a probability of success
curve requires a large number of shocks, 10 to 15, and so it is
impractical to always determine the entire curve for each patient. Therefore, several methods to estimate the 50% point
of the probability of success curve have been developed, including the step down (5), up/down (6), and the binary search
(7). All of these methods are described as measuring the defibrillation threshold in the literature. With the step down technique, the first shock delivered is of a strength that almost
always defibrillates. Then progressively smaller shocks are
delivered until a shock fails to defibrillate the heart. The up
down technique starts at a particular shock level, one thought
to be close to an estimated 50% success level, and shocks are
delivered at progressively higher or lower shock levels depending on the result of the previous trial until a reversal
from success to failure or failure to success occurs (8). The
binary search algorithm chooses a range in which the defibrillation threshold is thought to be, generally 0 J to 20 J (7).
The first shock is delivered in the center of the search space.
If the shock succeeds, then the next shock strength is the middle of the lower half of the initial search space. If the shock
fails, the next shock strength is the middle of the higher half
of the initial search space. The search space is progressively
bisected until the desired level of precision is reached.
These methods of estimating the 50% defibrillation success
level are sensitive to decisions made by the operator and underlying biases introduced by the algorithm itself. The binary
search algorithm requires that the operator estimate a range
that the defibrillation threshold is within. If the defibrillation
threshold is not within the range, then the algorithm will
never converge on an answer. The step-down method tends to
overestimate the 50% success point. A step size that is too
small for a step-down or updown method requires an excessive number of shocks to determine the 50% success point. If
the step size is too large, the precision of the estimate becomes very low. Empirically, people have used energy step
sizes of approximately 10% to 20%. McDaniel and Schuder
suggest that a log(energy step) equal to 0.05 is the best choice
(6). They base this on their laboratory data, which suggests
1020
80
70
60
50
40
21.8
Min SE without
retrospective
prior pdf
Min SE with
retrospective
prior pdf
19.4
16.9
14.5
Min SE
up-down
2.0
3.0
4.0
Number of measurements
12.1
9.69
Simulated rms error

(percent of simulated mean ED95)
Percent successful
100
Estimation of error (V2)
DEFIBRILLATORS
Probability of
defibrillation
(doseresponse curve)
116
(c)
Figure 6. A Bayesian approach to estimating the 95% probability of
successful defibrillation. (a) The method assumes a doseresponse
curve (open squares) which follows the logistic equation, although any
functional form could be used. Also shown is the cost function (closed
circles) that was chosen to give the lowest error. Cost functions that
minimize the absolute error or the patient risk are also possible. (b)
Contour plot of a prior probability density function (pdf) constructed
from a set of assumptions applicable to most implantable defibrillator
electrode configurations. and are variables that describe the logistic equation at the 95% probability point. is the subjects 95% probability point. One over is the slope of the logistic equation at the
95% probability point. For any animal, it is assumed that the 95%
probability point will be between 0 V and 800 V ( ) and that one over
the slope of the logistic equation () is between 0 V and 1700 V. (c)
The simulated performance of the minimum squared error (MinSE)
developed from (a) and (b).
DEFIBRILLATORS
that one standard deviation of the threshold is approximately

log(0.05) energy units.
A less widely used but promising method of estimation has
been developed by Malkin et al. (9). Bayesian estimation techniques are used to estimate both 50 and 95 points of the probability of success curve in a given patient (Fig. 6) (9,10). By
making several conservative assumptions about the probability search space and using a variable step sizes, Malkin et al.
showed that the 95% success level can be estimated with a
root mean square error of 15% (10). Although estimating the
50% success level is valuable in the research laboratory, estimating the 95% success level in patients would allow the physician implanting an ICD to set appropriate shock treatment
strengths with a great deal of confidence and so this methodology holds great promise.
EFFECT OF ELECTRODE SIZE AND LOCATION

ON DEFIBRILLATION EFFICACY
So far, we have discussed how to measure the efficacy of a
defibrillator by estimating some point on the probability of
success curve. In the next two sections we discuss some of the
factors that affect efficacy.
The electrodes that deliver the shock have a large effect
on defibrillation efficacy. The different forms of defibrillators
deliver electric shocks from different size electrodes that are
positioned on different parts of the body. ICDs initially delivered shocks to the ventricles of the heart from either two
patches on the epicardial surface of the heart [Fig. 1(a)] or a
patch on the left ventricular epicardium to a coil electrode in
the superior vena cava-right atrium. These electrode configurations had 10 J to 34 J defibrillation thresholds using monophasic waveforms. With the development of biphasic waveforms, it became possible to shock from a coil electrode in the
right ventricle to a coil in the superior vena cava-right atrium
with or without a patch on the left chest wall. This change
allowed patients to undergo implantation of an ICD without
the risk associated with a thoracotomy. Even more recently,
with the advent of smaller devices and the ability to implant
them in the pectoral region, the metallic shell of the device
has been used as the return electrode for a coil electrode in
the right ventricle [Fig. 1(b)]. With all of these electrode configurations, almost 100% of patients will have a defibrillation
threshold or estimated ED50 of less than 24 J if a biphasic
waveform is used (11).
Automatic internal atrial defibrillation is now being performed as an investigational technique. Currently, the best
electrode configuration for atrial defibrillation is a coil electrode in the right atrium and a second coil electrode in the
coronary sinus underlying the left atrial appendage (12). With
a biphasic waveform, atrial defibrillation thresholds vary
from 1.5 J to 10 J (1315). Unfortunately, the pain threshold
for patients with implanted atrial defibrillators is thought to
be less than 1 J (16), although the pain threshold has been
shown to be highly variable within a given patient and from
patient to patient. Newer techniques that use multiple electrodes and sequential shock delivery have lowered the atrial
defibrillation threshold significantly (17).
Shocks delivered from the body surface typically require
200 J to 360 J for a damped sinusoidal waveform (18), although recent evidence suggests that biphasic waveforms
117
may require less energy (19,20). Because only approximately

4% to 20% of the current delivered by transthoracic defibrillation electrodes ever reaches the heart, energy requirements
for external defibrillation greatly exceed the energy required
for internal defibrillation (2123).
THE EFFECT OF WAVEFORM SHAPE
ON DEFIBRILLATION EFFICACY
An understanding of how the magnitude and shape of the
waveform effects efficacy is vital to understanding defibrillators and defibrillation. Traditionally, the efficacy of defibrillation waveforms has been measured in terms of energy. Using energy as a measure of the efficacy of the defibrillation is
very useful when determining the engineering requirements
of a defibrillator, such as how big the components need to
be or the relative efficacy of different defibrillator waveformelectrode configurations. Several studies, however, have
shown that current may be a better measure of a defibrillation waveforms efficacy (24,25). These studies have shown
that, although the amount of energy necessary to defibrillate
a patient varies from individual to individual, the current
necessary to defibrillate remains relatively constant.
Variations in waveform shape have a large effect on a defibrillators ability to halt fibrillation. Currently, two different
waveform types are used clinically: monophasic and biphasic
(Fig. 7). In a monophasic waveform, the voltage on each electrode remains either positive or negative for the duration of
the entire waveform. In a biphasic waveform, the voltage on
the electrodes reverses part way through the duration of the
waveform. Within each type, waveforms can be described as
truncated exponential or damped sinusoidal. Most ICDs use
biphasic truncated exponential waveforms. In contrast, most
external defibrillators to date have employed damped sinusoidal monophasic waveforms. Because of the inductor necessary
to shape the damped sinusoidal waveform, these defibrillators
tend to be large and heavy. More recently, smaller, lighter
8 ms
10 ms
(a)
7.6 ms
(c)
(b)
4.2 ms
10 ms
(d)
Figure 7. Some of the different waveform shapes used in clinically

available defibrillators. (a) Truncated exponential monophasic waveform. This waveform is used in older ICDs. (b) Damped sinusoidal
monophasic waveform. This waveform is used in a majority of external defibrillators in use today. (c) Truncated exponential biphasic
waveform. This waveform is used in ICDs currently being implanted
and in newly available external defibrillators. (d) Damped sinusoidal
biphasic waveform. This waveform is used in external defibrillators
used in Russia (100).
(b)
15.0
Model response
(a)
10.0
5.0
0.0
(c)
(d)
10
15
20
Time (ms)
25
30
Leading edge
current
external defibrillators have been developed that employ truncated exponential biphasic waveforms, similar to those used
in ICDs (19,26)
Many studies have shown that certain biphasic waveforms
defibrillate with a lower current and energy than a monophasic waveform. It is important to choose the relative phase durations of the two phases of the biphasic waveform carefully
in order to realize an improvement in efficacy over the monophasic waveform. For waveforms with long time constants,
the first phase should be longer than or equal to the second
phase (27,28). For waveforms with a short time constant, the
second phase can be slightly longer than the first phase
(2931).
Several groups have shown that for square waveforms, defibrillation efficacy follows a strengthduration relationship
similar to cardiac stimulation (32,33); as the waveform gets
longer, the average current at the 50% success point becomes
progressively less, approaching an asymptote called the rheobase (34). Based on this observation, several groups have suggested that cardiac defibrillation can be mathematically modeled using a parallel resistorcapacitor (RC) network to
represent the heart (Fig. 8) (29,3537). Empirically, it has
been determined that the time constant for the parallel RC
network is in the range of 2.5 ms to 5 ms (29,31,36). In one
version of the model (29), a current waveform is applied to
the RC network. The voltage across the network is then calculated for each time point during the defibrillation pulse. The
relative efficacy of different waveform shapes and durations
can be compared by determining the current that is necessary
to make the voltage across the RC network reach a particular
value, called the defibrillation threshold.
Several observations can be made from this model. First,
for square waves, as the waveform duration gets longer, the
voltage across the network gets progressively higher and approaches an asymptote or rheobase. For truncated exponential waveforms, however, the model voltage rises, reaches a
peak, and then, if the waveform is long enough, begins to decrease (Fig. 8). Therefore, the model would predict that monophasic exponential waveforms should be truncated at a time
when the peak voltage across the RC network is reached. Current or energy delivered after that time is wasted and may
even be detrimental if the waveform gets too long (38). In
supporting this prediction, strengthduration relationships
measured in both animals (29) and humans (39) do not approach an asymptote but reach a minimum and remain there
as the waveform gets longer.
Secondly, the model predicts that the heart acts as a lowpass filter (37). Therefore waveforms that rise gradually
should have an improved efficacy over waveforms that turn
on immediately. This prediction has been shown to hold true
for external defibrillation (40), internal atrial defibrillation
(41), and internal ventricular defibrillation (42). Ascending
ramps defibrillate with a greater efficacy than do descending
ramps (42,43). Sweeney et al. showed that a square wave
duty cycle waveform (a waveform in which the current is rapidly turned on and off) defibrillates with the same efficacy as
a square waveform delivering the same total charge as the
duty cycle waveform (37). This idea has implications for new
defibrillators that would use a duty cycle concept to shape
a waveform.
Several groups have suggested that the optimal first phase
of a biphasic waveform is the optimal monophasic waveform
Leading edge
current
DEFIBRILLATORS
Model response
118
22
16
12
6
2
4
8
0
10
Time (ms)
15
20
Figure 8. The response of a parallel resistorcapacitor network representation of the heart to a monophasic and biphasic truncated exponential waveform with a time constant of 7 ms. The parallel resistor
capacitor network has a time constant of 2.8 ms. (a) Input
monophasic waveforms. Leading edge current of the input waveform
was 10 A. The waveforms were truncated at 1, 2, 3, 4, 5, 6, 8, and 10
ms. (b) Model response, V(t). Initially, as the waveform gets longer,
V(t) increases until it reaches a maximum at approximately 4 ms,
after which V(t) begins to decrease. (c) Input biphasic waveforms.
Leading edge current was 10 A. Phase 1 was truncated at 6 ms. Phase
2 was truncated after 1, 2, 3, 4, 5, 6, 7, and 8 ms. (d) The model
response does not change polarity until phase 2 duration is longer
than 2 ms.
(30,44). If this is true, then what does the model predict to be

the best second phase of a biphasic waveform? Empirically, it
seems that the role of the second phase is to return the model
voltage response as closely to zero as quickly as possible in
order to maximize the increased efficacy of the biphasic waveform over that of the monophasic waveform with the same
duration as phase one of the biphasic waveform. If the network voltage does not reach zero, or it overshoots zero, then
efficacy is lost (29,30). Swerdlow et al. have shown in humans
that the best second phase of a biphasic waveform is one that
returns the model response close to zero (31).
Together, these ideas allow one to optimize capacitor sizes
and phase durations for truncated exponential biphasic waveforms, the most commonly used waveforms in ICDs. The capacitor has to be large enough to be able to raise the network
voltage to its threshold value and still be able to hold enough
charge to drive the network voltage back to zero. For a 40
interelectrode impedance and a network time constant of 2.8
DEFIBRILLATORS
ms, the minimum capacitor size that can accomplish this is

75 F.
Only a few studies have examined the effects of the duration of fibrillation on the defibrillation threshold. Most of
these studies deal with internal defibrillation and with fibrillation durations of less than one minute (4547). In a dog
model of defibrillation, using internal electrodes and bidirectional monophasic defibrillation pulses delivered along two
pathways, Echt et al. showed that the energy necessary to
defibrillate rose from 27 13 J at 5 s of fibrillation to 41
14 J at 30 s of fibrillation (45). Jones et al. showed that in a
working rabbit heart model of defibrillation both monophasic
and biphasic waveform defibrillation thresholds increased
with duration from 5 to 15 to 30 s (47). At all durations, the
biphasic threshold was lower than the monophasic threshold.
This difference increased with fibrillation duration. In a study
using sequential trapezoidal defibrillation pulses in a pig
model of defibrillation, Fujimura et al. concluded that a delay
in defibrillation therapy of up to 90 s has no significant effect
on the ability to defibrillate the heart (48). Bardy et al. found
no difference between the mean defibrillation thresholds in
humans when fibrillation was allowed to continue for 10 versus 20 s (11.5 5.9 J versus 12.0 6.9, p NS) (46). Winkle
et al. showed that in humans the probability of successful defibrillation with low energy shocks (5.9 J) was higher for ventricular fibrillation lasting 5 s than for ventricular fibrillation
lasting 15 s, yet there was no significant difference between
the success rates of high energy shocks (24.2 J) delivered at
the same two durations (49). Together, these results suggest
that for ventricular fibrillation durations up to 90 s, the defibrillation threshold for monophasic waveforms increases with
duration while the results are inconclusive for biphasic waveforms.
from the defibrillation electrodes (Fig. 9). Just as there is a

minimum shock strength needed to defibrillate consistently,
it has been found that there is a minimum potential gradient
that must be created throughout the ventricles by the shock
to defibrillate consistently (50). The shock strength needed to
defibrillate can vary widely for different defibrillation electrode configurations; however, the minimum potential gradient that must be created throughout the heart for a particular
shock waveform is approximately the same even when the
defibrillation electrode configuration is altered (50,51). This
finding suggests that the minimum potential gradient generated throughout the heart by the shock is a more fundamental unit involved in the mechanism of defibrillation than is
the shock strength delivered to the electrodes. Yet the minimum potential gradient necessary for defibrillation is different for different waveforms. For example, it is approximately
6 V/cm for a typical monophasic waveform, but is approximately 4 V/cm for a typical biphasic waveform (51). There-
Gradient field
EPI
14
18
11
14
2 4
8
3 7
4
6
4
Potential Gradient
During a shock, different amounts of current flow through different parts of the heart. According to Ohms law, the current
density through each region of the heart is equal to the potential gradient in that region divided by the resistivity of that
region. Although current density is difficult to measure directly, techniques to measure the potential gradient are well
established. If we make the assumption that tissue resistivity
is constant in the heart, then the potential gradient is directly
related to current density. For shocks delivered from intracardiac electrodes, the distribution of potential gradients is
highly uneven. High potential gradients occur near the defibrillation electrodes. Low potential gradients occur distant
19
9
10
22
12
5
17
19
8
10
19
24
10
5
8
13
19
23
17
MECHANISMS OF DEFIBRILLATION
In the following sections, the interaction between the defibrillation shock and the fibrillating myocardium will be discussed. We start with how the distribution of the current from
the shock affects defibrillation. Then we discuss how the
shock interacts with the fibrillating myocardium. Finally, we
discuss how the shock changes the action potential, the transmembrane potential, and the ion channels in the membrane.
By looking at all of these different shockfibrillating heart
interactions, we will attempt to summarize what is known
about how an electrical shock causes the fibrillating heart to
return to sinus rhythm.
119
23
26
32
24
23
23
7
ENDO
4
2
6
11
17
3
3
5
7
5
7
8
10
13
15
17
4
Left anterolateral
view
27
21
12
21
11
28
32
37
27
21
9
24
21
6
7
8
19
11
24
22
8
Right posterolateral
view
Figure 9. The potential gradient field from a 500 V, 6 ms unsuccessful defibrillation shock delivered from a catheter electrode in the right
ventricular apex as cathode and a cutaneous patch on the lower left
thorax as anode. Left-hand panels demonstrate the heart from the
left anterolateral view; the two right-hand panels represent the right
posterolateral angle. Numbers represent the potential gradient in
volts per centimeter. Isogradient lines are separated by 10 V/cm.
(Dashed line) The upper border of the right ventricular outflow tract.
(Asterisks) The top row of electrodes in the atrium and right ventricular outflow tract where potential gradients could not be calculated
because no recording sites were above them. Neighboring electrodes
are required to calculate the potential gradient, defined as the change
in potential with distance. (Solid circles) Electrodes from which good
recordings were not obtained. (EPI) epicardial, (ENDO) endocardial.
Reproduced with permission from The Institute of Electrical and
Electronics Engineers (101).
120
DEFIBRILLATORS
fore, at this level of consideration of the mechanism of defibrillation, one of the reasons that some biphasic waveforms
require a smaller shock than some monophasic waveforms for
defibrillation is that they must create a lower minimum potential gradient throughout the heart.
Activation Sequence
There are two things that a shock must do in order to defibrillate the heart. First, it must stop all the fibrillation wavefronts on the heart. Second it must not restart fibrillation.
When shocks are given that are much lower than the strength
needed to defibrillate, activation after the shock appears at
numerous sights throughout the ventricles (52). As the shock
strength is increased, the potential gradients are increased
throughout the myocardium, and activation originates just in
those regions in which the potential gradients are lowest. For
shocks just slightly lower than the strength needed to defibrillate, postshock activation arises only in the small myocardial regions in which the potential gradients remain below
the minimum needed for defibrillation (Fig. 10). Activation
fronts arising from these low potential gradient regions propagate to activate the remainder of the myocardium for a few
cycles following the shock. Then reentry occurs, activation becomes disorganized, and the heart begins to fibrillate again.
Following shocks slightly stronger than that needed to defibrillate, postshock sites of early activation still arise in the
regions of lowest potential gradient; however, successive cycles of activation originate from these regions more slowly
than following unsuccessful, slightly lower strength shocks.
After a few cycles, these activations terminate without reinducing fibrillation (52,53). These results suggest that the reason a minimum potential gradient is required for defibrillation is that, above this minimum, activation fronts are not
generated by the shock that can interact and reinduce fibrillation (53). For shocks delivered through transvenous electrodes at strengths a few hundred volts higher than needed
to defibrillate, ectopic activation fronts first appear following
the shock in regions exposed to the highest potential gradients generated by the shocks, adjacent to the defibrillation
electrodes (54,55). In a similar fashion to activation fronts
arising following shocks just above the defibrillation threshold, these activation fronts also terminate without reinitiating
fibrillation. However, when the shock strength is increased
still further, e.g. to above 1000 V with transvenous electrodes,
the activation fronts arising from the high potential gradient
regions can reinduce ventricular fibrillation, so that the shock
fails even though a lower strength shock succeeds.
Cellular Action Potential
One effect of the shock field is to initiate a new action potential [Fig. 11(a)]. If the shock strength is large enough, new
action potentials can be generated both in tissue adjacent to
the defibrillation electrodes, and in regions throughout the
myocardium distant from the defibrillation electrodes (56,57).
Under certain conditions, the shock can have a second effect
on the action potential. It can cause prolongation of the action
potential and, as a result, a prolongation of the refractory period, without giving rise to a new action potential [Fig. 11(b)].
This action potential prolongation, called a graded response
by some (58), occurs if a shock of sufficient strength is given
when the cells are in their refractory period. Action potential
prolongation occurs when the shock potential gradient deliv-
ered to the tissue is above the minimum level needed to defibrillate. At shock strengths less than the minimum needed to
defibrillate, only an all-or-none response is observed [Fig.
4(a)] (59,60). Although these effects are important for electrical induction of reentry, the fact that action potential prolongation occurs in response to shock field strengths that occur
during defibrillation suggests that action potential prolongation may also be important for defibrillation (59,6165). Action potential prolongation and, hence, refractory period extension are hypothesized to play two different roles in the
mechanism of defibrillation. One, they are thought to prevent
the appearance of propagating activation fronts following the
shock (52,66). Two, by causing a more uniform dispersion of
refractoriness following the shock, they are thought to prevent the block and reentry that cause the activation fronts
that do appear following the shock from degenerating into fibrillation (6165).
In regions in which the shock potential gradient is high,
over 50 V/cm to 70 V/cm, the shock can have detrimental
effects, probably by causing electroporation of the cardiac cell
membranes (67,68). This can cause the transmembrane potential to temporarily hang up near the value of the plateau
of the action potential (69). The cell is electrically paralyzed
and cannot conduct an action potential during this time. At
yet higher potential gradients, probably over 150 V/cm, the
exposed myocardium gives rise to arrhythmic beats (50). This
may be the mechanism that causes the probability of the defibrillation success curve to decrease for very large shocks
(Fig. 5).
Waveform shape alters the shock strength at which these
detrimental effects on the myocardium occur. Yabe et al.
showed that for a 10 ms truncated exponential monophasic
waveform, conduction block occurred in dogs in regions where
the potential gradient was greater than 64 4 V cm1 (70).
Conduction block would last longer for shocks that created
even higher potential gradients in the myocardium. In contrast to monophasic shocks, conduction block occurred when
the potential gradient in the myocardium reached 71 6 V/
cm for a 5 ms/5 ms truncated exponential biphasic shock.
Jones et al. showed that adding a second phase to a monophasic waveform (i.e., making it a biphasic waveform) decreased
the damage done to cultured chick myocytes by the monophasic waveform alone (71). Both results show that biphasic
waveforms are less apt to cause damage or dysfunction in
high-gradient regions. The therapeutic index has been defined
as the range of energies over which a defibrillation waveform
is both safe and effective. Since biphasic waveforms defibrillate at lower energies and cause more myocardial damage
than monophasic waveforms, they have been described as
having a higher therapeutic index than monophasic waveforms.
Transmembrane Potential
For the shock to cause either a new action potential to be
triggered or to prolong an action potential, it must alter the
transmembrane potential. It has been estimated that only
about one quarter of the total current traversing the heart
crosses the membrane to enter the cells (72). Since the defibrillation electrodes are located extracellularly, current from
the shock that enters myocardial cells in some regions must
exit the cells in other regions. These currents, which flow
through the cell membrane, will introduce changes in the
DEFIBRILLATORS
121
First post shock beat

EPI
ENDO
149
127
149
149
146
151
165
136
162
135
37 154
196
162
138
39
160
144
176
131
160
169
151
176
167
159
144
126
116
69
143 165
122
134
142
118
131
157
150
151
142
146
133
149
155
149
153 141
156
155
159
152
206
189
147
157
137
135
144
150
155
143
171
199
192
139
160
166
175
152
144
133
151 149
60
202
148
144
142
152
154
151
190
213
179
156
133
142
202
169
81
177
141
Second post shock beat

ENDO
EPI
325
325
327
327
173
329
337 349
320
320
312
299
320
325
314
315
338
158
280
317
315 335
351
301
316
296
323
329
336
344
324
306
311
350
236
314
183
295
25
388
326
302
377
280
342
297 306
306
273
290
292
313
314
318
327
319
325
285
339
352
306
303
311
301
301
283
270
278
297
320
368
244
322
339
195 269
303
316
319
306
345
329
303
300
327
369
389
336
293
291
289
302
341
306
267
370
396
325
330
316
Third post shock beat

EPI
ENDO
470
470
470
467
479
32
422
450
422
452
473
437
474
521
426
458
442 461
477
426
441
412
424
466
461
506
483
405
408
516
560
436
377
407
424
460
466
453
477
31
339
402
470
485
517
480474
466
446
456
423
448
453 453
429
438
435
414
448
453
372
319
565
424
450
374
417 469
432
380
426 431
434
414
468
480
429
410
402
397
457
501
466
449
482
388
450
456
506
401
353
477
478
436
424
461
414
430
424
411
Figure 10. Postshock activation sequence. The first three cycles after the unsuccessful 500 V,
6 ms defibrillation shock shown in Fig. 9. Numbers represent activation times in milliseconds.
(U) Isochronal lines, separated by 20 ms. () sites of electrodes where adequate recordings were
not obtained. () represent conduction block; (---) Frame shift from one isochronal map to the
next. Such dashed lines are necessary whenever a dynamic process such as reentrant activation
is illustrated by a series of static maps. Reproduced with permission from The Institute of Electrical and Electronics Engineers (101).
transmembrane potential that include depolarization or hyperpolarization during the shock pulse. Several mathematical
formulations have been proposed to describe which regions of
the heart are depolarized and which are hyperpolarized during shocks from a particular defibrillation electrode configuration. These formulations include the cable equations, the
sawtooth model (73,74), the bidomain model (75,76), and the
secondary source model (77). In their simplest form, these formulations incorporate the extracellular and intracellular
spaces as low resistance media and the membrane as a high
resistance in parallel with a capacitance. Therefore, these
simple case models incorporate only passive myocardial properties. Recently, the models have been made more realistic by
the addition of active components to represent the ion chan-
122
DEFIBRILLATORS
1.6 V/cm, 2 ms
S1-S2 (MS)
0
(mV)
222
225
80_
100 ms
(a)
8.4 V/cm, 2 ms
S1-S2 (MS)
90
110
130, 150, 170
180
200
210
220
230
(mV)
80_
230
90
100 ms
(b)
Figure 11. (a) Recordings that illustrate the response to an S2 stimulus of 1.6 V/cm oriented
along the fibers. The S1S2 stimulus intervals for each of the responses are indicated to the
right of the recordings. The responses are markedly different even though the change in S2
timing was only 3 ms. An S1S2 interval of 222 ms caused almost no response, whereas an
interval of 225 ms produced a new action potential. (b) A range of action potential extensions
produced by an S2 stimulus generating a potential gradient of 8.4 V/cm oriented along the long
axis of the myofibers. The recordings were obtained from the same cell as (a). The action potential
recordings, obtained from one cellular impairment, are aligned with the S2 time. An S1 stimulus
was applied 3 ms before phase-zero of each recording. The longest and shortest S1S2 intervals
tested, 230 ms and 90 ms respectively, are indicated beneath their respective phase-zero depolarizations. The S1S2 intervals for each response after S2 are indicated to the right. Reproduced
with permission from the American Heart Association (59).
nels in the membrane. Because the extracellular space

throughout the body is primarily resistive, with very little reactive components, the defibrillation shock appears in the extracellular space of the heart almost immediately and without
significant distortion. For example, a shock in the form of a
square wave given across the defibrillation electrodes will appear almost immediately as a square wave in the extracellular space of the heart. Because of the capacitance and the ion
channels of the membrane, however, phase delays and alter-
ations of the appearance of the shock wave occur in the transmembrane potential. For example, a square wave shock may
appear as an exponential change in the transmembrane potential that reaches an asymptote (Fig. 12). Because of the
nonlinear behavior of the membrane introduced by the ion
channels, reversing defibrillation shock polarity does not just
reverse the sign of the change in the transmembrane potential but also alters the magnitude and time-course of the
change in transmembrane potential (Fig. 12).
DEFIBRILLATORS
Extracellular space
Transmembrane
potential
10 ms
(a)
Extracellular space
Transmembrane
potential
10 ms
(b)
Figure 12. The effect of a square wave shock on the extracellular

potential and the transmembrane potential. The square wave shock
appears immediately as a relatively undistorted square wave in the
extracellular space. It appears as an exponentially increasing change
in the transmembrane potential. When given during the action potential plateau, as shown in the figure, the depolarization obtained when
a shock of one shock polarity is delivered has a different magnitude
and time-course compared to the hyperpolarization obtained when a
shock of the opposite polarity is delivered. Reproduced with permission of the North American Society of Pacing and Electrophysiology
(102).
The one-dimensional cable model indicates that the tissue

near the anode during the defibrillation shock should be hyperpolarized, whereas the tissue near the cathode should be
depolarized (78). This hyperpolarization and depolarization
decreases exponentially with distance away from the electrodes. The distance at which the depolarization or hyperpolarization has decreased by 63% is called the membrane space
constant. The space constant for cardiac tissue is only 0.5 mm
to 1.0 mm (78,79). Therefore, the one-dimensional cable equations predict that tissue more than about 1 cm away from
the defibrillation electrodes (i.e. ten space constants) should
undergo almost no change in transmembrane potential
caused directly by the shock field. Thus, according to one-dimensional cable theory, the shock should not be able to directly excite a new action potential at distances more than 1
cm away from the electrodes. This prediction contrasts with
the experimental finding in hearts that new action potentials
can be created by shocks many centimeters away from the
shock electrodes (56,57).
The sawtooth formulation states that, because of the junctional resistance at the gap junctions between cells, cells in
the region away from and between shock electrodes undergo
hyperpolarization at the cell end facing the anode and depolarization at the cell end facing the cathode. Thus the change
in transmembrane potential during the shock assumes a sawtooth distribution with each tooth of the sawtooth corresponding to an individual cell (73,74,80,81). While a sawtooth
change in the transmembrane potential during a shock has
been observed in single, isolated cells (82) it has never been
observed in a syncytium of cardiac cells experimentally
(8386).
The bidomain formulation performs the mathematical legerdemain of representing in two or more dimensions the extracellular space as an everywhere continuous domain and
123
the intracellular space as an everywhere continuous domain

with both domains separated by the highly resistive cell membrane (87). When the ratio of the extracellular resistance
along fibers to across fibers is equal to the ratio of the intracellular resistance along fibers to across fibers, the bidomain
formulation predicts an effect similar to that predicted by the
cable equations. In this case, hyperpolarization occurs in tissue under an extracellular anodal electrode and the magnitude of hyperpolarization decreases exponentially with distance in the direction along or across fibers according to the
space constants along and across fibers. When the ratio of
the intracellular resistivities is not equal to the ratio of the
extracellular resistivities, however, bidomain theory differs
from the results of the cable equations. An important difference is that, while hyperpolarization still occurs immediately
adjacent to an anode during a shock, depolarization occurs
just a few millimeters away from the electrode along the long
axis of the myocardial fibers. Similarly, while depolarization
occurs immediately adjacent to a cathode, hyperpolarization
occurs along fibers just a few millimeters away (8890). Several recent experiments have verified this prediction (Fig. 13)
(88). The bidomain formulation does not predict a constant
relationship between the extracellular potential gradient generated by the shock and the change in transmembrane potential caused by the shock. Rather, the change in transmembrane potential depends on a complex distribution of
intracellular and extracellular current that involves the
change over distance of the potential gradient, the distance
from the electrode, and the direction of the myocardial fibers
over this distance. At first glance, this prediction seems to
conflict with the experimental finding that early sites of activation following failed defibrillation shocks occur in regions of
lowest potential gradient and that a certain minimum potential gradient is necessary for defibrillation (50,51). However,
for most commonly used defibrillation electrode configurations, the change in the potential gradient with distance is
lowest in those regions in which the potential gradient itself
is lowest. Therefore, these experimental findings may not
necessarily be in conflict with the predictions of bidomain
theory.
One limitation of the bidomain theory in its simplest form
is that it does not take into consideration the discontinuities
of the intracellular domain where the myocardium is crossed
by connective tissue septae, blood vessels, and scar tissue.
Any intracellular current that needs to cross such barriers
must leave the intracellular space on one side of the barrier
and reenter it on the other. Thus, depolarization should occur
on one side of the barrier and hyperpolarization on the other.
In other words, the connective tissue barrier will act as if it
is a pair of electrodes during the shock, acting as a secondary
source. For this reason, the gaps in the tissue where myocardial cells are not present have been considered to act as secondary sources. Recent studies by Gillis et al. (85) and White
et al. (91) suggest that such secondary sources are important
causes of depolarization and hyperpolarization throughout
the myocardial tissue during a shock.
Ion Channels
The electrical activity of the heart at its most fundamental
level is controlled by the ion channels located in the cell membrane. These channels selectively allow ions such as Na and
Ca into the cardiac cell and K out of the cell in response to
124
DEFIBRILLATORS
Anodal
500 ms
15
23
31
39
47
55
63
14
22
30
38
46
54
62
13
21
29
37
45
53
61
12
20
28
36
44
52
60
11
19
27
35
43
51
59
10
18
26
34
42
50
58
17
25
33
41
49
57
16
24
32
40
48
56
changes in the transmembrane potential. It is known that

both the fast as well as the slow Na and Ca channels are
active during early fibrillation when defibrillation shocks are
most likely to be given (92,93). It is thought that direct excitation of a new action potential by the shock is caused by activation of the sodium channels (9496). Results from computer
models have suggested that the role of the first phase of a
biphasic defibrillation waveform is to hyperpolarize the cardiac cell membrane from the 65 mV that is typically its
most negative transmembrane voltage during fibrillation to
closer to the 80 mV to 90 mV that is the resting transmembrane voltage. This decrease in transmembrane potential
is hypothesized to allow the transmembrane voltage-dependent Na channels to recover. Because the Na channels have
recovered, the second phase of the biphasic waveform can
more easily stimulate tissue and defibrillate the heart (97).
These results are in direct conflict with the ideas presented
earlier that the first phase of a biphasic waveform stimulates,
while the second phase keeps the heart from refibrillating.
More research is necessary to reconcile these results.
CONCLUSION
(a)
Anodal
0.17
0.06
0.50
1.00
0.89
0.47
0.91 1.06
0.8
0.6
0.20.4
0.0
0.22
0.12
0.21
0.35
0.24
0.31
0.38
0.19
0.31
0.53
0.43
0.50
0.44
0.41
0.38
0.27
0.50
0.45
0.52
0.44
0.58
0.53
0.39
0.22
0.55
0.55
0.33
0.35
0.41
0.42
0.33
0.38
0.41
0.37
0.32
0.35
0.40
0.41
0.14
0.19
0.0
0.29
0.2
0.31
0.32
0.32
0.31
0.35
0.16
0.25
0.31
0.24
0.25
0.21
0.0
0.63
0.
BIBLIOGRAPHY
4
0.11
0.34
0.05
We have examined how various aspects of the electric shock

interact with the heart at many different levels to stop fibrillation. Important aspects of the shock include its shape, the
electrodes that it is delivered from, and the potential gradient
field that is created in the heart because of it. The shock affects the activation sequence of the fibrillating heart, the cellular action potential, the transmembrane potential, and the
cellular ion channels in a specific fashion to stop fibrillation
and allow the heart to resume sinus rhythm. Understanding
these interactions will allow physicians, engineers, and researchers to build more effective defibrillators and thereby extend life.
(b)
1 mm
Figure 13. Fluorescence recordings showing the effect of an anodal

stimulation pulse in a region adjacent to a stimulation electrode and
contour plot of changes in transmembrane potential induced during
the pulse. (a) Each recording shows an S1-induced action potential
without an S2 stimulation pulse and then an S1-induced action potential during which an S2 stimulation pulse was applied from a line
of electrode terminals (row of large spots) illustrated above the recording region. Positive changes in transmembrane potential, i.e., depolarization during S2, occurred at most spots. (b) Contour plot showing the distribution of depolarization in the recording region. The
depolarization at each recording spot is given as a fraction of the
amplitude of the S1-induced action potential rising phase. Contour
lines are shown at intervals of 0.2 times the amplitude of the action
potential rising phase. Reproduced with permission from the Biophysical Society (88).
1. AVID Trial Executive Committee, Are implantable cardioverterdefibrillators or drugs more effective in prolonging life?, Am. J.
Cardiol., 79: 661663, 1997.
2. A. J. Moss et al., Improved survival with an implanted defibrillator in patients with coronary disease at high risk for ventricular arrhythmia, N. Engl. J. Med., 335: 19331940, 1996.
3. M. L. Weisfeldt et al., American Heart Association report on the
public access defibrillation conference, 1994. Automatic external
defibrillation task force, Circulation, 92: 27402747, 1995.
4. B. E. Gliner, Y. Murakawa, and N. V. Thakor, The defibrillation
success rate versus energy relationship: Part I-Curve fitting and
the most efficient defibrillation energy, Pacing Clin. Electrophysiol., 13: 326338, 1990.
5. M. H. Lehmann et al., Defibrillation threshold testing and other
practices related to AICD implantation: do all roads lead to
Rome?, Pacing Clin. Electrophysiol., 12: 15301537, 1989.
6. W. C. McDaniel and J. C. Schuder, An up-down algorithm for
estimation of the cardiac ventricular defibrillation threshold,
Med. Instrum., 22: 286292, 1988.
7. L. Wang et al., Dependent success rate of repeated shocks at
DFT determined by binary search, Pacing Clin. Electrophysiol.,
20: 1169, 1997.
8. J. D. Bourland, W. A. Tacker, Jr., and L. A. Geddes, Strengthduration curves for trapezoidal waveforms of various tilts for
DEFIBRILLATORS
transchest defibrillation in animals, Med. Instrum., 12: 3841,
1978.
9. R. A. Malkin, T. C. Pilkington, and D. S. Burdick, Optimizing
existing defibrillation thresholding techniques, presented at
Annu. Int. Conf. IEEE Eng. Med. Biol. Soc., Piscataway, NJ,
1990.
10. R. A. Malkin et al., Estimating the 95% effective defibrillation
dose, IEEE Trans. Biomed. Eng., 40: 256265, 1993.
11. F. P. van Rugge, L. H. Savalle, and M. J. Schalij, Subcutaneous
single-incision implantation of cardioverter-defibrillators under
local anesthesia by electrophysiologists in the electrophysiology
laboratory, Am. J. Cardiol., 81: 302305, 1998.
125
27. E. G. Dixon et al., Decreased defibrillation thresholds with large

contoured patch electrodes in dogs, J. Am. Coll. Cardiol., 9:
143A, 1987.
28. S. A. Feeser et al., Strength-duration and probability of success
curves for defibrillation with biphasic waveforms, Circulation,
82: 21282141, 1990.
29. G. P. Walcott et al., Choosing the optimal monophasic and biphasic waveforms for ventricular defibrillation, J. Cardiovasc.
Electrophysiol., 6: 737750, 1995.
30. M. W. Kroll, A minimal model of the single capacitor biphasic
defibrillation waveform, Pacing Clin. Electrophysiol., 17: 1782
1792, 1994.
12. R. A. S. Cooper et al., Comparison of multiple biphasic and monophasic waveforms for internal cardioversion of atrial fibrillation in humans, Circulation, 90: 113, 1994.
31. C. D. Swerdlow, W. Fan, and J. E. Brewer, Charge-burping theory correctly predicts optimal ratios of phase duration for biphasic defibrillation waveforms, Circulation, 94: 22782284, 1996.
13. R. A. S. Cooper, E. E. Johnson, and M. Wharton, Internal atrial

defibrillation in humans. Improved efficacy of biphasic waveforms and the importance of phase duration, Circulation, 95:
14871496, 1997.
32. H. A. Blair, On the intensity-time relations for stimulation by

electric currents. II, J. Gen. Physiol., 15: 731755, 1932.
14. C.-P. Lau and N.-S. Lok, A comparison of transvenous atrial

defibrillation of acute and chronic atrial fibrillation and the effect of intravenous Sotalol on human atrial defibrillation threshold, Pacing Clin. Electrophysiol., 20: 24422452, 1997.
15. A. Heisel et al., Low-energy transvenous cardioversion of atrial
fibrillation using a single atrial lead system, JCW, 8: 607614,
1997.
16. G. Tomassoni et al., Testing different biphasic waveforms and
capacitances: effect on atrial defibrillation threshold and pain
perception, J. Am. Coll. Cardiol., 28: 695699, 1996.
17. R A. S. Cooper, W. M. Smith, and R. E. Ideker, Dual current
pathways for internal atrial defibrillation in sheep: marked reduction in defibrillation threshold, J. Am. Coll. Cardiol., 29:
195A, 1997.
18. W. D. Weaver et al., Ventricular defibrillationa comparative
trial using 175-J and 320-J shocks, N. Engl. J. Med., 307: 1101
1106, 1982.
19. G. H. Bardy et al., Multicenter comparison of truncated biphasic
shocks and standard damped sine wave monophasic shocks for
transthoracic ventricular defibrillation. Transthoracic Investigators, Circulation, 94: 25072514, 1996.
20. H. L. Greene et al., Comparison of monophasic and biphasic defibrillating pulse waveforms for transthoracic cardioversion,
Am. J. Cardiol., 75: 11351139, 1995.
21. F. J. Claydon, III et al., A volume conductor model of the thorax
for the study of defibrillation fields, IEEE Trans. Biomed. Eng.,
35: 981992, 1988.
22. W. J. Karlon, S. R. Eisenberg, and J. L. Lehr, Effects of paddle
placement and size on defibrillation current distribution: a
three-dimensional finite element model, IEEE Trans. Biomed.
Eng., 40: 246255, 1993.
23. B. B. Lerman and O. C. Deale, Relation between transcardiac
and transthoracic current during defibrillation in humans, Circ.
Res., 67: 14201426, 1990.
24. R E. Kerber et al., Energy, current, and success in defibrillation
and cardioversion: clinical studies using an automated impedance-based method of energy adjustment, Circulation, 77: 1038
1046, 1988.
25. B. B. Lerman, J. P. DiMarco, and D. E. Haines, Current-based
versus energy-based ventricular defibrillation: a prospective
study, J. Am. Coll. Cardiol., 12: 12591264, 1988.
26. J. E. Poole et al., Low-energy impedance-compensating biphasic
waveforms terminate ventricular fibrillation at high rates in victims of out-of-hospital cardiac arrest. LIFE Investigators, J.
Cardiovasc. Electrophysiol., 8: 13731385, 1997.
33. L. Lapicque, in LExcitabilite en Fonction du Temps. Paris: Libraire J. Gilbert, 1926.

34. G. A. Mouchawar et al., Ability of the Lapicque and Blair
strength-duration curves to fit experimentally obtained data
from the dog heart, IEEE Trans. Biomed. Eng., 36: 971974,
1989.
35. W. Irnich, The fundamental law of electrostimulation and its
application to defibrillation, Pacing Clin. Electrophysiol., 13:
14331447, 1990.
36. M. W. Kroll, A minimal model of the monophasic defibrillation
pulse, Pacing Clin. Electrophysiol., 16: 769777, 1993.
37. R. J. Sweeney et al., Defibrillation using a high-frequency series
of monophasic rectangular pulses: observations and model predictions, J. Cardiovasc. Electrophysiol., 7: 134143, 1996.
38. J. C. Schuder et al., Transthorecic defibrillation in the dog with
truncated and untruncated exponential waveforms, Trans.
IEEE Biomed. Eng. 18 (6): 410415, 1997.
39. M. R. Gold and S. R. Shorofsky, Strength-duration relationship
for human transvenous defibrillation, Circulation, 96: 3517
3520, 1997.
40. G. P. Walcott et al., Comparison of damped sinusoidal and truncated exponential waveforms for external defibrillation, J. Am.
Coll. Cardiol., 27: 237A, 1996.
41. M. T. Harbinson et al., Rounded biphasic waveform reduces energy requirements for transvenous catheter cardioversion of
atrial fibrillation and flutter, Pacing Clin. Electrophysiol., 20:
226229, 1997.
42. R. E. Hillsley et al., Is the second phase of a biphasic defibrillation waveform the defibrillating phase?, Pacing Clin. Electrophysiol., 16: 14011411, 1993.
43. J. C. Schuder, G. A. Rahmoeller, and H. Stoeckle, Transthoracic
ventricular defibrillation with triangular and trapezoidal waveforms, Circ. Res., 19: 689694, 1966.
44. G. P. Walcott et al., Choosing the optimum monophasic and biphasic waveforms for defibrillation, Pacing Clin. Electrophysiol.,
17: 789, 1994.
45. D. S. Echt et al., Influence of ventricular fibrillation duration on
defibrillation energy in dogs using bidirectional pulse discharges, PACE, 11: 13151323, 1988.
46. G. H. Bardy et al., A prospective, randomized evaluation of effect of ventricular fibrillation duration on defibrillation thresholds in humans, JACC, 13 (6): 13621366, 1989.
47. J. L. Jones et al., Increasing fibrillation duration enhances relative asymmetrical biphasic versus monophasic defibrillator
waveform efficacy, CIRCRES, 67: 376384, 1990.
126
DEFIBRILLATORS
48. O. Fujimura et al., Effects of time to defibrillation and subthreshold preshocks on defibrillation success in pigs, PACE,
12(2): 358365, 1989.
68. S. B. Knisley and A. O. Grant, Asymmetrical electrically induced injury of rabbit ventricular myocytes, J. Mol. Cell.
Cardiol., 27: 11111122, 1995.
49. R. A. Winkle et al., Effect of duration of ventricular fibrillation

on defibrillation efficacy in humans, CIRC, 81: 14771481,
1990.
69. E. N. Moore and J. F. Spear, Electrophysiologic studies on the

initiation, prevention, and termiation of ventricular fibrillation, in D. P. Zipes and J. Jalife (eds.), Cardiac Electrophysiology
and Arrhythmias, Orlando: Grune & Stratton, 1985, pp. 315
322.
50. J. M. Wharton et al., Cardiac potential and potential gradient

fields generated by single, combined, and sequential shocks during ventricular defibrillation, Circulation, 85: 15101523,
1992.
70. S. Yabe et al., Conduction disturbances caused by high current

density electric fields, Circ. Res., 66: 11901203, 1990.
51. X. Zhou et al., Epicardial mapping of ventricular defibrillation

with monophasic and biphasic shocks in dogs, Circ. Res., 72:
145160, 1993.
71. J. L. Jones and R. E. Jones, Decreased defibrillator-induced dysfunction with biphasic rectangular waveforms, Am. J. Physiol.,
247: H792H796, 1984.
52. N. Shibata et al., Epicardial activation following unsuccessful

defibrillation shocks in dogs, Am. J. Physiol., 255: H902H909,
1988.
72. J. C. Eason, Membrane polarization in a bidomain model of electrical field stimulation of myocardial tissue, Duke Univ., Durham, NC, 1995.
53. P.-S. Chen et al., Comparison of activation during ventricular

fibrillation and following unsuccessful defibrillation shocks in
open chest dogs, Circ. Res., 66: 15441560, 1990.
73. R. Plonsey and R. C. Barr, Inclusion of junction elements in a

linear cardiac model through secondary sources: application to
defibrillation, Med. Biol. Eng. Comput., 24: 137144, 1986.
54. R. G. Walker et al., Sites of earliest activation following transvenous defibrillation, Circulation, 90: 1447, 1994.
74. R. Plonsey and R. C. Barr, Effect of microscopic and macroscopic

discontinuities on the response of cardiac tissue to defibrillating
(stimulating) currents, Med. Biol. Eng. Comput., 24: 130136,
1986.
55. R. G. Walker, W. M. Smith, and R. E. Ideker, Activation patterns following defibrillation with different waveforms, Pacing
Clin. Electrophysiol., 18 (part II): 835, 1995.
56. P. G. Colavita et al., Determination of effects of internal countershock by direct cardiac recordings during normal rhythm,
Am. J. Physiol., 250: H736H740, 1986.
57. K. F. Kwaku and S. M. Dillon, Shock-induced depolarization of
refractory myocardium prevents wave-front propagation in defibrillation, Circ. Res., 79: 957973, 1996.
58. C. Y. Kao and B. F. Hoffman, Graded and decremental response
in heart muscle fibers, Am. J. Physiol., 194: 187196, 1958.
59. S. B. Knisley, W. M. Smith, and R. E. Ideker, Effect of field
stimulation on cellular repolarization in rabbit myocardium: implications for reentry induction, Circ. Res., 70: 707715,
1992.
60. S. B. Knisley and B. C. Hill, Optical recordings of the effect of
electrical stimulation on action potential repolarization and the
induction of reentry in two-dimensional perfused rabbit epicardium, Circulation, 88 (part I): 24022414, 1993.
61. J. F. Swartz et al., Conditioning prepulse of biphasic defibrillator waveforms enhances refractoriness to fibrillation wavefronts, Circ. Res., 68: 438449, 1991.
62. R. J. Sweeney et al., Ventricular refractory period extension
caused by defibrillation shocks, Circulation, 82: 965972, 1990.
63. S. M. Dillon, Optical recordings in the rabbit heart show that
defibrillation strength shocks prolong the duration of depolarization and the refractory period, Circ. Res., 69: 842856,
1991.
64. S. M. Dillon and R. Mehra, Prolongation of ventricular refractoriness by defibrillation shocks may be due to additional depolarization of the action potential, J. Cardiovasc. Electrophysiol., 3:
442456, 1992.
65. S. M. Dillon, Synchronized repolarization after defibrillation
shocks: a possible component of the defibrillation process demonstrated by optical recordings in rabbit heart, Circulation, 85:
18651878, 1992.
66. P.-S. Chen, P. D. Wolf, and R. E. Ideker, Mechanism of cardiac
defibrillation: a different point of view, Circulation, 84: 913
919, 1991.
67. O. Tovar and L. Tung, Electroporation of cardiac cell membranes with monophasic or biphasic rectangular pulses, Pacing
Clin. Electrophysiol., 14: 18871892, 1991.
75. N. G. Sepulveda, B. J. Roth, and J. P. Wikswo, Jr., Current

injection into a two-dimensional anisotropic bidomain, Biophys.
J., 55: 987999, 1989.
76. N. Trayanova, A bidomain model for ring stimulation of a cardiac strand, IEEE Trans. Biomed. Eng., 41: 393397, 1994.
77. N. A. Trayanova, T. C. Pilkington, and C. S. Henriquez, A periodic bidomain model for cardiac tissue, presented at Annu. Int.
Conf. IEEE Eng. Med. Biol. Soc., Piscataway, NJ, 1991.
78. S. Weidmann, Electrical constants of trabecular muscle from
mammalian heart, J. Physiol., 210: 10411054, 1970.
79. A. G. Kleber and C. B. Riegger, Electrical constants of arterially
perfused rabbit papillary muscle, J. Physiol., 385: 307324,
1987.
80. W. Krassowska et al., Potential distribution in three-dimensional periodic myocardium: part II, Application to extracellular
stimulation, IEEE Trans. Biomed. Eng., 37: 267284, 1990.
81. W. Krassowska, T. C. Pilkington, and R. E. Ideker, Potential
distribution in three-dimensional myocardium: part I, Solution
with two-scale asymptotic analysis, IEEE Trans. Biomed. Eng.,
37: 252266, 1990.
82. S. B. Knisley et al., Optical measurements of transmembrane
potential changes during electric field stimulation of ventricular
cells, Circ. Res., 72: 255270, 1993.
83. X. Zhou et al., Optical transmembrane potential measurements
during defibrillation-strength shocks in perfused rabbit hearts,
Circ. Res., 77: 593602, 1995.
84. X. Zhou et al., Spatial changes in transmembrane potential during a shock, Pacing Clin. Electrophysiol., 18 (part II): 935, 1995.
85. A. M. Gillis et al., Microscopic distribution of transmembrane
potential during application of defibrillatory shocks in strands
and monolayers of cultured myocytes, Pacing and Clin. Electrophysiol., 19: 570, 1996.
86. J. P. Wikswo Jr., Tissue anisotropy, the cardiac bidomain, and
the virtual cathode effect, in D. P. Zipes and J. Jalife, (eds.),
Cardiac Electrophysiology: From Cell to Bedside, 2nd ed. Philadelphia: W. B. Saunders, 1995, pp. 348362.
87. L. Tung, A bidomain model for describing ischemic myocardial
DC potentials, MIT, Cambridge, MA, 1978.
88. S. B. Knisley, B. C. Hill, and R. E. Ideker, Virtual electrode
effects in myocardial fibers, Biophys. J., 66: 719728, 1994.
DELAY CIRCUITS
89. S. B. Knisley, Transmembrane voltage changes during unipolar
stimulation of rabbit ventricle, Circ. Res., 77: 12291239, 1995.
90. J. P. Wikswo, Jr., S.-F. Lin, and R. A. Abbas, Virtual electrodes
in cardiac tissue: A common mechanism for anodal and cathodal
stimulation, Biophys. J., 69: 21952210, 1995.
91. J. B. White et al., Myocardial discontinuities: a substrate for
producing virtual electrodes to increase directly excited areas of
the myocardium by shocks, Circulation, 97: 1998.
92. T. Akiyama, Intracellular recording of in situ ventricular cells
during ventricular fibrillation, Am. J. Physiol., 240: H465
H471, 1981.
93. X. Zhou et al., Existence of both fast and slow channel activity
during the early stage of ventricular fibrillation, Circ. Res., 70:
773786, 1992.
94. J. L. Jones and R. E. Jones, Threshold reduction with biphasic
defibrillator waveforms: role of excitation channel recovery in a
computer model of the ventricular action potential, J. Electrocardiol., 23: 3035, 1990.
95. L. Tung and J.-R. Borderies, Analysis of electric field stimulation of single cardiac muscle cells, J. Physiol., 63: 371386,
1992.
96. X. Zhou et al., Transmembrane potential changes caused by
shocks in guinea pig papillary muscle, Am. J. Physiol., 271:
H2536H2546, 1996.
97. J. L. Jones and R. E. Jones, Threshold reduction with biphasic
defibrillator waveforms: role of excitation channel recovery in
computer model of the ventricular action potential, J. Electrocardiol., 23: 3035, 1991.
98. T. D. Valenzuela et al., Estimating effectiveness of cardiac arrest intervention: a logistic regression survival model, Circulation, 96: 33083313, 1997.
99. J. C. Schuder et al., Transthoracic ventricular defibrillation in
the 100 kg calf with untruncated and truncated exponential
stimuli, IEEE Trans. Biomed. Eng., BME 27: 3743, 1980.
100. N. L. Gurvich and V. A. Markarychev, Defibrillation of the heart
with biphasic electrical impulses, Kardiologiia, 7: 109112,
1967.
101. A. S. L. Tang et al., Measurement of defibrillation shock potential distributions and activation sequences of the heart in threedimensions, Proc. IEEE, 76: 11761186, 1988.
102. G. P. Walcott et al., On the mechanism of ventricular defibrillation, Pacing Clin. Electrophysiol., 20: 422431, 1997.
GREGORY P. WALCOTT
RAYMOND E. IDEKER
University of Alabama at
Birmingham
DEGREE PROGRAMS. See ELECTRICAL ENGINEERING CURRICULA.
127
370
ELECTROCARDIOGRAPHY
Physiological Basis
ELECTROCARDIOGRAPHY
Electrocardiography is the study of the hearts electrical activity recorded from the surface of the body. Such recordings
represent a total or integrated view of all of the electrically
excitable cells in the heart. A sensitive medical recording device, called an electrocardiograph, is attached to the body
with special electrodes and records the voltage changes on
chart paper. This voltage versus time recording is the electrocardiogram. Both the device and its graphical output are abbreviated by the familiar acronym ECG and, depending on
its contextual use, one could be referring to either one. Because much of the original work in this field was performed
by Willem Einthoven in Holland, the abbreviation of EKG
was based on the Dutch root word kardio and is interchangeably used with ECG.
The ECG provides the physician with a significant series
of waves from which one can measure the rate, rhythm, and
many aspects of the health of the various cardiac muscle tissues that comprise the heart. The actual recording devices
have kept pace with advances in modern technology, so that
todays recording devices use integrated electronics and embedded microprocessors to record, analyze, and store the signals generated by the heart. In addition, a wide variety of
medical devices rely on an ECG signal, in part, to perform
their primary function. Examples of these devices are treadmill systems where the heart is monitored under exercise
workload conditions; cardiac pacemakers that monitor the
heart rhythm from internally implanted electrodes to determine if it is necessary to stimulate the heart because of loss of
function of the hearts natural pacemaker; and sophisticated
imaging systems that require synchronization with the cardiac cycle to minimize the effects of cardiac motion. Thus, the
ECG is still evolving as a tool for studying the heart even
though it is perhaps the oldest test instrument in medicine.
When two wires are placed anywhere on the body surface and
then attached to the inputs of a bioelectric amplifier, it is possible to record the voltage generated by the heart. There are
standard positions for placing the recording electrodes on the
body, but generally the potential difference measured between any two recording sites is the summation of electrical
signals generated by billions of cardiac cells. The adult heart
is a bit larger than a fist, and the sequence of its electrical
activation is directly related to the contractional sequence of
the various heart chambers. It is important to note that the
electrical signals are the triggering event for the mechanical
motion of the heart and that these electrical events precede
the heart contraction. The electromechanical coupling is a significant phenomenon for the overall function of a healthy
heart, but it is possible to have an electrically normal heart
while the mechanical function could be significantly impaired
and vice versa. This article focuses on the electrical activity
of the normal heart, for which it is important to understand
some fundamental aspects of the hearts anatomy and physiology (1).
There are four chambers in the heart. The two upper
chambers are the atria, and the two lower chambers are the
ventricles. Another way to divide the heart is into the right
and left side with the result that the four chambers are the
right atrium, right ventricle, left atrium, and left ventricle.
Figure 1 is a schematic representation of the four chambered
heart with its physical connections to the veins which deliver
blood into the heart and the arteries, the vessels that carry
blood away from the heart. One could begin anywhere in the
To the body
To the
lungs
From the
lungs
From the
body
Left atrium
Right
atrium
Left ventricle
Right ventricle
Figure 1. A cutaway diagram of the heart showing the major chambers and vessels. The blood flow into and out of the heart is indicated
by the arrows.
ELECTROCARDIOGRAPHY
Depolarization
(Phase 0)
Early repolarization
(Phase 1)
Plateau (Phase 2)
0.0 V
Repolarization
(Phase 3)
Resting
state
(Phase 4)
90 mV
300 ms
Figure 2. A cardiac action potential recorded with a microelectrode
inside of a single cardiac cell. There are five phases describing the
three electrical states of each cell.
circulatory system to describe the sequence of events that

make up the cardiac cycle, but the most common starting
point is the right atrium. This approach takes the view that
there are two pumps in series, each with a low-pressure collecting chamber, the atria, and each with a high-pressure
pumping chamber, the ventricles. In fact the atria contract
sequentially whereas the ventricles contract together. This
parallel view of the cardiac cycle is examined when discussing
electrical activation of the heart.
The right atrium collects blood from all of the veins in the
body except those of the lung. When the right atrium is triggered to contract, it forces blood into the right ventricle. As
the right ventricle fills, it contracts and forces blood to the
lungs where the blood exchanges the excess carbon dioxide,
an end product of metabolism, for more oxygen, a necessary
metabolic component. The pulmonary veins return this oxygen-enriched blood to the left atrium that in turn empties into
the left ventricle. The left ventricle is a high-pressure pump
that forces blood to all of the body organs and tissues except
of course the lungs. This is done through the arterial blood
vessels which evolve to microscopic tubes called capillaries
which then reform into larger vessels forming the venous return system to the heart.
Before discussing the electrical activation sequence of the
heart, it is necessary to understand some of the basic concepts
of cellular electrophysiology. A complicated cellular process of
ion flow across the cell membrane occurs with each heartbeat.
It is possible to monitor these currents and their resulting
transmembrane potentials when cardiac tissues are studied
in vitro, or outside the body, and placed in a warmed, nourishing solution. Such studies form the basis of almost all current
knowledge of cardiac electricity, but a detailed discussion of
this work is well beyond the scope of this report. The article
BIOELECTRIC PHENOMENA in this encyclopedia covers this material. Figure 2 is a recording of a transmembranous action potential from a microelectrode that can actually be impaled
into a single cardiac cell. It shows how the voltage from the
inside of a cardiac cell varies with respect to the outside of
the cell and how it varies during the time of cardiac activation. Cardiac cells have three electrical states that comprise
the cardiac cycle, the resting state, depolarization, and repolarization. Scientists who study the cardiac action potential
refer to the resting state as Phase 4. The electrical states and
phases of the action potential are labeled along the waveform
371
at points representing their respective occurrence. Hence the

resting state and Phase 4 are synonymous. The resting state
is the time between beats and, depending on the heart rate,
is the most common state of the heart. For example, cardiac
action potentials range in duration from 200 ms to 350 ms, so
if the heart rate is 60 beats per minute (1 bps), then the resting state accounts for 65% to 80% of an individual cells electrical activity. During the resting state the inside of the cell
has a negative potential of about 90 mV with respect to the
outside of the cell, which is assumed to be at 0.0 V.
As an aside to this discussion it is interesting to note that
this potential difference occurs across a very small distance,
the thickness of the cell membrane. This distance is on the
order of 1 nm. This results in a field strength of (90 mV)/
(1 nm) 90 MV/m. This is a very large electric field considering that air conducts electricity, for example, static spark or
lighting, at field strengths of about 300 kV/m. Maintenance
of this large electric field across the cell membrane is most
likely due to the lipid (fat) bilayer which forms the bulk of the
cell membrane.
Activation of a cardiac cell occurs when ionic currents flow
from neighboring cells or from a nonphysiological stimulus
such as a mechanical impulse or a pacemaker battery pulse.
The membrane potential of a single cell can shift very rapidly
from 90 mV to 20 mV (a total swing of 110 mV) in approximately 1 ms. This rapid change is called depolarization or
Phase 0 of the cardiac action potential. This rapid change in
membrane potential causes neighboring cells to depolarize,
and thus the cardiac impulse spreads throughout the myocardium from cell to cell. Although the stimulus for initiating
depolarization may vary, a key factor required for depolarization is changing the resting membrane potential to a specific
threshold voltage. Once the membranous potential reaches
this threshold, the individual cellular dynamics take over,
and now all of the following phases are independent of the
original stimulus. The speed at which cell-to-cell propagation
spreads across and through heart tissues depends on the particular cell type and is discussed later.
Immediately following the depolarization state is the return to the resting state, called the repolarization state. This
process has three well-defined phases. Phase 1 is an early,
rapid, low-amplitude decline in potential. Phase 2 is a relatively isopotential period called the plateau, and Phase 3 is
the return to the resting state.
All of the cellular events previously described imply that
each cell acts as a tiny current source with a resulting potential field. The current spreads throughout the body because
all of the tissues enclosed within the skin form what is called
a volume conductor and, for the most part, these tissues are
good conductors of electricity. Thus, what is measured on the
body surface or from within the body for that matter is a
summed view of all of the individual potential fields generated by each individual cell.
The hearts primary function is to pump blood. This mechanical process is triggered by an electrical signal generated
by each cell once it is stimulated. Cardiac muscle cells are
bricklike in appearance. They have a base which is about
10 m on each side and are about 100 m long. Rather than
viewing activation and contraction cell-by-cell, it is often simpler and more convenient at times to consider these electrical
and mechanical processes on a more macro or tissue level.
Hence one could refer to atrial activation or ventricular con-
372
ELECTROCARDIOGRAPHY
1.0 s
SA node
SA node
240
Right atrium
50
Left atrium
Right
atrium
AV node
His bundle
Left
bundle
Right
Left
bundle
ventricle
Left atrium
AV node
50
50
Right
ventricle
His bundle
15
Purkinje fibers
Figure 3. A schematic diagram of the cardiac conduction system.
The major components of the conduction system are shaded and
shown in their approximate anatomic position, but they are not to
scale.
Right and left

bundle branches
30
Right and left

ventricles
80
R
T
traction. This integrated view of the hearts mechanical function is called a syncytium (1).
Now let us consider the heart on a more anatomical basis
and describe the actual sequence of activation and how this
relates to generating the measured waves of the ECG. Figure
3 is a cross-sectional sketch of the heart. All four chambers
are respectively labeled. A network of structures has been
added that comprise the hearts specialized conduction system. Consider Figs. 3 and 4 together to understand the sequence of cardiac activation better. Figure 4 is a timing diagram that indicates the length of time during which each
particular structure has cells undergoing phase zero depolarization. The bottom trace in Fig. 4 is a stylized ECG recording
whose component waves are labeled allowing one to compare
the surface waves with the internal cardiac sources. Note that
not all of the internal structures are observed on the standard ECG.
The electrical activation of the heart begins with automatic
depolarization of an irregular mass of cells in the upper portion of the right atrium called the sinoatrial (SA) node. Once
these cells end their repolarization, there is a gradual incline
of Phase 4 toward the threshold voltage. Thus these cells do
not require an initial excitative current from any other cell,
and they comprise the hearts natural pacemaker. The right
atrial muscle cells respond to these neighboring and depolarizing SA nodal cells by initiating their own depolarizing currents. Notice that there is considerable timing overlap be-
ECG
Q
Figure 4. A timing diagram showing the activation times for each of

the major muscle chambers and the conduction system. Note that the
durations of the positive pulses (labeled in milliseconds under each
pulse) indicate the time during which cells in each structure undergo
depolarization. The bottom trace is a stylized ECG showing the correspondence of atrial and ventricular depolarization with the surface
manifestations of the P-wave and QRS-complex, respectively.
tween the SA node and the right atrium because the SA node
does not have a single point of interconnection with the right
atrium. The speed at which the cardiac impulse travels is relatively slow in the SA node, on the order of 0.05 m/s.
Conduction velocity plays an important role in cardiac activation. It is easily measured by placing two electrodes with
known spacing on the heart. The occurrence of depolarizing
voltages is easily identified at each site. Together with the
time measured between these two events, the conduction velocity is determined. Conduction velocity is the ratio of the
time between the two events divided by the distance between
the two recording sites. Table 1 lists conduction velocities for
the various cardiac tissues. Note also that physiologists refer
to conduction velocity as the measure of speed of activation
through electrically active tissues. In this sense conduction
Table 1. Conduction Velocity of Cardiac Tissues a

SA
Node
Atrial
Muscle
AV
Node
His
Bundle
Bundle
Branches
Purkinje
Fibers
Ventricular
Muscle
0.05
0.5
0.05
1.0
1.02.0
4.0
0.5
All values are given in m/s.
ELECTROCARDIOGRAPHY
velocity is a scalar measurement. In addition the bricklike

shape of the cardiac cells and their regular pattern of end-toend construction with staggered rows on top of each other also
affect the speed of activation. In the ventricles, for example,
the speed of conduction is three to four times faster in the
direction of the long axis of the cells than across their short
axis. The reason for this is based on the number of interconnections which exist at cell ends rather than at the sides of
the cells. This directionally dependent property is called anisotropy. Hence, if one artificially stimulates the ventricles at
a single site, the patterns of activation away from this site are
generally in concentric ellipses rather than concentric circles.
Atrial activation spreads from right to left and from top to
bottom. The timing diagram shows that the duration of SA
nodal activation is about 240 ms. Some of the right atrial
muscle begins activation during the period of SA nodal activation. The right atrium begins activation about 50 ms prior to
the left atrium. Total atrial activation takes about 100 ms
with a conduction velocity of about 0.5 m/s. As seen in the
bottom trace of Fig. 4, the ECG registers the P-wave during
atrial depolarization. When the activation wave front reaches
the lower floor of the right atrium, another nodal structure
called the atrioventricular (AV) node is activated. This is also
a slowly conducting structure (0.05 m/s) and is considered the
initial conduit of activation from the atria to the ventricles.
Although the atria and ventricles border each other, the only
electrical connection between them is through the AV node.
The relatively slow conduction through the AV node allows
time for the atria to contract and fill the ventricles. It takes
about 50 ms to 70 ms for activation to traverse the AV node
as indicated by the timing diagram. The AV node connects to
a small fiber, about 1.0 mm in diameter and about 10.0 mm
to 15.0 mm long, which actually leaves the right atrium and
penetrates to the top of the ventricles. This fiber is called the
common bundle or the bundle of His, named after Wilhelm
His, a German physiologist, and his early work on conduction
system anatomy, although he was not the first person to actually identify this structure. Conduction velocity accelerates
through the His bundle to about 1.0 m/s, and it has a total
activation time of about 10 ms to 15 ms. At this point the
electrical impulse is traveling atop the wall or septum that
separates the left and right ventricles. The His bundle and its
early ramifications are not electrically connected to the ventricular muscle tissue. These ramifications separate into bundles on the right and left sides of the septum and are appropriately called the right and left bundle branches. Conduction
velocity accelerates to about 2.0 m/s and lasts about 30 ms to
35 ms in these bundles. A further arborization of this conduction system tissue occurs over the inner surface of the ventricles, called the endocardium. The Purkinje fibers are the final
stage of the conduction system and form a fine fiber mesh
covering over some parts of the endocardium. This widespread, thin layer of tissue is where the conduction system
and ventricular muscle tissue finally form electrical interconnections. A large part of the endocardium is initially activated
by the Purkinje network. This is followed by conduction from
muscle cell to muscle cell which results in total electrical activation of the ventricles. The conduction velocity through the
ventricular muscle is about 0.5 m/s, and it takes about 80 ms
to 100 ms to depolarize the ventricular myocardium. The timing diagram shows that the two ventricles are activated at
373
about the same time and have roughly the same duration of
activation.
A sample ECG tracing is shown at the bottom of the timing
diagram. The first rounded wave, called the P-wave, is the
body surfaces manifestation of atrial depolarization. Toward
the end of the P-wave the AV node and HisPurkinje system
depolarize. The standard ECG does not show any evidence
that these structures depolarize because they produce very
small signals compared with the large muscle masses of the
atria and the ventricles. A specialized branch of electrocardiography, called high-resolution electrocardiography (2), uses
computer-based enhancement techniques to record these lowlevel signals from the body surface, which are described later
in this article. Once the ventricles are depolarized, the ECG
shows a rapidly changing voltage called the QRS-complex.
The initial downstroke of this complex is called the Q-wave,
and the initial upstroke is called the R-wave. The final downstroke is called an S-wave. The nature of the QRS-complex
depends highly on the specific lead and any underlying cardiac abnormalities. Thus, there are times when the ventricular depolarization wave may have only a RS-pattern, or may
be just a deep Q-wave. Generally, the complex is called a
QRS-complex even if the pattern does not strictly follow the
QRS-sequence. Following the QRS-complex is a smooth
wave called the T-wave. This wave represents the repolarizing currents of the ventricles. Notice that the atria do not
produce a similar repolarizing waveform. The atrial repolarization wave temporally overlaps with the QRS-complex and
is masked by its higher voltage.
Figure 5 is a larger, stylized ECG recording showing both
amplitude and timing scales. In ECG terminology an interval is the period from the beginning of one wave to the beginning of the next wave. An example is the PR-interval shown
in this figure. A segment is the period between two waves
as demonstrated by the ST-segment in this figure. Finally
this tracing shows the U-wave, a small wave after the Twave. The U-wave is associated with repolarization, but its
actual origin has yet to be definitively determined.
Measurements of the amplitude and duration of each wave
and the above-mentioned intervals and segments are used by
electrocardiographers to diagnose cardiac pathologies such as
ventricular enlargement (hypertrophy), blocks of conduction
in the bundle branches, damage to the ventricles due to heart
attacks, and abnormalities of rhythm. Such diagnostic interpretation of the ECG is an important skill, and there are a
number of texts devoted to the subject (3).
200
ms
1.0
mV
P
PRinterval Q
S
T
U
ST-segment
Figure 5. A stylized ECG recording showing the individual waves

(P, QRS, T, U) and the approximate time and voltage scales. Both the
PR-interval and ST-segment are also labeled.
374
ELECTROCARDIOGRAPHY
History
The electrocardiogram was first recorded by Augustus D.
Waller, an English physiologist, in 1887. His pet bulldog, Jimmie, was his first subject using a device called the capillary
electrometer. The device, crude by todays standards, used a
voltage-sensitive column of mercury that reflected a beam of
light from its meniscus onto a moving photographic plate.
Waller is credited by many to have actually coined the term
electrocardiogram. However, Willem Einthoven, a Dutch
physiologist, is usually credited with bringing the ECG into
clinical practice with a string galvanometer. This device used
a thin wire between poles of a magnet. A movement of the
wire occurred which was proportional to the current flow. Motion of the wire could be used to scrape a carbon residue off a
slowly rotating smoked drum. The evolution of ECG recording instrumentation closely followed the developments of
electronic technologies, such as the vacuum tube, the transistor, the integrated circuit, and the microprocessor. The use of
computers to automate the interpretation of the ECG was a
very early application of computers in medicine (4). Since the
1960s the algorithms for a fully automatically interpreted
ECG have been developed and optimized to the point where
physician overreading, although still a technical requirement,
is seldom necessary for normal ECGs. Complex arrhythmias
have proven difficult for automated interpretation. It usually
requires a highly trained physician to render an accurate
reading. Systems in use today often resemble a fully functioning computer system with very specialized software. Figure 6
shows a block diagram of a modern ECG system in which

embedded microprocessors are used to digitize and analyze
the 12-lead ECG. In addition, microprocessors control the digital input/output systems which include a graphics screen,
full keyboard, strip chart recorder, and floppy disk drive.
Theoretical Basis
As ions cross the cell membranes, a tiny current source exists
inside the body. The body generally consists of fluids, muscles, fat, and other organ tissues which all act as good conductors of electricity. The body is bounded by a skin interface and
hence is referred to as a volume conductor. Theoretically, the
potential field established by the flow of ionic currents in the
body tissues appears at all points within and on the surface
of the body. Because of the large differences in the resistances
between the environment and the body tissues these voltage
potential fields are not found beyond the body surface. Hence,
there is no current flow from the body to free space, or put
another way our bodies do not supply electrical energy to the
outside world.
For signal frequencies in the range of the ECG (0.05 Hz to
200 Hz) the body is considered purely resistive and does not
contain inductive or capacitive elements (5,6). Thus, as the
heart depolarizes, events on the body surface, which reflect
these cardiac sources, appear instantaneously. There is no
time delay between changes in the source and the appearance
of these events on the body surface. This instantaneous propagation from the heart to the body surface should not be con-
J2
Printer
control
J6
Transformer
and
rectifier
J7
J8
Power
supply
J3
J5
Printer
(printhead and
paper transport
CPU
assembly
J15
DRAM
Disk
control
J14
Disk
drive
J4
Battery
J10
Serial
comm
Modem or
direct
connect
CPU
J1
I/O
control
Keyboard
and LCD
display
J9
ROM
Figure 6. A block diagram of a modern
computer-based ECG recorder. Such a device will amplify, digitize, and analyze the
ECG and has all the features of a modern
embedded microprocessor-based instrument (courtesy of Hewlett-Packard Company, Palo Alto, CA).
Patient
module
DSP
J12
Expansion
bus
Preview display
and power
supply
ELECTROCARDIOGRAPHY
fused with the measurable conduction velocities within the

heart (see Table 1).
There are literally billions of individual cells in the heart
that depolarize during the cardiac cycle. The sequence of depolarization from the various structures within the heart was
schematized in Fig. 4. As one proceeds down to the cellular
level, there are only groups of cells depolarizing at any given
moment. It is possible to represent those groups of cells which
are simultaneously depolarizing as an equivalent source. A
common representation of the equivalent cardiac source is
that of a dipole with a time-varying magnitude, orientation,
and position within the body. This representation is a firstorder model useful for understanding the generation of the
main waves of the ECG recorded from a bipolar lead with
electrodes on the body surface.
The equivalent heart source can be expressed as a threedimensional heart vector H in a Cartesian coordinate system. When two electrodes are placed on the body surface, another lead vector L is formed by connecting the two electrodes. The voltage recorded between two electrodes from the
cardiac source vector is directly proportional to the dot product, H L H L cos(). More specifically, the voltage between the two electrodes is given by the component of H in
the direction of L. This concept is referred to as the lead field
theory. To visualize the lead field concept better, we can use
the principle of reciprocity. Briefly, this old network theorem
states that when a voltage source at one location within a
circuit produces a current between two nodes in the circuit,
then one can inject the same measured current in these two
nodes to duplicate the original voltage at its original location.
Figure 7 demonstrates how this can be applied to the ECG.
The top panel shows an outline of a torso with a drawing of
the heart in its approximate anatomical position. The two
dots represent two body surface electrodes with current flowing between them. The lines crossing through the torso and
heart represent the flow of current or the reciprocal lead
field. Note that the heart is in the densest region of the current lines. Components of the heart vector lying along the
lead axis, i.e., parallel to the lead field, produce the largest
relative voltages. In the lower panel the recording electrodes
are placed along the right side of the torso. The current lines
do not substantially cross through the heart, and hence this
lead axis, i.e., parallel to the lead field, would be a poor one
for recording the ECG. This concept of reciprocally energizing
the recording leads and mentally visualizing the extent of the
current lines which pass through the heart is useful in understanding the lead field concept and determining whether two
electrodes placed on various body sites will record a large amplitude ECG.
METHODOLOGY
The Recording Technology
The modern ECG system still relies on attaching a set of
wires to the skin to couple the potentials on the body surface
directly to electronic amplification systems. The electrodes
that contact the skin are usually made of a silver/silver chloride (Ag/AgCl) electrode. A conductive electrode jelly or paste
acts as the interface between the skin and the metal. This
electrodetissue interface relies on establishing a stable
chemical reaction between the ionic charge carriers in the
375
Figure 7. Each panel has an outline of a male torso with the diagram of the heart positioned approximately in its proper place within
the chest. The solid dots represent electrodes on the body surface.
The lines represent current flow between the electrodes and demonstrate the lead field concept described in the text.
body and the electron charge carrier in the metal electrode

(7). The Ag/AgCl electrode is preferred because it is a nonpolarizing electrode through which current freely passes. Generally, this chemical reaction stabilizes in a minute or so and
does not interfere or alter the nature of the electrical signal
from the heart. There are times, however, when the chemical
reaction does not readily equilibrate, for example, when the
subject is ambulatory, resulting in a recording susceptible to
artifact.
Then the electrode wires are directed to the inputs of a
special differential amplifier used to record bioelectric events.
These bioelectric amplifiers must meet a number of technical
requirements to record the millivolt level ECG signals and to
ensure safety when connected to a human being. In general,
the bioelectric amplifier has differential inputs with a high
input impedance (100 M) and a bandwidth between 0.05
Hz and 150 Hz. There are two sets of standards which most
manufacturers rely on for the ECG. They are published by
the American Heart Association (8) and the Association for
the Advancement of Medical Instrumentation (9).
376
ELECTROCARDIOGRAPHY
Digitizing the ECG signal is relatively straightforward

with a standard analog-to-digital converter (ADC). The dynamic range of the ADC is the voltage range over which the
analog input voltage is converted to a binary number. Hence
the ADC may have an input range of 1.0 V. Another figure
of merit for the ADC is the number of bits in the converted
binary digit. Most commercial ECG systems use a 16-bit converter which has a dynamic range of 216 or about 96 dB. This
is a very high dynamic range for the typical ECG signal, but
it does allow the ECG system to accommodate a widely varying baseline drift which can occur when the tissueelectrode
interface is not well established. This artifact, known as baseline wander, can be problematic when interpreting the ECG.
There are several digital methods which correct for the baseline wander, and the large dynamic range of the ADC allows
these algorithms to operate without the amplified ECG signal
reaching the limits of the 1.0 V ADC range.
Automated Measurements and Analysis
Once the ECG signals are digitized, there are many forms of
measurement and analysis that are automatically performed
to aid the medical professional in interpreting the clinical information contained in the ECG. The section on applications,
covers several of these, but one of the most common functions
performed in each application is detecting each beat (10). In
this case it is assumed that each beat means every ventricular contraction or every QRS-complex. There are times when
atrial and ventricular activity are not synchronous and automated analysis requires detecting both atrial (P-wave) and
ventricular activity (QRS-complex). Generally the first step in
automated analysis is detecting each QRS-complex. It is the
most prominent deflection of the ECG has the largest amplitude (1.0 mV), and the most rapid change of potential. This
rapid change in potential can be detected by taking the derivative (dV/dt) of the ECG and searching for the largest value
of the derivative. Of particular concern in this approach is
that noise which may contaminate the ECG recording also
produces large derivative values, but the noise is not usually
larger in overall amplitude than the QRS-complex. Once each
possible beat is detected by searching for the largest first derivative, other algorithms can be used to examine the shape
of the QRS-complexes and to classify them as normal or abnormal. By measuring intervals, amplitudes, and other wave
characteristics, a number of ECG applications can be automated. The following section describes several of these applications where computer based algorithms are used to replace
a human operator, and also to create new forms of analysis
not amenable to human measurement.
APPLICATIONS
Five ECG applications are presented in this section. In most
cases, these instruments are the result of older technology
which has matured with the evolution of computer-based instrumentation: 12-lead ECG, monitored ECG, stress ECG,
high-resolution ECG, and intracardiac ECG.
The 12-Lead ECG
Einthovin demonstrated the clinical value of the ECG and
some of its theoretical underpinnings using the three bipolar
limb leads: I, II, and III. These leads are defined as follows:
I = VLA VRA
II = VLL VRA
III = VLL VLA
where the terms VLA, VRA, VLL represent the voltages recorded
at the left arm, right arm, and left leg, respectively. Note that
since each measurement is the difference between two voltages sharing a common reference, the choice of the reference
location is arbitrary, and its symbol disappears in the algebra. This is shown in the top panel of Fig. 8. The lines connecting the three limbs define a triangle known as Einthovens triangle, and it demonstrates Einthovens law:
III = II I
In 1934, Frank N. Wilson (11), an American physician, introduced a concept whereby the signals from the two arms
and left leg were averaged by connecting them together with
a set of equal resistors, as shown in the lower panel of Fig.
8. This common terminal was used as a reference for other
electrodes attached to the body. Thus, new leads were formed
using what is now commonly called the Wilson central terminal or WCT. The voltage at the WCT is defined as
VWCT = (VLA + VRA + VLL )/3
II
III
I = VLA VRA
II = VLL VRA
III = VLL VLA
III = II I
2VLA VRA VLL

2
2VRA VLA VLL
aVR =
2
2VLL VLA VRA
aVF =
2
aVL =
Augmented
leads
vi
Vi = vi VWCT
i = 1 to 6
VWCT
Figure 8. The electrode positions for the 12-lead ECG are demonstrated. The top panel shows the formation of Einthovens triangle
from leads I, II, and III. The voltages on the limbs are from the right
arm (VRA), the left arm (VLA), and the left foot (VRA). The middle panel
shows the formation of the augmented leads which are linear combinations of the limb leads. The bottom panel shows the Wilson central
terminal (WCT) formed by averaging the limb voltages through an
equal value set of resistors. The WCT is the reference for the chest
leads (V1, V2 . . . V6).
ELECTROCARDIOGRAPHY
Initially, Wilson proposed a new set of leads abbreviated VR,

VL, and VF formed by the pairings of the right arm and the
WCT, the left arm and the WCT, and the left foot and the
WCT, respectively. Thus the three additional leads are given
as follows:
VR = VRA VWCT = VRA (VLA + VRA + VLL )/3

VL = VLA VWCT = VLA (VLA + VRA + VLL )/3
VF = VLL VWCT = VLL (VLA + VRA + VLL )/3
E. Goldberger (12), an American physician, recognized that
each term had a duplicated voltage and defined the modified
WCT where only the two nonduplicated limb voltages were
averaged. These augmented leads were identical in appearance to those originally defined by Wilson but were 33%
larger in amplitude. In this era, such an increase in amplitude was significant considering the quality of the amplifiers
used to record the ECG. The augmented leads are defined as
follows:
aVR = (2VRA VLA + VLL )/2

aVL = (2VLA VRA + VLL )/2
aVF = (2VLL VLA + VRA )/2
Similarly a set of electrodes is attached across the front
and left side of the chest and recorded with respect to the
WCT. The approximate placement of these chest electrodes
shown in the lower panel of Fig. 8 are called V1 through V6.
The standard 12-lead ECG consists of the following leads: I,
II, III, aVR, aVL, aVF, V1, V2, V3, V4, V5, V6. These leads are
often formatted in the fashion shown in Fig. 9 where a 2.5s
window of each lead is displayed and a 10s period of a single
lead for rhythm analysis (bottom trace). Note that the time
and voltage scales are 25 mm/s and 1.0 mV/cm, respectively.
These scales are relatively low in resolution but have been in
use for about 75 years. The 12-lead ECG forms the traditional
diagnostic ECG and although a relatively low tech approach
is used for display to the physician, the large empirical data
base used by physicians will most likely rely on this format
for generations to come.
The text in the upper portion of Fig. 9 includes information
about the fictitious patient, a set of basic measurements, such
as heart rate and interval measurements, and a set of diagnostic statements generated by the interpretive program. The
automation of ECG interpretation is usually divided into two
phases. In the first phase a set of algorithms is used to obtain
a multitude of wave amplitudes, duration, and beat-to-beat
measures. Figure 10, published many years ago by Hewlett
Packard, demonstrates that each beat can generate over 20 of
these types of measurements. Most of these parameters are
derived from the morphology of the waves and most do not
relate to an actual physiological event. For example, the durations of the individual components of the QRS-complex, for
example, QD, RD, SD, are not used by physicians when reading
ECGs. However, such measurements may provide the software algorithms with values which aid in discriminating various abnormal QRS-complexes.
The second phase of the interpretation is generating diagnostic statements, samples of which are shown in the top portion of Fig. 9. For the most part these phrases are given the
most attention by the physicians. They are usually rule-based
377
outputs derived from the measurement matrix. In almost all

cases these diagnostic statements are overread by the physician and changed if necessary.
The automated process is performed on almost all ECGs
obtained in clinics and hospitals. Perhaps one of the greatest
advantages of these automated systems is the digital ECG
database. These systems are considered ECG management
systems which allow for a tremendous saving in space needed
for ECG storage, instantaneous recall of older recordings, and
comparing serial changes in the ECG to chart the progress of
some diseases.
The Monitored ECG
There are two primary applications where a patients ECG is
continuously monitored. Intensive care units within a hospital often monitor the ECG of critically ill patients to observe
the patients rate and rhythm. When the patient is suspected
of having a life-threatening arrhythmia, it is best to monitor
the patient in an environment where a rapid response and
therapeutic intervention can be lifesaving. The other application of the monitored ECG is in the ambulatory, outpatient
setting. The patients ECG can be monitored by a belt-worn
device. In the 1950s Norman J. Holter (13), an American
physicist, demonstrated that the ECG could be monitored
while the subject was physically active. However, the technology of the day resulted in a very heavy backpack device
weighing 85 lb and was impractical for routine use. Recording
devices and their associated electronics and batteries eventually became small enough to allow for a belt-worn tape recorder using originally reel-to-reel technology, but now relying exclusively on cassettes. These tape devices still have
many inherent limitation, such as poor noise figures, low dynamic range, and limited frequency response. If one is merely
recording the patients rate and rhythm, then the tape technology is adequate. However, high-resolution ECGs might
also be useful when obtained from the ambulatory patient,
and the tape technology is definitely limited for this application.
Newer digital recorders are currently available whereby
the ECG is digitized and stored in either high density memory chips or on actual hard disk drives. Current versions of
the latter have removable drives with capacities exceeding
500 Mbytes. Depending on the application, the ECG may be
sampled between 250 Hz and 1000 Hz. Originally, only one
ECG lead was recorded on tape-based systems, but the new
systems are not limited by the poor frequency response of
tape systems or the physical size constraints of magnetic recording heads. With digital systems the number of simultaneous (or near simultaneous) recordings is not particularly limited, but three or four ECG leads are a practical number.
Electrode positions for these monitored leads do not follow the
conventions of the 12-lead ECG and are often similar to the
bipolar limb leads, where the electrodes are placed a few
inches apart over the chest, creating several lead fields
through the heart.
For hospital monitoring, where the patient is being evaluated for a critical cardiac condition, only a few leads are recorded, but a full 12-lead ECG is periodically recorded. In
some cases the patients, although not acutely ill, are given
the freedom to walk about the hospital and their ECG is telemetered via radio frequencies by an antenna/receiver net-
378
ELECTROCARDIOGRAPHY
work. In such cases the goal is to monitor the patients

rhythm in normal activities.
The massive amount of ECG information obtained during
continuous monitoring is overwhelming. In the hospital approach, the ECG signals are usually fed to a large system of
monitor screens where specially trained technicians view the
actual recording of 10 to 50 patients. In conjunction with computer-based software the high risk situations are quickly
identified with appropriate communication to the medical
staff, for example, code blue. This is not the case with outpatient monitors where the patient returns to the hospital 1 or
2 days later. Then the entire record is scanned with an interactive software analysis program. Often just a compressed
printout of the continuous ECG can be quickly inspected for
an abnormal rhythm. An example of this full disclosure
mode is shown in Fig. 11. A 7.5 min recording is shown in
this format. Note that the first several minutes have a normal
rhythm (there are several other abnormalities in this tracing,
but they are beyond the scope of this article). The abnormal
beats begin to appear in groupings of two or three. A condi-
tion known as nonsustained ventricular tachycardia appears

in the fourth trace from the bottom.
The event recorder is an extension to ambulatory recordings. In this case the patient wears a recorder for many
days or even weeks. When the patients experience a symptom, such as chest palpitations or dizziness, they push an
event button on the recorder which causes the recorder to
save 1 min to 2 min of data before and after the event. The
patient can call the physician office and transmit these data
via a modem for rapid interpretation. The most advanced version of the event recorder is an implantable device that monitors the ECG for months or years and uses special monitoring
software to record suspicious events without patient activation. This type of unit can be interrogated at regular intervals
over the phone or during regular visits.
The Exercise ECG
The 12-lead ECG is obtained while the patient is resting. The
monitored ECG is obtained either during an evolving disease
Figure 9. An example of a 12-lead ECG in standard format with computer-based measurements

and diagnostic statements (courtesy of Hewlett-Packard Company, Palo Alto, CA).
ELECTROCARDIOGRAPHY
379
QRSt
VAT
QRS
area
P area
P area
QRS notch
P area
T area
ST
shape
Type 3
Type 2
Tarea
Type 1
RA
ST
P
PA
PD
PA
STE
QD
T
TA
TA
STON
QA
P
PD PR
Q
Segment
PR-interval
SA
RD
SD
QRSP1
STM
80MS
TD
TD
STSlope
J point
S-interval
Figure 10. A stylized ECG showing the many amplitude, duration, and area measurements
used to develop the measurement matrix for automated ECG interpretation (courtesy of HewlettPackard Company, Palo Alto, CA).
process or while the patient is undergoing routine activities.

Evaluation of cardiac performance is done with a number of
diagnostic tools, but an exercise stress test monitors the ECG
prior to an exercise protocol (usually on a treadmill with adjustable speed and elevation) during the exercise period and
finally during a warm-down period. Of particular interest is
achieving a target heart rate which is usually based on the
patients age. The primary objective of the stress test is to
monitor the ST-segment for small voltage changes on the order of at least 100 V. This is a rather difficult measurement
because of the amount of motion artifact resulting from the
very active patient on the treadmill. Great attention must be
paid to stable electrode placement on the chest. A number of
computer algorithms can reduce the noise.
The High-Resolution ECG
Computer processing of the ECG has its origins in aiding or
replacing the physician in making tedious measurements.
Elucidating the standard waves (P-wave, QRS-complex, Twave) was the primary aim. Once in the digital domain, a
number of digital signal processing techniques are performed
on the ECG, particularly in the realm of noise reduction and
signal enhancement. This allows visualizing very low level
signals (1.0 V) from such sources as the bundle of His and
the left and right bundle branches. These structures are depolarized during the PR-segment (see Fig. 4) when no other cardiac signals are considered present. In fact, in traditional
ECG theory the PR-segment is considered an isoelectric interval and is often used as a 0.0 V reference for the other waves.
The initial computer-based approaches to record such a highresolution ECG began in the early 1970s (14). Up to that time
the HisPurkinje signals were recordable only by placing
electrode catheters inside the heart and in very close proximity to the respective structures. The noninvasive recording of
HisPurkinje signals was the advent of a new generation of
ECG analysis. As the techniques for high-resolution ECG
evolved perhaps the most clinically significant application
was in recording so-called cardiac late potentials.
Cardiac late potentials typically arise from ventricular
cells which surround a dead region of the heart caused by a
heart attack. These bordering regions with surviving cells appear on the outer edge of the scar tissue and also permeate
into the scarred region. It is possible that complete but circuitous pathways of viable cells can actually traverse the scar
tissue. Such a matrix is often the site where life-threatening
arrhythmias originate and are sustained. During normal
heart rates partial activation of these arrhythmia pathways
have been revealed, originally with electrodes in direct cardiac contact and then eventually by using high-resolution
ECG techniques similar to those used to record HisPurkinje
signals noninvasively (15). The activation of border zone cells
is often delayed past activation of the normal ventricular cells
because they are poorly conducting due to the heart attack.
The artificially long pathways resulting from the mix of dead
and surviving cells within and surrounding the infarct can
also result in depolarizing signals which outlast the end of
normal activation. These signals are not part of normal cardiac activation. The use of computer-based enhancement techniques has been the only way to identify and quantify them.
It has been shown in hundreds of studies that the presence of
these late potentials, after patients have had heart attacks,
indicates that they have a very high risk of future life-threatening arrhythmias.
The primary method used to record both HisPurkinje and
late potential signals is achieved by means of signal averag-
380
ELECTROCARDIOGRAPHY
03:37:00
03:37:30
03:38:00
03:38:30
03:39:00
03:39:30
03:40:00
03:40:30
03:41:00
03:41:30
03:42:00
03:42:30
03:43:00
03:43:30
03:44:00
Figure 11. A portion of a full disclosure ECG from an ambulatory ECG. This mode of presentation, although significantly condensed, allows the trained reader to assess rhythm alterations
rapidly. This is a very abnormal recording.
ing (16). One assumes that the signal of interest, which is

very low level, repeats on a beat-to-beat basis. Also, the interfering noise (usually the signals associated with chest wall
muscles depolarizing during breathing) is not temporally
linked to the signal of interest and is random. Once digitized,
the QRS-complex is detected very precisely so that the computer can finely align a window of data surrounding the QRScomplex and perform a point-by-point addition of each incoming beat. This averaging process is similar to one used by any
scientist making a physical measurement. It is common to
make several such measurements and to average them to get
closer to the true value. The mathematics of signal averaging
shows that under ideal conditions the signal-to-noise ratio increases by the square root of the number of measurements
(beats). Thus if 100 beats are averaged, the noise decreases
by a factor of 10.
Figure 12 best demonstrates the signal-averaged ECG using an example of cardiac late potentials. Figure 12(a) is an
example of three ECG leads obtained from anatomically orthogonal leads called X, Y, and Z. These are bipolar leads,
like leads I, II, and III, but the electrodes are placed on the
chest along a set Cartesian axes with the heart at the origin.
The voltage and timescale used in panel (a) are close to those
used in the standard ECG. Panel (b) is a 300 ms window of
the ECG approximately centered on the QRS-complex. The
signal amplitude is five times greater than that used in panel
(a). This single cardiac complex, however, is the result of av-
eraging 200 cardiac cycles. One can actually observe small

undulations at the end of and after the QRS-complex. These
are late potentials, but they are still difficult to observe and
quantify. One method used to elucidate overlapping signal
components is to use a selective filter which reduces or increases the amplitude of certain spectral components of the
signal (17). The bass and treble controls of a stereo amplifier
are familiar to most as a means spectrally manipulating a
signal. In the case of late potentials a special high-pass filter
is used to reduce low frequency components and to pass the
higher frequency components. The result of this is shown in
panel (c). The QRS-complex appears as a very large multiphasic signal because of the very large amplification and the frequency-selective nature of the filter. The late potentials are
clearly seen as the post QRS components. One commonly used
presentation format is to combine the XYZ leads into a vector
magnitude (X2 Y2 Z2)1/2. This is shown in panel (d), and
the late potentials are shown in the shaded region. By making
specific measurements on this filtered, vector magnitude, a
number of parameters have been used to characterize the patient at high risk from life-threatening arrhythmias. The
power of this method is significantly increased when used in
conjunction with other clinical tests.
The Intracardiac ECG
There are many applications where electrical activity measured directly from the heart surface is the primary form of
ELECTROCARDIOGRAPHY
data recorded for analysis. A modern electrophysiological

evaluation of the heart relies on both the body surface ECG
and direct recordings obtained from within the heart by electrode catheters. Such catheters are introduced into a leg or
arm vein or artery and advanced, under fluoroscopic control,
into the interior of one of the four chambers of the heart. An
electrode catheter is an insulated set of wires bundled within
a polyurethane sheath. The diameters of these catheters
range from about 1.0 mm to 2.5 mm. As many as 16 wires
may be in the total assembly with ring electrodes, exposed on
the outer surface of the catheter, attached to each internal
wire. In addition, there are usually structural internal wires
used to stiffen the catheter. With a proper controller at the
rear of the catheter, a trained operator can flex the catheter
in a loop of almost 180. Together with the torsional properties of the catheter, almost every point within the heart can
be probed for electrical events. Direct contact recordings are
381
called electrograms to distinguish them from body surface

electrocardiograms.
Figure 13 is an example of a His bundle recording. The top
two traces are leads II and V6 of the ECG, and the bottom
trace is the voltage difference from two electrodes on the indwelling electrode catheter. This internal view of cardiac activation combined with the His bundle electrogram has been
called His bundle electrocardiography (18). Atrial activation
on the catheter recording is called the A deflection and ventricular activation called the V deflection. The His bundle
potential is the central H deflection. Because the catheter is
located very close to the His bundle and AV node, it is assumed that the A deflection arises from atrial muscle tissue
close to the AV node. When combined with the surface lead
information, a number of new intervals can be obtained.
These are the PA-, AH-, and HV-intervals. The PA-interval is
a measure of atrial muscle activation time, the AH-interval is
19.250
0.346
211
Time (ms)
Time (ms)
(a)
(c)
0
27
0
24
63
30
00
00
30
00
24
21
00
00
21
00
18
15
00
12
90
0.038
0
0.077
1.925
60
21
0.115
209
0.154
3.850
30
70
0.192
18
5.775
0.231
7.700
0.269
15
y
9.625
197
12
11.550
60
Voltage (mV)
0.256
3.850
0.231
3.465
x
0.205
Time (ms)
Time (ms)
(b)
(d)
0
30
0
24
Figure 12. The high-resolution ECG derived by signal averaging the XYZ leads. Panel (a) is a
3s run of the normal scale XYZ leads. Panel (b) is a 0.3s window of the averaged XYZ leads with
a fivefold increase in the voltage scale. Panel (c) is a 40 Hz high-pass version of panel (b). Panel
(d) is the filtered vector magnitude derived from the XYZ with standard measurements indicated
(see text). (Reprinted from the Archives of Internal Medicine, vol. 148, page 1862, 1988, Copyright
1988, American Medical Association.)
21
LAS
0
30
0
24
0
21
0
21
0
18
0
15
12
0.025
90
0.385
60
0.051
30
0.770
205
67
21
0.077
18
1.155
0.102
15
1.540
0.128
1.925
0.154
12
QRS = duration = 138 ms

rms = 20 V
LAS = 54 ms
0.179
90
2.310
Voltage (mV)
2.695
60
3.080
30
Voltage (mV)
13.475
Voltage (mV)
83
0.308
90
15.400
30
17.325
382
ELECTROCARDIOGRAPHY
QRS
P
methods for identifying patients at high risk from life-threatening arrhythmias and for enabling lifesaving devices.
II
BIBLIOGRAPHY
V6
A H V
His bundle
electrogram
Figure 13. The top two traces are ECG leads II and V6, and the
bottom trace is a bipolar catheter recording, properly positioned inside the heart, showing the His bundle deflection (H), and intracardiac atrial (A) and ventricular (V) activity.
a measure of AV nodal activation time, and the HV-interval

is a measure of the ventricular conduction system activation
time.
The modern electrophysiological evaluation, or EP study,
may involve as many as 64 individual recordings within the
heart. In addition, current can be passed through these electrodes to stimulate the heart. A variety of atrial and ventricular stimulation protocols can be used which then allows the
cardiac electrophysiologist to identify pathways and mechanisms involved in most forms of arrhythmias. Besides this
diagnostic function, it is now possible to locate abnormal
structures or regions of the heart which are critical to arrhythmogenesis. By passing high-energy, radio-frequency
waves through one or more of the internal electrodes, it is
possible to cauterize or ablate the suspect tissue without
causing any widespread injury to the rest of the heart. In
many forms of arrhythmias, this ablation therapy can produce a cure for the patient.
In addition to the EP study and ablation therapy, internal
electrodes are the primary form of signal recording for both
the cardiac pacemaker and implantable defibrillator. These
devices sense cardiac activation from permanent indwelling
catheters and deliver energy to the heart through them. In
the case of the cardiac pacemaker these are low level shocks
that maintain the patients heart rhythm. In the case of the
implantable defibrillator the device monitors the patients
rhythm until a serious or life-threatening arrhythmia occurs,
and then a high-energy pulse is delivered to convert the
rhythm back to normal. Both devices rely heavily on continuous monitoring of the cardiac signals obtained from internal
catheter recordings using sophisticated implanted microprocessors and accurate means of signal detection and analysis.
CONCLUSION
The heart is a vital organ that generates an electrical signal
which is a byproduct of the electrical triggering of the mechanical function of pumping blood. Recording and analyzing
these electrical signals has a long history. The application of
modern computer-based processing has automated this process and has also opened the way to many new and exciting
1. R. M. Berne and M. N. Levy, Physiology, 3rd ed., St. Louis, MO:

Mosby, 1993.
2. J. A. Gomes (ed.), Signal Averaged Electrocardiography, Dordrecht: Kluwer Academic, 1993.
3. G. S. Wagner, Marriotts Practical Electrocardiography, 9th ed.,
Baltimore, MD: Williams and Wilkins, 1994.
4. J. M. Jenkins, Computerized electrocardiography, CRC Crit. Rev.
Bioeng., 6: 307, 1981.
5. R. Plonsey, Bioelectric Phenomena, New York: McGraw-Hill,
1969.
6. R. Plonsey and R. C. Barr, Bioelectricity: A Quantitative Approach, New York: Plenum, 1988.
7. J. G. Webster (ed.), Medical Instrumentation: Application and Design, 3rd ed., New York: Wiley, 1998.
8. J. J. Bailey et al., Recommendations for standardization and
specifications in automated electrocardiography: Bandwidth and
digital signal processing: A report for health professionals by an
ad hoc writing group of the committee on electrocardiography and
cardiac electrophysiology of the Council on Clinical Cardiology,
American Heart Association, Circulation 81 (2): 730739, 1990.
9. Voluntary standard for diagnostic electrocardiographic devices,
ANSI/AAMI EC11a, Arlington, VA: Association for the Advancement of Medical Instrumentation, 1984.
10. W. J. Tompkins and J. G. Webster, Design of Microcomputer
Based Medical Instrumentation, Englewood Cliffs, NJ: PrenticeHall, 1981.
11. F. N. Wilson, F. S. Johnston, and I. G. W. Hill, The interpretation
of the galvanometric curves obtained when one electrode is distant from the heart and the other near or in contact with the
ventricular surface, Amer. Heart J., 10: 176, 1934.
12. E. Goldberger, A simple, indifferent, electrocardiographic electrode of zero potential and a technique of obtaining augmented,
unipolar, extremity leads, Amer. Heart J., 23: 483, 1942.
13. N. J. Holter, New method for heart studies: Continuous electrocardiography of active subjects over long periods is now practical,
Science, 134: 12141220, 1961.
14. E. J. Berbari, High resolution electrocardiography, CRC Crit. Rev.
Biomed. Eng., 16: 67103, 1988.
15. E. J. Berbari et al., Recording from the body surface of arrhythmogenic ventricular activity during the ST segment, Amer. J.
Cardiol., 41: 697702, 1978.
16. P. Lander and E. J. Berbari, Principles and signal processing
techniques of the high-resolution electrocardiogram, Progr.
Cardiovasc. Dis., 35 (3): 169188, 1992.
17. M. B. Simson, Use of signal in the terminal QRS complex to identify patients with ventricular tachycardia after myocardial infarction, Circulation, 64: 235242, 1981.
18. B. J. Scherlag, P. Samet, and R. H. Helfant, His bundle electrogram: A critical appraisal of its uses and limitations, Circulation,
46: 601613, 1972.
Reading List
P. W. Macfarlane and T. D. Veitch Lawrie (eds.), Comprehensive Electrocardiology: Theory and Practice in Health and Disease, England:
Pergamon, 1989, Vols. 13.
EDWARD J. BERBARI
Indiana University/
Purdue University, Indianapolis
394
ELECTROENCEPHALOGRAPHY
An electroencephalogram (EEG) is a record of electric signals
generated by the cooperative action of brain cells or, more
precisely, the time course of extracellular field potentials generated by synchronous action of brain cells. The name is derived from the Greek words enkephalos (brain) and graphein
(to write). An electroencephalogram can be obtained by means
of electrodes placed on the scalp or directly on or in the cortex.
In the latter case it is sometimes called an electrocorticogram
(ECoG) or subdural EEG (SEEG). An EEG recorded in the
absence of external stimuli is called a spontaneous EEG; an
EEG generated as a response to an external stimulus is called
an event-related potential (ERP). The amplitude of an EEG
measured with scalp electrodes is 50 V to 200 V.
In the EEG the following rhythms have been distinguished
(1): delta (0.5 Hz to 4 Hz), theta 4 Hz to 8 Hz), alpha (8 Hz to
13 Hz), and beta (above 13 Hz, usually 14 Hz to 40 Hz) (Fig.
1). The term gamma rhythm for 35 Hz to 45 Hz activity is
now seldom used. The contribution of different rhythms to the
EEG depends on the age and behavioral state of the subject,
mainly the level of alertness. There are also considerable intersubject differences in EEG characteristics. The EEG pattern changes in different neuropathological states and is also
influenced by metabolic disorders (1).
The delta rhythm is a predominant feature in EEGs recorded during deep sleep. During deep sleep, delta waves
have usually large amplitudes (75 V to 200 V peak-to-peak)
and show strong coherence with signals acquired in different
locations on the scalp.
Theta rhythms rarely occur in humans and primates, except during infancy and childhood. In humans, activity in the
theta band is mostly attributed to the slowing of alpha
rhythms due to pathology. However, theta rhythms are predominant in rodents; in their case the frequency range is
broader (4 Hz to 12 Hz) and the waves have a high amplitude
and characteristic sawtooth shape. It is hypothesized that
theta rhythms in rodents serve as a gating mechanism in the

information transfer between the brain structures (2).
In humans, alpha rhythms occur during wakefulness and
are most pronounced in the posterior regions of the head.
They are best observed when the eyes are closed and the subject is in a relaxed state. They are blocked or attenuated by
attention (especially visual) and by mental effort (3).
Mu rhythms have a frequency band similar to alpha, but
their topography and physiological significance are different.
They are related to the function of the motor cortex and are
prevalent in the central part of the head. Mu rhythms are
blocked by motor functions (2,3).
Beta activity is characteristic for states of increased alertness and focused attention, as was shown in several animal
and human studies. It has been observed at the onset of voluntary movements and is present during the processing of
sensory information (3).
In general, it can be concluded that the slowest cortical
rhythms are related to an idle brain and the fastest are for
information processing. The EEG is observed in all mammals,
the characteristics of primates EEGs being closest to the human. Cat, dog, and rodent EEGs also resemble human EEGs,
though their spectral content is somewhat different. In lower
vertebrates EEG-like activity is also observed, but it lacks the
rhythmical behavior found in hgiher-vertebrate recordings.
The EEG is affected by central nervous system (CNS) disorders, including epilepsy, craniocerebral traumas, tumors,
cerebral inflammatory processes, degenerative and metabolic
CNS disorders, cerebral anoxia, psychiatric disorders, cerebral palsy, migraine, dementia, and pharmacological substances.
HISTORICAL REVIEW OF ELECTROENCEPHALOGRAPHY

The discovery by Luigi Galvani (1837 to 1882) of intrinsic
electrical transmission in the peripheral and central nervous
system and the discovery by Alexandro Volta (1745 to 1827)
in generating and storing electricity were historical milestones in neurophysiology and EEG research (34). Later, the
introduction of the first vacuum-tube amplifier by Alexander
yyy
;;;
;;;;
yyyy
;;;;;
yyyyy
;;;
yyy
;;;;
yyyy
yyyyy
;;;;;
;;;
yyy
;;;;
yyyy
;;;;;yyyy
yyyyy
;;;
yyy
;;;;
;;; yyyy
yyy
;;;;
T4
A2
F8
C4
F8
T3
T4
T6
C4
A2
T6
C4
F4
Fp2
F4
O2
P4
Cb2
P4
O2
Fp2
Nasion
F2
C2
F2
C2
P2
C2
P2
Fp1
F3
F3
F2
Fp1
C3
A1
T3
P3
C3
F7
T3
T5
O2
P3
Union
O3
T5
Cb1
A1
Figure 1. Electrodes (4,5).
Forbes (1882 to 1965) into neurophysiological research had

significant impact on EEG research (34).
Richard Caton (1842 to 1926) is regarded as the first scientist to investigate brain potentials. He worked on the exposed
brains of cats and rabbits, measuring electric currents by
means of a galvanometer, a beam of light reflected from its
mirror being projected onto a scale placed on a nearby wall
(4). The results showed that feeble currents of varying directions pass through the multiplier when the electrodes are
placed at two points of the external surface, or one electrode
on the gray matter, and one on the surface of skull. This first
observation can be regarded as a discovery of electroencephalographic activity. The second concerns the steady dc potential.
Adolf Beck (1863 to 1939) also investigated spontaneous
activity of the brains of rabbits and dogs. He was the first to
observe the rhythmical oscillations of brain electrical activity
(4). He also observed the disappearance of these oscillations
when the eyes were stimulated with light, the first discovery
of the so-called alpha blocking. Later, Napolean Cybulski
(1854 to 1919) presented the electroencephalogram in a
graphical form by applying a galvanometer with a photographic attachment, and was the first to observe epileptic
EEG activity elicited by an electric stimulation in a dog (4).
Progress in recording techniques, namely the application
of a double-coil galvanometer, made possible the recording of
human EEG activity. Hans Berger (1873 to 1941) was the
first to investigate human EEG activity during sleep and
changes in EEG patterns that occur with different states of
consciousness (5). His works on EEG of patients with localized and diffused brain disorders opened the way to clinical
electroencephalography, which became a diagnostic aid in
hospitals after the first World War.
THE NEUROPHYSIOLOGICAL BASIS OF THE EEG

In the brain there are two main classes of cells: nervous cells,
called neurons, and glial cells. In both of them the resting
potential is approximately 80 mV, the inside of the cells being negative. The difference of potential across a cell membrane comes from the diffeences in concentration of the cations K, Na, the anion Cl, and large organic anions. Ca
ions are less abundant, but they have an important regulatory role. The potential difference is maintained by the active
transport of K to the inside of the cell and Na to the outside;
the energy for this transport is supplied through metabolic
processes.
Neurons have the ability to generate action potentials
when the electrical excitation of the membrane exceeds a
threshold. The permeability for Na ions increases rapidly,
and influx of Na ions in the cell causes a rapid increase in
the potential, but subsequent increase of membrane permeability to K ions leads to their outflow from the cell. Since
the permeability for Na ions decreases after about 2 ms, the
inside of the cell again becomes negative with respect to surrounding medium. The negativity is even greater than before
the neuron became hyperpolarized. By this the action potential is created. The action potentials obey the all or nothing
firing rule, such that for subthreshold excitations action potentials are not generated, and for suprathreshold stimuli a
pulse of a constant amplitude is generated.
395
The synapses of the neuron are in contact with the membranes of the other neurons. When the action potential arrives at the synapse, it secretes a chemical substance, called
a mediator or transmitter, which causes a change in the permeability of the postsynaptic membrane to the ions. As a result, ions traverse the membrane, and a difference in potential (postsynaptic potentials, or PCPs) across the membrane
is created. When the negativity inside the neuron is decreased
(e.g., by the influx of Na), the possibility of firing is
higheran excitatory postsynaptic potential (EPSP) is generated. An inhibitory postsynaptic potential (IPSP) is created
when the negativity inside the neuron is increased (by the
flux of Cl ions) and the neuron becomes hyperpolarized. Unlike the action potential, the PSPs are graded potentials: their
amplitudes are proportional to the amount of secreted mediator, which depends on the excitation of the input neuron.
Postsynaptic potentials typically have amplitudes of 5 mV to
10 mV and time spans of 10 ms to 50 ms. In order to obtain
suprathreshold excitation, the amplitudes of many postsynaptic potentials have to be superimposed in the soma of a neuron. A neuron can have very abundant arborizations, making
up to 10,000 synaptic junctions with other neurons.
The electrical activity of neurons generates currents along
the cell membrane in the intra- and extracellular spaces, producing an electric field conforming approximately to that of a
dipole. Microscopic observation of this electric field requires
the synchronization of electrical activity of a large number
of parallelly oriented dipoles (6). Indeed, parallelly oriented
pyramidal cells of the cortex are to a large degree synchronized by virtue of common feeding by thalamocortical connections (2). The condition of synchrony is fullfilled by the PSPs,
which are relatively long in duration. The contribution from
action potentials to the electric field measured extracranially
is negligible.
The problem of the origins of EEG rhythmical activity has
been approached by electrophysiological studies on brain
nerve cells and by the modeling of electrical activity of the
neural populations (2,3). The question arises whether the
rhythms are caused by single cells with pacemaker properties
or by oscillating neural networks. It has been shown that
some thalamic neurons display oscillatory behavior, even in
the absence of synaptic input (7). There is evidence that the
intrinsic oscillatory properties of some neurons contribute to
the shaping of the rhythmic behavior of networks to which
they belong. However, these properties may not be sufficient
to account for the networks rhythmic behavior (2). It seems
that cooperative properties of networks consisting of excitatory and inhibitory neurons connected by feedback loops
play the crucial role in establishing EEG rhythms. The frequency of oscillation depends on the intrinsic membrane properties, on the membrane potential of the individual neurons,
and on the strength of the synaptic interactions.
The role of EEG oscillations in information processing has
not been fully recognized. However, there is strong evidence
that coherent oscillations in the beta range in a population of
neurons might be the basic mechanism in feature binding of
the visual system (8). Indeed, it seems that this observation
is not limited to the visual system and that synchronized oscillatory activity provides an efficient way to switch the system between different behavior states and to cause a qualitative transition between different modes of information
processing. In this way, neuronal groups with a similar dy-
396
namic functional state can be formed, subserving perceptual

processes. It has also been postulated that the role of synchronized oscillatory EEG activity in the alpha and theta range is
to serve as a gating mechanism for the flow of the information
through the network. Bursts of oscillatory activity may constitute a mechanism by which the brain can regulate changes of
state in selected neuronal networks and change the route of
information (2).
RECORDING STANDARDS
The EEG is usually registered by means of electrodes placed
on the scalp. They can be secured by an adhesive such as
collodion or embedded in a special snug cap. The resistance
of the connection should be less than 5 k, so the recording
site is first cleaned and diluted alcohol, and conductive electrode paste applied to the electrode cup.
Knowledge of the exact positions of electrodes is very important for both interpretation of a single recording and comparison of results; hence the need for standardization. The
traditional 1020 electrode system (9) fixes the positions of
19 EEG electrodes (and two electrodes placed on earlobes:
A1, A2) in relation to specific anatomic landmarks, such that
10% to 20% of the distance between them is used as the electrode interval [Fig. 1(ac)]. The first part of derivations name
indexes the arrays row from the front of the head: Fp, F, C,
P, and O. The second part is formed from numbers, even on
the left and odd on the right side, or z or 0 for the center.
Progress in topographic representation of EEG recordings demands a larger number of electrodes. Electrode sites halfway
between those defined by standard 1020 system have been
introduced in the extended 1020 system (10).
The EEG is a measure of potential difference; in a referential (or unipolar) setup it is measured relative to the same
electrode for all derivations. This reference electrode is usually placed on an earlobe, nose, mastoid, chin, neck, or scalp
center. There is no universal consensus regarding its best location. In the bipolar setup (mortgage) each channel registers
the potential difference between two particular scalp electrodes. Data recorded in a referential setup can be transformed into any bipolar montage, for the sake of display or
futher processing. The average reference montage can be obtained by subtracting from each channel the average activity
from all the remaining derivations. The Hjorth transform references each electrode to the four closest neighbors, which is
an approximation of the Laplace transform (LT). The LT, calculated as a second spatial derivative of the signal, represents
the scalp current density (11).
In contrast with the open question of the reference, the
necessity of artifact rejection is universally acknowledged.
The main problem lies in the lack of a working definition for
an EEG artifactit can stem from muscle or heart activity
(EMG, ECG), eye movement (EOG), external electromagnetic
fields, poor electrode contact, the subjects movement an so
on. Corresponding signals (EMG, EOG, ECG, and body movements) registered simultaneously with EEG are helpful in the
visual rejection of artifact-contaminated epochs.
An EEG is usually digitized by a 12 bit analog-to-digital
converter (ADC) with the sampling frequency ranging from
100 Hz for spontaneous EEGs to several hundred hertz for
ERPs to several kilohertz for recording short-latency far-field
ERPs. Prior to sampling, low-pass antialiasing filters are

used; high-pass filters are applied in order to eliminate artifacts of the lowest frequencies.
SLEEP EEG
A sleep EEG displays a characteristic alternating pattern.
The classical description of sleep involves division into
stages (12): stage 1 (drowsiness), stage 2 (light sleep), stage 3
(deep sleep), stage 4 (very deep sleep), and REM (dreaming
period accompanied by rapid eye movements.). A polysomnogram includes not only an EEG, but also an electrooculogram
(EOG), electromyogram (muscular activity), and respiration.
It may also include measurement of blood flow, an electrocardiogram (ECG), and the oxygen level in the blood. The EOG
is recorded by means of electrodes placed at the canthi of the
eyes. As a result of the corneoretinal standing potential (the
cornea is positive relative to the fundus), the eye movements
produce changes in the potential between electrodes.
The EOG and EMG help to differentiate REM from the
awake state: while these sleep states have similar spectral
characteristics, in REM eye movements are more pronounced,
and there is a loss of muscular activity. The sequence of sleep
stages is usually illustrated in the form of a hypnogram (Fig.
2). The recognition of states is based on the contribution of
the different rhythms and the occurrence of characteristic signal structures absent in the waking EEG, namely, sleep spindles, vertex waves, and K complexes. Sleep spindles are
rhythmic waves of frequency 11 Hz to 15 Hz characterized by
progressively increasing and then gradually decreasing amplitude. A vertex wave is a compound potential: a small spike
discharge of positive polarity preceding a large spike, which
is followed by a negative wave of latency around 100 ms and
often another small positive spike. Vertex waves are a kind
of auditory evoked response (AER), as can be judged from
their shape and place of occurrence. The K complex consists
of an initial sharp component, followed by a slow component
that fuses with a superimposed fast component. The sharp
component may be biphasic or multiphasic. Sometimes the K
complex is described only as having slow and fast components; the initiating sharp component is equated with a vertex wave (1).
Sleep stages can be briefly characterized as follows:
Stage 1. Decrease of alpha rhythm, appearance of mixed
frequencies in the 2 Hz to 7 Hz band of low amplitude,
occasional vertex waves and slow rolling eye movements
Stage 2. Spindles, vertex waves, K complexes
Stage 3. Preponderant slow rhythm, K complexes, some
spindles
Stage 4. Very slow rhythm of high amplitude, some K complexes
REM. Decrease of amplitude, faster rhythms, rapid eye
movements, and decrease of muscular activity
The evolution of slow wave activity and characteristic spindles during overnight sleep is shown in Fig. 2.
It has been found recently that when the sleep becomes
deeper the sources that drive EEG activity move from the
posterior regions of the head (prevalent in the waking state
with eyes closed) to the centrofrontal regions (13). There is a
397
Amplitude
175
150
125
100
75
50
25
Occurrence rate
18
16
14
12
10
8
6
4
2
0
Power
1
R
2
3
4
2000
0
0
Hours
Figure 2. Hypnogram, sleep spindles, vertex waves, and K complex.
tendency to perceive sleep as a continuous process, revealing

a microstructure which may be described in terms of cyclic
alternating patterns. They consist of a phase A of enhancement of electric activity and a subsequent phase B characterized by attenuation of EEG activity. Each phase lasts only
between 2 s and 60 s.
The sleep pattern changes greatly during childhood and
adolescence. In old age the contribution of stages 3 and 4 decreases markedly. The changes of the sleep pattern may be
caused not only by a normal aging, but also by degenerative
diseases. An abnormal polysomnogram is often present in
sleep disorders and in some psychiatric disorders (e.g., depression). Therefore, investigation of the sleep pattern is an
important clinical tool.
MATURATION OF THE EEG
The maturation of the brain as evidenced by the EEG has its
peak at 30 years; then it stabilizes, forming a plateau, and
starts to decay. The rate of decay is correlated with mental
health.
The first continuous signal resembling an EEG can be seen
in premature babies of conceptual age 32 weeks to 35 weeks.
EEG development in infancy and adolescence is characterized
by a shift of the EEG rhythm toward higher frequencies. In
newborns, slow delta rhythms predominate; then the basic
frequency shifts toward theta at the age of 12 months. The
posterior slow activity characteristic of young children constantly diminishes during adolescence. Alpha rhythm appears
at the age of 10 years (1). In young adults (21 years to 30
years) the EEG still shows mild signs of immaturity, including contributions of 1.5 Hz to 3 Hz and 4 Hz to 7 Hz waves
during the waking state, normally not seen past the age of 30.
The sleep pattern changes dramatically during maturation. For newborn babies REM takes most of the sleep time,
and in young children only REM and non-REM stages can
be distinguished.
Maturation changes in electrocortical activity of fetal animals also involve an increase of power in the higher frequency
bands, as was shown for fetal lambs by means of wavelet
transform (14). Increased correlation between EEG, respiratory activity, and blood pressure was also found with increasing age (15). However, morphine destroys these correlations.
These observations indicate that maturation is connected
with increased CNS integration.
Physiologically, the maturation process is connected with
the development of dendritic trees and myelination. Myelin
layers produced by glial cells cover the axons of neurons and
act as an insulator of the electrically conductive cells. The
propagation of electrical activity is faster and less energy-consuming in myelinated fibers.
EVENT-RELATED POTENTIALS
ERPs are the stimulus-induced synchronization and enhancement of spontaneous EEG activity (16). Among them, the
most clinically used are the evoked potentials (EPs), usually
defined as changes of EEG triggered by particular stimuli:
visual (VEP), auditory (AEP), somatosensory (SEP). The basic
problem in the analysis of EPs is detecting them within the
usually larger EEG activity. EPs amplitudes are one order of
magnitude smaller than that of the ongoing EEG (or even
less). Averaging is a common technique in EP analysis; it
makes possible the reduction of background EEG noise on the
assumption that the background noise is a random process
but the EP is deterministic.
398
The EP pattern depends on the nature of the stimulation,

the placement of the recording electrode, and the actual state
of the brain. Visual EPs are best seen in the posterior regions
of the head, auditory potentials at the vertex, and somatosensory EPs at the brain hemisphere contralateral to the stimulus (e.g., stimulation of the right hand will give rise to an EP
in the left hemisphere).
EPs are usually described in terms of the amplitudes and
latencies of their characteristic waves. The components occurring at different times are different in nature; they are
called early and late EPs. The early EPs of latency 10 ms
to 12 ms (sometimes called far fields) are connected with
the response of the receptors and peripheral nervous system;
late EPs (near field potentials) are generated in the brain.
In late EPs, exogenous components (primarily dependent
upon characteristics of the external stimulus and endogenous
components) and endogenous components (dependent upon
internal cognitive processes) can be distinguished. Endogenous components of latencies above 100 ms to 200 ms are
influenced by attention to the stimulus. The later components, around 300 ms, reflect recognition and discrimination
between stimuli.
EPs are widely used in clinical practice as a tests of the
integrity of the sensory pathways of their different dysfunctions. They are also helpful in the diagnosis of diffused brain
diseases (e.g., multiple sclerosis or psychiatric disorders). Particularly in the diagnosis of psychiatric disorders, identification of contingent negative variation (CNV) is helpful (1).
CNV is a potential consisting of a slow surface negativity
that depends upon the association or contingency of two successive stimuli. A first stimulus serves as a preparatory signal for the imperative stimulus, to which a response is made.
Early CNV is considered an indicator of arousal, whereas late
CNV is associated with attention to the experimental task.
CNV is a sensitive test of weakness in higher mental functions (e.g., schizophrenia, Alzheimers disease, migraine, and
anxiety) (18).
ERP potentials, also known as Bereitschaft (readiness) potentials, precede voluntary actions such as speech or movements. Usually they involve event-related desynchronization
(decrease of power in the alpha band) and an increase of high
frequencies (17).
EPILEPTIC SEIZURE DISORDERS
Epilepsy is caused by the massive synchronization of neuronal electrical activity. During an epileptic seizure, groups of
neurons discharge synchronously, creating a large-amplitude
signal and leading to uncontrollable oscillations. Tumors, infections, trauma, or metabolic and toxic disorders may be responsible for the synchronized discharges. Epilepsy is the second most common neurological disease (18). Its clinical
symptoms may involve the loss of awareness, drop attacks,
facial muscle and eye movements, aggressive outbursts, prolonged confusional states, and flexor spasms of the whole
body.
Seizure types can be divided into three main categories as
follows (18):
1. Local: the synchronized electrical activity starts in a
well-localized part of the brain. The seizure, lasting a
few seconds, is accompanied by jerking or spasms, as

well as by a loss of consciousness.
2. Generalized: the EEG patterns are bilaterally symmetrical and roughly synchronous; the epileptic activity is
spread over wide areas of both hemispheres simultaneously from the onset of attack.
3. Unclassifiable: different from types 1 and 2.
In epileptic discharges the membrane potential of cortical
and deeper neurons changes in a dramatic way, which leads
to the massive bursts of action potentials and large fluctuations of intra- and extracellular fields. The seizure initiation
is probably connected with the breakdown of the local inhibitory mechanisms. The crucial factor in the generation of epileptic activity is the synchronization of neural pools. The
mechanisms of this synchronization are probably connected
with recurrent excitation operating through positive feedback
loops. An important problem for diagnosis is the localization
of the epileptic focus, which in severe cases can sometimes be
removed by surgical intervention. Usually intracranial electrodes are placed in the suspected region, found from the
scalp EEG, in order to better localize the focus. The tests involving measurement of ERPs are performed in order to check
if the removal of that part of brain will impair vital brain
functions. The epileptic focus will not necessarily be detected
by imaging techniques such as tomography, so the information contained in the EEG, and possibly also a magnetoencephalogram (MEG), is essential for localization of epileptic foci.
EEG ANALYSIS
The original method of EEG analysis is visual scoring of the
signals plotted on paper. Modern computer analysis can extend electroencephalographic capabilities by supplying information not directly available from the raw data. However, visual analysis is still a widespread technique, especially for
detection of transient features of signals. In most cases the
agreement of an automatic method with visual analysis is a
basic criterion for its acceptance.
Due to its complexity, the EEG time series can be treated
as a realization of a stochastic process, and its statistical
properties can be evaluated by typical methods based on the
theory of stochastic signals. These methods include probability distributions and their moments (means, variances,
higher-order moments), correlation functions, and spectra.
Estimation of these observables is usually based on the assumption of stationarity, which means that the statistical
properties of the signal do not change during the observation
time. While the EEG signals are ever changing, they can be
subdivided into quasistationary epochs when recorded under
constant behavioral conditions.
EEG signal can be analyzed in the time or the frequency
domain, and one or several channels can be analyzed at a
time. The applied methods involve spectral analysis by the
fast Fourier transform (FFT), autoregressive (AR) or autoregressive moving-average (ARMA) parametric models, time
frequency and time-scale methods (wavelets), nonlinear analysis (including the formalism for chaotic series), and artificial
neural networks.
The estimation of power spectra is one of the most frequently used methods of EEG analysis (Fig. 3). It provides
information about the basic rhythms present in the signal
and can be calculated by means of the FFT. Spectral estimators with better statistical properties can be obtained by application of parametric models such as AR and ARMA models
or, for time-varying signals, the Kalman filter. For quasistationary EEGs, and AR model is sufficient. The AR model represents a filter with a white noise at the input and the EEG
series at the output; it is compatible with a physiological
model of alpha rhythm generation (19). The AR model also
provides a parametric description of the signal, and makes
possible its segmentation into stationary epochs. It also offers
the possibility of detecting nonstationarities by means of inverse filtering (1).
Interdependence between two EEG signals can be found
by a cross-correlation function or its analog in the frequency
domaincoherence. Cross-correlation can be used for comparison of EEGs from homologous derivations on the scalp.
A certain degree of difference between these EEGs may be
connected with functional differences between brain hemispheres, but a low value of cross-correlation may also indicate
pathology. Cross-covariance functions have been extensively
used in the analysis of ERPs for the study of the electrophysiological correlates of cognitive functions (20). Coherences are
useful in determining the topographic relations of EEG
rhythms. Usually, ordinary coherence calculated pairwise between two signals is used. However, for the ensemble of channels taken from different derivations the relationship between the signals may come from common driving by another
site. In order to find intrinsic relationships between signals
from different locations, partial coherences should be calculated: EEG signals recorded from the ensemble of electrodes
are realizations of one EEG process and are usually correlated (21).
The representation of EEG activity in a spatial domain is
usually performed by mapping. However, it is more effective
for a human observer to look at the map than at the table of
numbers. A map may help to make a direct comparison be-
FFT spectra
10
20
30
40
AR spectra
10
20
30
40
4s. epochs
Hz
Figure 3. FFT and AR power spectra.
399
tween the topographic distribution of EEG features and an

anatomic image (given, e.g., by a tomographic brain scan).
Three types of features are most commonly mapped for clinical applications: (1) direct variables such as amplitude, (2)
transformed variables such as the total spectral power or the
relative spectral power in a frequency band, (3) the results of
statistical test applied to given EEG feature.
The appearance of a map depends very much on the electrode reference system. Therefore, especially in many cases it
is recommended to use the spline-generated surface Laplacians, which are reference-independent. This approach approximates the source current density and cancels a common
component due to volume conduction (6,11). The maps can be
superimposed on 3-D images obtained by means of CT or MRI
scans. This approach was used to map chosen temporal segments with epileptic events, extracted by means of wavelet
analysis (22).
The problem of automatic computer-assisted EEG diagnosis is approached by means of pattern recognition techniques
that involve choosing a number of characteristic features, and
clustering and classification of these features.
One of the first automatic diagnostic methods (23) was
based on the observation that an increased amount of slow
EEG activity might be analogous to the slow activity seen in
the immature EEG. For each electrode, the maturity calculated on the basis of spectral features was compared with the
actual maturity. A significant discrepancy was considered an
abnormality. In another diagnostic system (1), the ratio of
slow to fast EEG activity and the degree of asymmetry between homologous derivations were taken into account. The
most extended system, called neurometrics (24), is based on
standardized data acquisition techniques and EEG and ERP
feature extraction. Statistical tranformations are performed
in order to achieve Gaussian distributions before application
of multivariate statistical methods such as factor analysis,
cluster analysis, and discriminant analysis. Profiles of neurometric features that deviate from age-matched normals have
been obtained for patients suffering from cognitive disorders,
psychiatric illnesses, and neurological dysfunctions.
Recently, pattern recognition and classification problems
in EEG research have been modeled in the form of artificial
neural networks (ANNs). The multilayer perceptron with
backpropagation of errors is the most common such technique
and has been used for spike detection (25). Self-organizing
ANNs have been used for recognition of topographic EEG patterns (26,27).
The methods of analysis described so far are based on the
assumption of the quasistationarity of the EEG time series.
However, the understanding of brain processes involves analysis of dynamic features of brain activity offered by time
frequency methods operating on a short time scale. The first
method aiming at dynamic analysis is the windowed Fourier
transform with a sliding window. Substantial progress has
also been achieved with wavelet analysis. The wavelet transform (WT) describes signals in terms of coefficients representing their energy content in specified timefrequency regions.
This representation is constructed by means of decomposition
of the signal over a set of functions generated by translating
and scaling one function called a wavelet. WTs have been successfully used for reconstruction of a single AEP, for parametric description of SEPs, and for other biomedical applications
reviewed elsewhere (28).
400
Frequency
However, the time and frequency resolution in WTs are

subject to certain restrictions that lead to poor frequency resolution at high frequencies, as shown in Fig. 4. The representation also depends on the setting of the time window, which
makes WT suitable mainly for the evaluation of time-locked
signals such as EP, and less appropriate for detecting structures appearing more or less randomly in the signal. This
problem has been approached by application of time-shiftand frequency-shift-invariant timefrequency distributions of
the Cohen class. However, in the resulting Wigner plots the
cross terms are present and sophisticated mathematics has to
be applied to diminish their contribution. Also, the Wigner
plots obtained by these methods, being continuous functions,
do not provide the parametrization of signal structures.
These problems can be solved by a matching pursuit (MP)
algorithm introduced by Mallat and Zhang (29), which decomposes the signal into waveforms of well-defined frequency,
time of occurrence, time span, and amplitude. A Wigner plot
of the EEG obtained by means of MP parameterization is
shown in Fig. 5. It is easy to perceive the absence of the cross
terms observed in Wigner distributions obtained by other
methods. The parametrization makes possible the statistical
evaluation of EEG features and automatic detection of desired signal structures (30). The application of MP to the detection of EEG structures is shown in Fig. 5. See Ref. 28 for
the details of modern timefrequency methods.
The determination of the geometry and orientation of cortical sources of electrical activity is a complex problem. Electri-
Time
0 signal
A
A
A
A
D
D
Figure 4. Wavelets.
0
5
5
10
10
15
15
20 Hz 20 s
Figure 5. MP: 3-D map.
cal activity propagates along neuronal tracts and by volume

conduction. The potentials measured by scalp electrodes are
attenuated by media by different conductivity (cerebrospinal
fluid, skull, skin), which results in the decrease of their amplitude by a factor 10 to 20. The determination of source localization from the field distribution involves solution of the inverse problem and is nonunique. In solving the inverse
problem, usually one or several dipole sources are assumed
and their positions and orientation are estimated by an iterative fit to the measured field (e.g., Ref. 31). A possibility of
approaching the inverse problem without assuming dipole
sources is offered by low-resolution tomography (32).
Recently, MEGrecording of the magnetic field of the
brainhas proven to be helpful in solving the inverse problem. The magnetic field is perpendicular to the electric field
that produces it. Therefore, in a MEG the sources tangential
to the brain surface will be more visible, contrary to the EEG,
where the contribution of radial sources is larger. The combination of EEG and MEG is an optimal solution. Unfortunately, magnetoencephalographs are still very expensive.
Methods of brain activity localization such as positron
emission tomography (PET) and nuclear magnetic resonance
(NMR) give a measure of metabolic rate or glucose consumption, not the brain electrical activity itself. Although their
spatial localization properties are good, their time resolution
is much lower than that of EEG and MEG. Therefore, they
are not likely to replace EEG, which is a totally noninvasive
and low-cost technique capable of providing information
about relationships between cortical sites.
1
2
3
MODELS OF EEG GENERATION AND

CHAOTIC PHENOMENA IN EEG
...
...
log N
The most successful models of EEG developed so far are based

on the consideration of neural populations characterized by
pulse density and slow electrical activity amplitude due to
postsynaptic potentials. The dynamic behavior is described in
terms of differential equations (3,18). It has been shown that
populations of excitatory and inhibitory cells connected by a
feedback loop produce rhythmic activity of frequency and
bandwidth depending on the coupling strength determined by
synaptic interactions (18).
A model of the olfactory system was considered in a linear

and a nonlinear regime (3). Nonlinear characteristics of the
transition between slow activity and pulses produced chaotic
behavior of neural populations. A chaotic system may be described by its trajectory in the phase space, which usually
becomes confined to a limited region of the phase space called
a basin of attraction. It was postulated that the complexity of
dynamics depended on the behavioral state; the recognition
of the stimulus was connected with a dynamics described by
strange attractor of lower dimension than the attractor describing the state of alertness (3). The physiological observation of chaotic attractors is difficult; the multitude of processes running in parallel in the brain and their changing
dynamic pattern make the resulting EEG process apparently
stochastic (except for the special situations when large pools
of neurons are synchronized, e.g., in epileptic seizure). In the
search for chaotic behavior of EEGs, it is important to check
the results by comparison with surrogate data (signals where
the phase relations were destroyed), since the procedures of
calculation of attactor dimensions and Lyapunov coefficients
are subject to large systematic and statistical errors. The
most striking chaotic dynamics is observed in the vicinity of
an epileptic focus, which paves the way for diagnostic applications (33).
The EEG historically has been an important diagnostic
tool, and more recent investigations have proven its significance for understanding information processing by the brain.
BIBLIOGRAPHY
401
ellipsoidal surfaces, IEEE Trans. Biomed. Eng., 40: 145153,

1993.
12. A. Rechtschaffen and A. Kales, A Manual: Standardized Terminology, Technique and Scoring System for Sleep Stages of Human
Subjects, Los Angeles: Brain Information Service/Brain Res.
Inst., Univ. California, 1968.
13. K. J. Blinowska et al., Methods of topographical timefrequency
analysis of EEG in coarse and fine time scale, in F. L. Silva, J.
C. Principe, and L. B. Almeida (eds.), Spatiotemporal Models in
Biological and Artificial Systems, Amsterdam: IOS Press, 1996.
14. M. Akay et al., Timefrequency analysis of the electrocortical activity during maturation, Biol. Cybernet., 71: 169176, 1994.
15. M. Akay and H. Szeto, Investigating the effects of drugs on the
fetal EEG using wavelet transform, Biol. Cybernet., 72: 431
437, 1995.
16. E. Basar, EEG Brain Dynamics: Relation between EEG and Brain
Evoked Potentials, Amsterdam: Elsevier, 1980.
17. G. Pfurtscheller, A. Stancak, and C. Neuper, Post-movement beta
synchronization. A correlate of an idling motor area? Electroenceph. Clin. Neurophys., 98: 281293, 1996.
18. J. S. Dunkan, S. D. Shorvon, and D. R. Fish, Clinical Epilepsy,
New York: Churchill Livingstone, 1995.
19. A. van Rotterdam et al., A model of the spatialtemporal characteristics of the alpha rhythm, Bull. Math. Biol., 44: 283305,
1982.
20. A. S. Gevins, Dynamic functional topography of cognitive tasks,
Brain Topography, 2: 3756, 1987.
21. M. Kamipski, K. J. Blinowska, and W. Szelenberger, Topographic
analysis of coherence and propagation of EEG activity during
sleep and wakefulness, Electroenceph. Clin. Neurophysiol., 102:
216227, 1997.
1. E. Niedermayer and F. Lopes da Silva (eds.), Electroencephalography: Basic Principles, Clinical Applications, and Related Fields,
3rd ed., Baltimore: Williams & Wilkins, 1993.
22. L. Senhadij et al., Wavelet analysis of EEG for three-dimensional

mapping of epileptic events, Ann. Biomed. Endo., 23: 522543,
1995.
2. F. H. Lopes da Silva, The generation of electric and magnetic

signals of the brain by local networks, in R. Greger and U. Windhorst (eds.), Comprehensive Human Physiology, Heidelberg:
Springer-Verlag, 1996.
23. M. Matousek, P. Petersen, and S. Freeberg, Automatic assessment of randomly selected routine EEG records, in G. Dolce and
H. Kunkel (eds.), CEANComputerized EEG Analysis, Stuttgart:
Fisher, 1975, pp. 421428.
3. W. J. Freeman, The physiology of perception, Sci. Amer., 264:

7885, 1991; C. A. Skarda and W. J. Freeman, How brain makes
chaos in order to make sense of the world, Behav. Brain Sci., 10:
161195, 1987.
24. E. R. John et al., Neurometrics: Computer assisted diagnosis of

brain dysfunctions, Science, 29: 162169, 1988.
4. M. A. B. Brazier, A History of the Electrical Activity of the Brain.

The First Half-Century, London: Pitman, 1961.
5. P. Gloor, Hans Berger on the Electroencephalogram of Man, Amsterdam: Elsevier, 1969.
6. P. L. Nunez, Electric Fields of the Brain, New York: Oxford Univ.
Press, 1981.
7. H. Jahnsen and R. Linas, Electrophysiological properties of
guinea-pig thalamic neurones: An in vitro study, J. Physiol.
(Lond.), 349: 205226, 1984.
8. C. M. Gray et al., Mechanisms underlying the generation of neuronal oscillations in cat visual cortex, in E. Basar and T. H. Bullock (eds.), Induced Rhythms in the Brain, Birkhauser, Boston,
1992, pp. 2945.
9. H. Jasper, Report of the Committee on Methods of Clinical Examination in Electroencephalography, Electroenceph. Clin. Neurophysiol., 10: 370375, 1958.
10. R. T. Pivik et al., Committee report: Guidelines for the recording
and quantitative analysis of electroencephalographic activity in
research context, Psychophysiology, 30: 547558, 1993.
11. S. L. Law, P. L. Nunez, and R. S. Wijesinghe, High resolution
EEG using spline generated surface Laplacians on spherical and
25. T. Kalayci and O. Ozdamar, Wavelet preprocessing for automated

neural network detection of EEG spikes, IEEE Eng. Med. Biol.
Mag., 14 (2): 160166, 1995.
26. S.-L. Jountsdiniemi, S. Kaski, and T. A. Larsen, Self-organizing
map in recognition of topographic patterns in EEG spectra, IEEE
Trans. Biomed. Eng., 42: 10621067, 1995.
27. A. J. Gabor, R. R. Leach, and F. U. Dowla, Automated seizure
detection using a self-organizing neural network, Electroenceph.
Clin. Neurophysiol., 99: 257266, 1996.
28. M. Akay (ed.), TimeFrequency and Wavelets in Biomedical Signal
Processing, Piscataway, NJ: IEEE Press, 1997.
29. S. G. Mallat and Z. Zhang, Matching pursuit with time-frequency
dictionaries, IEEE Trans. Signal Process., 41: 33973415, 1993.
30. P. J. Durka and K. J. Blinowska, Analysis of EEG transients by
means of matching pursuit, Ann. Biomed. Eng., 23: 608611,
1995.
31. C. Bumgartner et al., Investigation of multiple simultaneously
active brain sources in the electroencephalogram, J. Neurosci.
Methods, 30: 175184, 1989.
32. R. D. Pascual-Marqui, C. M. Michel, and D. Lehmann, Low resolution electromagnetic tomography: A new method for localizing
electrical activity in the brain, Int. J. Psychophysiol., 18: 4965,
1994.
402
ELECTROLUMINESCENCE
33. F. H. Lopes da Silva, J. P. Pijn, and D. Velis, Signal processing

of EEG: Evidence for chaos or noise. An application to seizure
activity in epilepsy, in I. Gath and G. F. Inbar (eds.), Advances
in Processing and Pattern Analysis of Biological Signals. New
York: Plenum, 1996.
34. E. S. Goldensohn, Animal electricity from Bologna to Boston,
Electroenceph. Clin. Neurophys., 106: 94100, 1998.
METIN AKAY
Dartmouth College
KATERZYNA BLINOWSKA
PIOTR DURKA
University of Warsaw
ELECTROFILTER. See ELECTROSTATIC PRECIPITATOR.
ELECTROMYOGRAPHY
523
ELECTROMYOGRAPHY
ANATOMY AND PHYSIOLOGY
Muscles convert chemical energy into mechanical energy.
Since they can pull but not push, at least two muscles are
needed for each joint connecting two body segments: the agonist and the antagonist. Cocontraction of both muscles causes
524
ELECTROMYOGRAPHY
carefully controlled movements and stiffening of the joint.

Skeletal muscles are striated, whereas muscles of internal organs (stomach and intestines, vessels, uterus) are smooth.
Muscles consist of long thin cells (fibers) whose membranes
are excitable. The membrane presents a resting voltage of
about 70 mV between the inside and the outside. If the
membrane voltage is disturbed to the point that a threshold
is reached, a phenomenon called action potential, similar to
the firing of a monostable multivibrator (one shot), is triggered. This voltage transient evolves in time for 25 ms and
propagates in space from the point of trigger to the two ends
of the fiber with a velocity of 36 m/s. The traveling action
potential generates a field of current flowing into the surrounding conductive medium (volume conductor) and producing voltages detectable between any two points in the surrounding medium as well as on the skin surface.
Motor neurons establish a connection between the spinal
cord and the muscles. When it reaches the muscle, each motor
neuron branches into a number of terminals, each of which
makes an electrochemical connection, called the end-plate or
neuromuscular junction (NMJ), with a single muscle fiber.
Each fiber has only one NMJ and belongs to only one motor
neuron. A motor neuron and the muscle fibers it innervates
form a motor unit (MU) whose fibers are activated synchronously and may range in number from approximately 10 to
2000. A muscle may contain 50 to 1500 MUs whose fibers
are scattered in overlapping territories. Fibers differ in size
(diameter may range from 10 m to 100 m, length may
range from a few millimeters to 300 mm), metabolism, and
electrical and mechanical behavior. Two main types can be
identified: Type I fibers are smaller, are more fatigue-resistant, produce lower forces, and have lower contraction speed
and lower conduction velocity. Type II are usually larger, are
less fatigue-resistant, produce higher forces, and have higher
contraction speed and conduction velocity. The two fiber types
have different metabolism, and their distribution is affected
by genetic factors as well as training (1).
Each action potential transmitted along a motor neuron is
called a firing and triggers a mechanical contraction (single
twitch) of the MU. As the firing frequency increases from a
few pulses per second to 15 to 20 pulses per second the MU
has no time to relax between pulses and remains contracted,
generating a tetanic contraction. The brain controls and
smoothly adjusts the level of muscle force by controlling the
number and the firing frequencies of the activated MUs.
ELECTROPHYSIOLOGY OF MUSCLES AND EMG DETECTION

The summation of the single-fiber action potentials generated
by a MU is the MU action potential (MUAP). A sequence of
firings generates a MUAP train. The electromyographic
(EMG) or myoelectric signal is the voltage detected with two
(or more) electrodes within the volume conductor (needle or
wire detection) or on the surface of the skin (surface detection). Figures 1(a) and 1(b) show the depolarization zone of a
cell, its schematic representation as a current tripole, and its
contribution to the surface and needle potentials for two
depths. The deeper the source, the smaller and more diffused
is its two-dimensional potential distribution on the surface.
As this distribution travels on the surface and in the direction
of the fiber, it generates a time-varying potential at the detection point(s).

Figure 1(c) shows a schematic for a three-fiber MU, the
traveling action potentials of one fiber, and the needle and
surface detection technique. Figure 2 shows examples of needles used for EMG detection. The detection volume of an electrode system is the volume from which signal contributions
are above noise level. This volume depends on the interelectrode distance; it is a hemisphere of about 0.51 mm diameter
in the case of needles and a few centimeters in the case of
surface detection. At the present state of the art, surface detection provides information about superficial muscles only.
Surface signals have amplitudes ranging from the noise level
to a few millivolts, with most of the power comprised between
10 Hz and 300 to 400 Hz.
Needle detection allows the monitoring of potentials generated by fibers that may belong to a few different MUs, and it
provides local information with good morphological details
that allow identification and separation of the contributions
due to different MUs as well as the recognition of MUAP
shapes that reflect pathologies. Needle-detected signals have
amplitudes ranging from the noise level to a few millivolts,
with most of the power ranging from 10 Hz to 1 kHz.
Surface detection provides global information with poor
morphological details. Distant but large muscle portions may
provide signal contribution comparable to those of smaller
and closer portions. When these contributions come from
muscles different from the one in or on which the electrodes
are placed, they are referred to as crosstalk.
When many MUs firing asynchronously contribute to the
detected signal, algebraic summation takes place; the signal
appears random and is referred to as the interference pattern.
This is the signal observable during voluntary muscle contractions. If the motor neurons (or their terminal branches)
are activated synchronously by an external electrical stimulator, the different MUAPs are synchronized and add to form
the compound action potential or motor wave (CAP or Mwave).
APPLICATIONS OF EMG
Today the main clinical application of the EMG techniques
is based on the use of needles for diagnosing neuromuscular
diseases that modify the morphology of MUAPs. However,
surface techniques are becoming more popular because they
are noninvasive and inexpensive and provide global information. The most important applications of surface techniques
are listed below.
Estimation of Nerve Fiber Conduction Velocity. A peripheral
nerve is electrically stimulated and the muscle response (Mwave) is detected. Measurement of the distance between the
stimulation and detection sites, along with appropriate measurement of the stimulusresponse delay, allows estimation
of the nerve fibers conduction velocity (2).
Myoelectric Manifestations of Muscle Fatigue. As a voluntary
or electrically elicited muscle contraction is sustained in time
under isometric conditions, the EMG signal becomes progressively slower. This change precedes the inability to sustain
the required effort (mechanical fatigue), is referred to as myoelectric manifestations of muscle fatigue, and depends on the
fiber type constituency of the muscle. It is likely that current
ELECTROMYOGRAPHY
V
Vs
X
X
70 mV
I
Propagation direction
h1
h2
Current tripole
0
X
Current field
I
V
X
X
Current source
Current source
Current sink
(a)
(b)
V
Innervation zone
+
_
t
Reference
electrode
Needle
Subcutaneous layer
1
2
3
Depolarization zone
Z
0 mV
CV = 4m/s
Termination
zone
70 mV
Action potentials of fiber #3
(c)
Figure 1. (a) Depolarization zone of a muscle fiber, description of the membrane current, and
monopolar surface potentials Vs generated on the skin by two depolarized zones at two depths
h1 and h2. (b) Muscle fiber transmembrane voltage, current, and tripole model of the transmembrane current. (c) Schematic representation of a motor unit (example with three fibers only) and
of the signal detected by a differential amplifier (surface electrodes) or by a coaxial needle. Physical dimensions are not in correct proportions.
525
526
ELECTROMYOGRAPHY
1 mm
(a)
cle group) helps him/her in learning strategies to increase the

voluntary control or decrease the involuntary activity of the
muscle and recover whatever degree of voluntary control is
possible after a lesion (10).
Occupational Medicine and Ergonomics. EMG is used in ergonomic studies to evaluate how workplace factors such as
tasks, posture, tool design, layout, and so on, influence the
activity of a set of muscles (11).
0.5 mm
.
...
..........
FINE WIRE AND NEEDLE EMG; DECOMPOSITION

INTO THE CONSTITUENT MUAP TRAINS
The EMG signal may be detected invasively with fine wires

or needle electrodes, or noninvasively with surface electrodes.
Fine wire electrodes are made of a spiral or multitread insulated stainless steel wire, with diameter 25 m or 50 m,
with a bared hook-shaped terminal portion about 1 to 2 mm
long. The wire is positioned inside the muscle by a hypodermic needle which is then withdrawn leaving the wire in position. This technique is applied for kinesiological studies and
gait analysis. Technical details about fine wire electrodes can
be found in Ref. (10).
Concentric (coaxial) needle electrodes are more selective
than wires and allow the detection of the interfering contributions of up to 10 to 15 MUs. The separation of the constituent
MUAP trains provides information on the firing rate of the
individual MUs and on the change of this rate during increasing or decreasing contraction level. This task becomes progressively more difficult as the contraction level increases and
the resolution of partially overlapping MUAP (superpositions)
becomes more important. At the present state of the art it is
possible to separate 10 to 12 MUAP trains during voluntary
contractions of level up to the maximum. Figure 3 is a diagram of the system which has been provided by Basmajan
and De Luca (10) and used by Guiheneuc, Haas and Meyer,
McGill and Dorfman, Jabre, Kamen, and De Luca, whose
work is collected in Ref. 12. The algorithms used for decompo-
(b)
Surface
Figure 2. (a) Examples of EMG needles. A, single-fiber needle with

one recording surface; B, single fiber needle with multiple recording
surfaces; C, concentric needle; DF, dual electrodes; G, monopolar
needle (insulated, only tip exposed); H, macro-EMG electrode. (b)
Macro-EMG detection and macro-EMG electrodes [from Jabre in Desmedt (12)].
research will lead to a noninvasive estimation of the percentage of Type I and Type II fibers thereby reducing the need for
muscle biopsies (35).
Myoelectric manifestations of muscle fatigue are also observable in intermittent isometric contractions, isokinetic
contractions, and, in general, in dynamic contractions. Particularly important fields of interest concern respiratory muscle
fatigue and back muscle fatigue in occupational medicine (6).
Gait Analysis and Muscle Activation Intervals. During movements, such as gait, sport activities, or rehabilitation exercises, it is important to detect the time and the level of individual muscle activations. Surface EMG is the appropriate
tool for this purpose. Crosstalk and relative movement between muscle and electrodes still represent important confounding factors (7,8).
Control of Myoelectric Prosthesis. The motors of artificial
limbs (mostly hands, wrists, and elbows) may be controlled by
surface EMG signals detected from muscles above the level
of amputation. Many systems of this kind are commercially
available (9).
Biofeedback. Providing a patient with real-time information about the level of activity of a particular muscle (or mus-
Raw
EMG signal
Electrode
Muscle
Decomposition
-Motoneurons
Individual motor
unit action potential
trains (MUAPTs)
Figure 3. A schematic representation of the decomposition of the

needle EMG signal into its constituent motor unit action potential
trains [from Basmajan and De Luca (10)].
ELECTROMYOGRAPHY
sition are based on (a) MUAP shape information and (b)

MUAP firing statistics. An interesting observation made possible by precision decomposition is that the firing rates of different motor units often fluctuate together as if they were entrained by a common drive mechanism. This phenomenon was
first studied in detail by Basmajan and De Luca (10) and was
reported, among others, by Kamen and De Luca in Ref. 12.
Special needle electrodes (Fig. 2) allow the detection of single-fiber extracellular potentials (single-fiber EMG). Macro
EMG is a technique that uses single-fiber firing information
to trigger the acquisition of the signal detected between the
needle cannula and a skin reference electrode. By averaging
the signal detected between the cannula and the reference,
using the single-fiber signal as a trigger (the technique is similar to that used for EEG evoked potentials), the contribution
of the MU to which the single fiber belongs can be separated
from signals of other MUs. By slow withdrawal of the MacroEMG needle and search for stable single-fiber signals, one can
scan a large portion of the muscle and study many MUs. This
technique is called scanning Macro EMG (described by Jabre
in Ref. 12).
A technique that has been extensively used to provide a
quantitative evaluation of needle (as well as surface) EMG is
the turn and amplitude (T&A) analysis. The term turn or
count is defined as the occurrence of a peak separated by the
preceding and following ones by at least a V (usually 100
V) and at least a t (usually 0.3 ms). The distribution, mean,
and variance of the interturn amplitude and time intervals
provide indices of signal complexity that may be related to
pathological situations (as reported by Gilai in Ref. 12). The
frequency of crossings of a 100 V band centered on zero (zero
crossings) is also used in clinical practice. This frequency is
related to the second-order moment of the power spectral density of the EMG signal. Other descriptions of complexity are
based on chaos and fractal approaches.
SURFACE EMG DETECTION, SPATIAL

FILTERS, AND LINEAR ARRAYS
Detection of surface EMG is performed with small bar (1 mm
diameter, 5 to 10 mm long, 5 to 10 mm apart) or disk electrodes (3 to 10 mm diameter). Detection is said to be monopolar when the voltage is measured between one point above
the muscle and one reference, electrically unrelated, location.
In this case the detection volume is large, and thus crosstalk
and power line interference may be serious problems. This
configuration is the most sensitive to nontraveling potentials
generated at the innervation and termination zones (far-field
potentials). Detection is said to be bipolar (or single differential) when a differential amplifier is used to detect signals
present between two points on the same muscle, usually in
the direction of the fibers. This system, which samples the
voltage in two spatial locations and computes their difference,
is the simplest form of spatial filter. As shown in Fig. 4(a),
neither a common mode voltage (spatial dc component) nor
any sinusoidal distribution of potential in space with half
wavelength /2 e/2n, for any integer n, will be detected. On
the other hand, sinusoidal distributions with /2 e/(2n
1) will generate differential signals with amplitude equal to
the peak-to-peak amplitude of the monopolar distribution. In
general a spatial potential distribution will have many sinus-
527
oidal components (spatial Fourier series or transform) for

which the detection system will present different sensitivities
(hence the name of spatial filter). If e min the single differential system approaches the one-dimensional spatial differentiation. This detection modality is the most appropriate to
outline the innervation zone [see Fig. 5(a)]. Another commonly used detection method is the double differential filter
depicted in Fig. 4(b) whose dual version is often used for muscle fiber conduction velocity estimation (13) because it is the
least sensitive to nontraveling signal components [see Figs.
5(a) and 14]. Two-dimensional filters can be used as well to
improve selectivity. The two-dimensional double differentiator (Laplacian) is depicted in Fig. 4(c). More sophisticated detection techniques are being developed for high-resolution
EMG (see also Fig. 10) (14).
Linear surface electrode arrays provide useful geometric
information about individual MUs. Figure 5(a) shows an example of monopolar, single differential and double differential
detection of surface EMG with a linear array. Figure 5(b) depicts an example of single differential recording showing
three MUs with different features. It can be observed that the
technique for recognition and classification of motor units is
in this case very different from the one used with needle detection. While morphological information about the MUAP is
lost because of tissue filtering, additional information about
innervation zone location, fiber length, and conduction velocity is gained from the array and may be used for recognition,
classification, and clustering.
The surface of the detection electrodes affects the contact
noise (the smaller the electrode the higher the noise), and has
a smoothing effect on the signal because of the averaging of
the potential underneath the contact area (the larger the electrode the greater the smoothing effect).
MYOELECTRIC MANIFESTATIONS OF MUSCLE FATIGUE

DURING VOLUNTARY AND ELECTRICALLY
ELICITED ISOMETRIC CONTRACTIONS
Voluntary and electrically elicited contractions provide two
ways of studying the same system. Figures 6(a) and 6(b) show
the model of generation of EMG during these two conditions.
In the first case the detected signal is stochastic, while in the
second it is deterministic because the contributions of the active motor units are synchronized. It is well known that during a sustained contraction the surface myoelectric signal becomes progressively slower, and this slowing reflects
physiological changes of the muscle fiber membranes. During
electrically elicited contractions, this slowing appears to be
a combination of scaling (stretching in time and in amplitude)
and change of shape of the M-wave. During voluntary contractions, the slowing is more difficult to quantify since the
signal is random. A quantitative evaluation of this slowing
during stimulation may be obtained by stretching the initial
M-wave until it matches (in the mean square sense) the subsequent and progressively wider ones. The scaling coefficient
would indicate how slowing evolves in time. This approach
is not suitable for voluntary signals since they never repeat
and there is no reference signal to stretch. It is therefore
preferred to apply the stretching and matching procedure in
the frequency domain to the power spectral density of the signal, considering that, for any signal, scaling in time by a
528
ELECTROMYOGRAPHY
VD =V1 V2
Impulse response
Impulse response of the spatial filter

hD(x) = (x + e/2) (x e/2)
Transfer functions of the spatial filter
in spatial and time frequency
HD(fs) = 2jsin( efs)
e
1
HD(f) = 2jsin( efs/V)
2
x
Vm
Traveling ac
component
dc component
(a)
HD (f)
V/e
V/2e
+
Impulse response of the spatial filter
hDD(x) = (x + e) 2 (x) + (x e)
+
Transfer functions of the spatial filter

in spatial and time frequency
Impulse response
HDD(fs) = 4jsin2( efs)

HDD(f) = 4jsin2( efs/V)
V2
V1
V3
x
Vm
HDD (f)
f
x
V/2e
(b)
VL = V1 + V3 + V4 + V5 4V2
V1
1
4
V4
V2
V3
VL
1
1
V5
1
(c)
Figure 4. Detection techniques and spatial filtering effects. v conduction velocity, Vm monopolar voltage. (a) Single differential (or bipolar) detection. Impulse response and transfer function
of the spatial filter. (b) Double differential detection. Impulse response and transfer function of
the spatial filter. (c) A Laplacian spatial filter performing a discrete two-dimensional differentiation.
V/e
ELECTROMYOGRAPHY
Monopolar
detection
Array
Double differential
(DD) detection
Bipolar (SD)
detection
15
14
Innervation
zone
529
16
Reference electrode
Nontraveling
components
Innervation and
termination zones
Conduction
velocity est.
(a)
11
10
9
8
7
6
5
4
3
2
1
0
10
20
30
Time (ms)
Innervation zones
(b)
40
Figure 5. (a) Linear array detection of a single motor unit action potential. A nontraveling potential is seen in the monopolar detection and is due to the extinction
of the depolarized zones at the tendon endings. Innervation and termination zones
can be clearly seen in the single differential recording. Conduction velocity may be
well estimated using the double differential signals. Signals are simulated using
the model depicted in Fig. 14. (b) Example of array detection of three motor units,
with different properties, from the biceps brachii of a healthy subject. Interelectrode
distance is 5 mm. Straight lines have been added to outline the bi-directional propagation and the different innervation zones [from Merletti and Lo Conte (15)].
factor k implies a scaling in frequency by a factor 1/k. This

approach in the frequency domain is applicable to both situations and is widely used. Its operation is described in the following.
Consider the experimental EMG signals x1(t) and x2(t)
x1(kt), taken during time epochs 1 and 2 (e.g., two 0.5 s
epochs) during a sustained contraction, with power spectral
densities P1( f) X1( f)2, and P2( f) P1( f/k)/k2. The mean frequency (MNF, also called centroid or first moment) and me-
dian frequency (MDF) of a general P( f) are defined as
f mean =
fP( f ) df
P( f ) d f
(1)
and
f med
0

P( f ) df =
f med
P( f ) df = 0.5
0
P( f ) d f
MU #1
MU #2
Motor
neurons
Contributions of
each MU
MU #N
(a)
MU #1
MU #2
Motor
neurons
Contributions
of each MU
MU #N
(b)
Initial values: MNF = 131 Hz
RMS = 0.342 mV CV = 5.49 m/s
Torque = 60% MVC
EMG signal
160
Percent of initial value
1.0
MNF = 129 Hz
0.5
1.0
RMS
140
EMG PSD
1.0
0.5
0.0
0.5
1.0
0.5
0.0
0.5
120
100
0.0
MNF = 100 Hz
1.0
0.5
0.0
1.0
Torque
1.0
0.5
0.0
0.5
80
MNF
60
0.5
0.0
1.0
40
20
1.0
MNF = 81 Hz
CV
0
20
40
60
80
1.0
0.5
0.0
0.5
100
Time (s)
Fatigue plot (MNF, RMS, CV, Torque) at 60% MVC
0
(c)
Figure 6
530
50
100 150
Time (ms)
1.0
200
0
1.0
MNF = 68 Hz
0.5
100 200 300

Frequency (Hz)
0.0
ELECTROMYOGRAPHY
Initial values: MNF = 119 Hz
RMS = 1.24 mV CV = 5.1 m/s
EMG signal (mV)
140
130
RMS
(20 s fit)
Percent of initial value
120
110
100
CV
90
80
MNF
70
60
50
10
15
Time (s)
20
25
30
Fatigue plot (MNF, RMS, CV)

0
EMG PSD
6
4
MNF = 119 Hz
2
0
2
4
6
4
MNF = 84 Hz
2
0
2
4
6
4
MNF = 74 Hz
2
0
2
4
6
4
MNF = 70 Hz
2
0
2
4
10 15 20 25 30
0
100 200 300
Time (ms)
Frequency (Hz)
531
1.0
0.5
0.0
1.0
0.5
0.0
1.0
0.5
0.0
1.0
0.5
0.0
(d)
Figure 6 (Continued ) (a) Schematic diagram of generation of voluntary EMG. (b) Schematic
diagram of generation of electrically elicited EMG. (c) Experimental data from a voluntary contraction of a tibialis anterior muscle. All variables are normalized to their initial value to obtain
the fatigue plot. Notice that voluntary torque could be maintained at 60% MVC (maximal voluntary contraction) for 60 s; but myoelectric variables started to change from the beginning of the
contraction, showing myoelectric manifestations of muscle fatigue. EMG signal detected bipolarly
with 10 mm interelectrode distance. PSD, power spectral density function; RMS, root mean
square value; MNF, mean frequency; CV, conduction velocity [from Merletti and Lo Conte (11)].
(d) Experimental data from an electrically elicited contraction of a vastus medialis muscle. Notice
the change of shape of the M-wave. Detection as in c; f 30 pulses per second [from Merletti
and Lo Conte (15)].
It can be shown that f mean2 kf mean1 and f med2 kf med1. In general the two signals (and the relative spectra) will not be exactly scaled and the ratios f mean2 /f mean1 and f med2 /f med1 will not
be identical but can provide an estimate of k and quantify the
scaling phenomenon. If we define A as the average rectified
value (ARV) and R as the root mean square value (RMS), then
it is A2 A1 /k and R2 R1 / k. The normalized plots of MNF,
MDF, ARV, RMS, and CV (see next section for a discussion of
CV estimation) versus time describe signal changes and are
often referred to as the fatigue plot (15).
Figures 6(c) and 6(d) show an example of fatigue plots and
signals in both experimental situations. It has been demonstrated from animal experiments that the rate of decay of
MNF or MDF in each musclethat is, the estimated scaling
factor kis related to the percentage of Type I and Type II
fibers in the muscle (5). This finding suggests the possibility
of noninvasive fiber type estimation and, if confirmed by human biopsies, is expected to be very relevant in future research concerning rehabilitation and sport medicine. More advanced approaches are being developed to separate the
contribution of scaling from that of spectral shape change and
to relate them to different underlying physiological phenomena.
Applications concern rehabilitation, sports and occupational medicine. An important clinical application of myoelectric manifestations of muscle fatigue concerns the analysis of
back muscle impairment for the investigation of back problems and low back pain (6,16). Issue 4 of volume 34 of the
Journal of Rehabilitation Research and Development (1997) is
devoted to this topic. In particular, the works of Roy et al. (17)
and Oddson et al. (18) focus on the classification of muscle
impairments and on the development of clinical protocols.
The spectral approach requires the signal to be quasi-stationary, that is, its statistical properties must not change
during each time epoch. This requirement is not satisfied during dynamic contractions when the EMG is often generated
in short bursts. More sophisticated methods, based on timefrequency representations and wavelet expansions, are being investigated for the quantification of myoelectric manifestations of muscle fatigue in dynamic conditions.
MEASUREMENT OF MUSCLE FIBER CONDUCTION VELOCITY

The velocity at which a muscle fiber conducts an action potential along its length is an indication of its functional state.
Thus the measurement of conduction velocity (CV) is used to
study muscles in both clinical and research applications. The
measurements can be made with voluntary or evoked potentials and with invasive or noninvasive methods. The range of
values for CV in muscle is 3 to 6 m/s, with an average value
near 4 m/s (1923).
532
ELECTROMYOGRAPHY
As in the case of any propagating signal, the conduction

velocity value, v, is typically determined (see Fig. 7) from the
measurement of propagation time delay, , over some known
distance, d; that is, v d/. The time delay can be measured
from observation of the signals S1(t) and S2(t), or some function of the signals, at two electrode sites spaced a distance d
along the axis of the fiber, where S2(t) S1(t ). This latter
condition can be difficult to achieve in practical EMG work
because of the physical dimension of the electrodes and of possible misalignment between fibers and electrodes in either the
vertical or horizontal planes. These issues result in measurement errors.
Voluntary EMG
The CVs of single muscle fibers are usually measured with
two intramuscular needle electrodes (19,20), in which case the
delay is measured from the time displacement of identical
points on the two waveforms. In the event that two or more
fibers are detected by the electrodes, they can be individually
identified, using decomposition techniques, and their respective CVs measured. In this manner and with different electrode positions, a measure of the fiber CV distribution can
be obtained.
The CV of a single motor unit must be defined because the
motor unit action potential (MUAP) is the spatial and temporal sum of signals from the constituent fibers, each of which
can have a slightly different CV. With M1(t) and M2(t) being
the MUAPs seen at the two electrodes, we have
M1 (t) =
N
Si (t) and M2 (t) =
i=1
N
Si (t i )
(2)
i=1
where i d/vi, N is the number of fibers in the unit, and

Si(t) and Si (t) are the ith fibers signals at electrodes 1 and 2,
respectively. Now it is clear that M2(t) M1(t ) only if
i , and Si (t) Si(t) for all i; otherwise M2(t) is different
from M1(t) and delay is not well-defined. Indeed there is a
range of delays, and if we proceed to measure delay from two
similar points at the onsets of M1(t) and M2(t), we will have
the delay corresponding approximately to maximum velocity
and, from the tails, approximately the minimum velocity. If
some midpoints, say the signal peaks, are chosen for the delay measurement, then some average velocity is obtained.
However, the precise nature of this average is impossible to
define without information regarding the probability distributions for the fiber delays and waveforms.
In the case of EMG signals involving more than a few motor units, particularly when obtained with surface electrodes,
the possibility of determining conduction delay and hence velocity from instantaneous values as above is problematic. This
is because variables such as fiber depths, end-plate positions,
and so on, which affect the spatial and temporal sum output,
cause the signals from the two electrodes to be significantly
different. It is common in such cases to measure CV from the
cross-correlation function (21) or from a linear system impulse response identification approach (22). Both of these approaches give what amounts to a weighted-average CV where
the average is across fibers and the weighting depends primarily on the contribution of a fiber to the total EMG power.
The conduction delay is obtained from the cross-correlation
function by observation of the time shift of the function peak
from zero lag (see Fig. 8). This approach has the attractive
feature that the cross-correlation between signals of different
motor units goes to zero (assuming uncorrelated unit innerva-
+
S1(t)
Muscle
fibers
Figure 7. A three-bar electrode configuration with interelectrode spacing d for

the measurement of muscle fiber conduction velocity.
d
Propagation direction with
velocity V
S2(t)
ELECTROMYOGRAPHY
0.2 mV
X(t)
Y(t)
2 ms
533
plate variation will introduce a bias in the estimate. In both

invasive and noninvasive evoked measurements the signal
quality is higher than for the voluntary case as the random
innervation of the central nervous system is replaced with a
deterministic stimulus (see Fig. 6). However, the repeatability
is not necessarily better.
CV Distribution Estimation
+1
Cross-correlation function
RXY( )
2 ms
The muscle fiber CVs are distributed over a range, and the
measurement techniques described above give only a number
related to that distribution. Noninvasive techniques for CV
distribution estimation based on surface EMG give significantly more information about the muscle state, and they
would be useful for both clinical and research purposes.
Two recent approaches are based on measurements of the
cross and auto power spectra ratio of the EMGs from the two
sites (22,24). From Eq. (2) with Si(t) defined as the ith motor
unit train and assuming uncorrelated units with identical firing statistics, the ratio, ( f ), of the cross to auto power spectra is given by
( f ) =
N
Pii ( f ) exp( j2 f i )
i=1
1
Figure 8. Cross-correlation function between two EMG signal channels recorded from human biceps showing propagation delay as a
shift in peak of the function [from Li and Sakamoto (26)]. Interelectrode distance: 5 mm.
tion processes), thus eliminating erroneous delay measured

from firings of different units. In addition, the peak value of
the cross-correlation function, normalized with respect to the
signal powers, provides the cross-correlation coefficient,
which is an index of similarity between the signals and quality of the estimate of CV.
A confounding factor which is of particular concern with
the cross-correlation and impulse response identification approaches is the presence of coherent nondelayed EMG components in the two electrode signals. These nondelayed components appear due to, among other things, the termination of
the fibers at the tendon [see Fig. 5(a)] (13). This effect, which
is to give a positive bias to the CV estimate, can be greatly
reduced by using the double differential electrode configuration [see Fig. 4(b) and Fig. 5(a)].
Other techniques which have been developed but which
are not widely used include the spectrum dip, zero-crossing,
and polarity correlation. For a review of these various approaches and their application, see the work of Arendt-Nielsen and Zwarts (23).
Evoked EMG
The CV can also be measured, invasively or noninvasively,
from the EMGs or M-waves acquired at two sites in response
to an electrical stimulation of the fibers. In the case of invasive needle measurements, direct muscle fiber stimulation
may be obtained near a tendon and end-plate variation across
the fibers is not a problem (20). With surface stimulation, axonal branches are excited, many units can be involved and end-

N
Pii ( f )
(3)
i=1
where Pii( f ) is the autospectra for the ith unit, and N is the
number of units. Note that only autospectra terms appear in
Eq. (3) because of the assumption of uncorrelated units, in
which case all the cross-spectra terms are zero. Now assuming identical MUAPs (a strong limitation) across the units,
the Pii( f ) are identical and Eq. (3) becomes
( f ) = N 1
N
exp( j2 f i )
(4)
i=1
For large N, Eq. (4) is approximately equivalent to the statistical expectation operator over i, in which
( f )
=
f () exp( j2 f ) d = F ( f )
(5)
where f () and F( f ) are the probability density function for

and its Fourier transform.
Thus the inverse Fourier transform of the measured ( f )
is an estimate of the probability density function (PDF) for
the fiber conduction delays.
Figure 9 shows a measurement of the PDF of CV for the
biceps brachii muscle with the impulse response method proposed by Hunter et al. (22). Although still in the development
stage and with considerable improvement required, the techniques show promise and are worthy of further research
effort.
The latter two techniques obtain the CV distribution from
surface EMG, with many units contributing to an interference
signal. An alternative approach is to separate the contributing unit signals and obtain the CV for each component unit
and thus determine the distribution. This approach is based
on the use of spatial filters constructed from two-dimensional
electrode arrays which can be designed to spatially focus on
specific regions of muscle and hence separate the component
units and their velocities (14,25). Figure 10 gives an example
of the detection of motor units from surface EMG obtained
534
ELECTROMYOGRAPHY
0.4
Probability density
0.3
0.2
0.1
0
0
8
Velocity (m/s)
12
Figure 9. Results from the measurement of conduction velocity probability density function for biceps using the impulse response function
method [from Hunter et al. (22)].
trode pair placement with respect to the muscle innervation

and termination zones. If the electrode pair is positioned so
that it is near the innervation zone, a positive bias in the CV
estimation will result. Electrode pairs too close to the tendon
can also introduce nondelayed components in the EMGs,
thus, again, giving a positive bias to the CV measurement.
These errors can be reduced using different or multiple pairs
of a linear electrode array placed along the muscle fiber and
by use of the double differential electrode configuration (13).
Figure 5(b) shows that electrodes located in between two innervation zones (e.g., electrode pairs 5, 6, 7, 8) detect potentials propagating in opposite directions. They are suitable for
CV estimation only if individual motor unit action potentials
are identified. Electrical noise will introduce additional errors
in the determination of parameter values such as the peak
position of the cross-correction function, corresponding points
on the two waveform, etc. For a discussion of noise-induced
errors, see Rababy et al. (27).
CROSSTALK AND MUSCLE MOVEMENT ARTIFACTS

with an electrode array. This approach is very promising and
deserves considerable attention.
Sources of Error in CV Estimation
The most obvious source of error in CV estimation is misalignment of the electrode pair principal axis with the muscle
fibers. Misalignment produces a bias error which depends on
(a) the degree of misalignment and (b) the electrical properties of the medium. A second important source of error is elec-
Crosstalk is the component of EMG signal that is detected

above a specific muscle but is generated by another. This component may be a serious confounding factor in dynamic EMG
when the time and intensity of activation of a muscle are of
interest. One way to study it is to selectively stimulate a specific muscle and observe the signals present on the neighboring muscles (28,29). This technique, however, is appropriate for investigating the problem but not for solving it in
a clinical setting. A clinically satisfactory solution is not yet
Monopolar
Bipolar
d
d
5
2
1
2
1
3
+1
2
1
+1
+1
3 4
+1
+1
1
4
+1
Doubledifferentiated
d = 2.5 mm
NDdifferentiated
600 V
10ms
Figure 10. Surface EMG acquired with different electrode configurations showing single motor
unit signals [from Rau and Disselhorst-Klug (14)].
ELECTROMYOGRAPHY
available. The crosstalk phenomenon is strictly related to the

biophysics of volume conduction through a layered medium
(anisotropic muscle, isotropic subcutaneous fat, isotropic
skin). Particular situations (e.g., a rather conducting skin on
top of an insulating layer of fat) may strongly enhance it (30).
During dynamic contractions the muscle moves under the
skin, and the geometry of the electrode-muscle system
changes. In particular, a sliding of the innervation zone near
or below a bipolar electrode set may cause a marked change
of signal amplitude that might be mistakenly interpreted as
a change of muscle activation. This possibility may be seen in
Fig. 5(a) and Fig. 5(b) for a single motor unit but may be
easily generalized for many motor units innervated in the
same location. For example, the innervation zone of the first
two motor units in Fig. 5(b) is between electrode pairs 8 and
9. A sliding of 5 to 10 mm would bring the zone under electrode pairs 7 or 8 with marked decrement of the EMG amplitude detected by such electrodes and an increment of the
signal detected by pair 9. This sliding may happen cyclically
during a repetitive movement, and its effect may be misinterpreted as a change of muscle activation level.
Other artifacts are due to a relative movement between
electrodes and skin or to large changes of electrode contact
impedance. Automatic detection and correction of these confounding factors is an engineering challenge. Signal filtering
is usually not sufficient, and adaptive techniques are being investigated.
MOVEMENT AND GAIT ANALYSIS

During movements of the human body the muscles involved
in moving the joints are activated by the brain according to
specific patterns. In particular, the pattern of activation of
back muscles during lifting or leg muscles during gait has
great relevance in clinical evaluation (6,8,31,32). A simple
model of dynamic EMG assumes the signals to be Gaussian
zero-mean noise modulated by a function representing the intensity of activation. Such a function may be estimated by
amplitude demodulation of the EMG, that is, by rectification
and smoothing, a process often referred to as linear envelope
detection. The choice of the smoothing filter is critical: on-line
analog filtering with equivalent time constants ranging from
25 ms to 130 ms have been used as well as digital noncausal
FIR filters with symmetric coefficients and zero phase shift,
producing very different results. Indeed the analog filters
time constant or the impulse response length of the digital
filter should automatically adapt to the local properties of the
signal in order to track equally well fast and slow variations
without introducing significant delays (32,33). Figure 11
shows an example of dynamic EMG recording from a group of
muscles during normal gait. Signals are detected with bipolar
surface electrodes. It is evident that the identification of the
on/off timing of the muscles and of their intensity of activation is not an easy task. It should also be considered that
many additional confounding factors, besides the level of muscle activation, affect the EMG amplitude. Among these are
crosstalk, the thickness of the subcutaneous layers, the location of the electrodes, and the amount of muscle shortening
and sliding under the electrodes. It is evident that these factors mask the relationship between level of muscle activation
or contraction force on one hand and the EMG amplitude on
535
the other. Cyclical activities (such as gait) allow averaging of

repeating patterns and the attenuation of those factors that
are not synchronized with the pattern (8).
The analysis of nonrepetitive movements is more complex.
For example, it is known that improper shoulderneck muscle load in the workplace is a major factor for developing musculoskeletal disorders. Surface EMG techniques have been
used to evaluate muscle involvement during occupational
work. Such evaluation, as well as gait analysis, requires some
form of normalization and standardization. The problem is
discussed in detail in the recent review by Mathiassen et al.
(34).
EMG CONTROL FOR POWERED LIMB PROSTHESES
Motor information from the CNS is present in the electrical
signals from nerves (ENG) and residual muscle fibers (EMG)
of both traumatic and congenital amputees, and it can be acquired and used for the control of powered limb prostheses. A
major challenge in this application is the extraction of the
control information from the available EMG. The CNS can
vary (a) the number of motor units and their respective firing
rates and (b) the combination or pattern of activated muscles.
These variables give the surface EMG certain characteristics
(power, spectral content, etc.) and an EMG processor must,
with appropriate algorithms, measure these characteristics
and in turn decide which prosthesis function to select in order
to satisfy the intent of the CNS.
The number of units and their firing rates, along with the
pattern of muscle activity above the level of amputation, are
reflected in the average rectified value (ARV) and the local
waveform or spectral content, respectively, of the EMG. The
type and characteristics of the EMG processor used for prosthesis control depend, in part, on whether EMG ARV, waveform, or both form the basis of control. A block diagram showing the relationship of the EMG control system to the normal
system is given in Fig. 12. EMG prosthesis controllers can,
from the point of view of their EMG processors, be classified
as either multistate or pattern based. Extensive reviews of
EMG control systems for powered limb prostheses are given
in Refs. 35 and 36.
Multistate EMG Controllers
The multistate controller subdivides the EMG amplitude or
ARV range into a number of levels or states and assigns a
particular prosthesis function to each statea form of amplitude modulation. To select a function the operator generates
EMG with the appropriate ARV. The processor estimates this
ARV and, on the basis of the estimate, decides which function
to operate. The ARV estimator typically consists of a fullwave rectifier followed by a first-order low-pass filter. The
ARV estimation error will, with a predictable probability,
cause some decision errors in function operation. The decision
error can be reduced by increasing the time constant of the
filter, but only at the expense of prosthesis dynamic response.
Commercial systems are now available and for the most
part the number of states is two or three. The two-state controller is used to switch on/off a given degree-of-freedom, and
the three-state controller is used to select one of two directions of a given degree-of-freedom. In order to provide more
functionality for the user (i.e., more prosthesis degree-of-free-
536
ELECTROMYOGRAPHY
doms) it is necessary with current systems to either have parallel two- or three-state controllers (one for each degree-offreedom) or have a single controller with as many states as
required. Unfortunately, it has been found that multistate
systems with more than three states have unacceptable error
performance due to excessive demand put on the operators
ability to reliably generate the appropriate EMG, and hence

are neither practical nor commercially available.
EMG Pattern Controllers
In order to increase the number of prosthesis functions without putting excessive demand on the operator, control strate-
TA
SOL
MG
LG
VAM
SM
BF
GLM
Figure 11. Example of surface EMG detected from a group of muscles during gait of a normal
human subject. TA, tibialis anterior; SOL, soleus; MG, medial gastrocnemius; LG, lateral gastrocnemius; VAM, vastus medialis; SM, semitendinosus; BF, biceps femoris; GLM, gluteus maximus
(courtesy of Dr. Carlo Frigo, Centro di Bioingegneria Politecnico di Milano and Fondazione Don
Gnocchi).
ELECTROMYOGRAPHY
CNS
Muscle
Joint
Control
Battery
537
Output
Prosthesis
Figure 12. Block diagram showing relationship between normal and myoelectric control systems
(shaded area is removed in amputation surgery) [from Parker and Scott (36)].
gies have been developed based on the EMG signals present

in normally occurring activation patterns of agonist/antagonist group of muscles. The essence of this approach is to select
from the EMG an appropriate feature set whose values are
(a) repeatable for a given muscle group activity pattern and
(b) sufficiently different across patterns to allow for reliable
classification.
The early work in this direction used several EMG channels, and the feature set consisted of the binary 1 or 0 depending on whether or not a channels EMG activity exceeded
a threshold. The use of multiple EMG channels causes significant electronic hardware reliability problems and difficulties for the prosthetist in prosthesis fabrication. Thus the
single EMG channel has become the goal and standard for
pattern-based EMG multifunction controllers. That recognizable and repeatable EMG patterns can be obtained from a
single channel is due to the differing contributions by individual agonist/antagonist muscles to the total channel EMG during different muscle activation patterns. Graupe et al. (37)
were among the first to demonstrate this, and others have
followed up to produce a new generation of EMG controllers.
Not only is the single-channel system more reliable and easily
fabricated, but it also has reduced size and cost. The reliability and size advantages were significant factors in the success
of self-contained limb prostheses. Function selection performances of 80 to 90% have been obtained for five-function controllers.
Other recent developments have been in the direction of
multifunction EMG controllers which allow simultaneous operation of several functions. The major challenge in this work
is to obtain sufficient simultaneous and preferably independent EMG inputs which can be generated without excessive
burden on the operator. To this end, artificial neural networks
(ANN) have proven very successful in EMG pattern classification, and they are of particular significance in this application because of their trainability, adaptability, and robustness.
The EMG spectral components of a single-channel EMG
can form a feature vector as input to an ANN, and the ANN
can be trained under supervision to recognize members of a
set of spectra corresponding to the functions of a multifunction prosthesis (38). Alternatively, a time-domain feature set
can be used as the ANN input. Hudgins et al. (39) have demonstrated that the initial 300 ms of dynamic contraction
EMGs contain deterministic components that are repeatable
in time and which differ over contraction functions (see Fig.
13). Thus time samples from these deterministic components
can form feature sets for an ANN trained to classify by function. A 30:8:4 perceptron ANN-based controller is used in this
application and is currently implemented on a TMS320 DSP
micro for clinical testing.
An approach to prosthesis control with significant promise
is to estimate from EMGs the biomechanical variables of a
joint model and to drive the mechanical prosthesis accordingly [see Wood et al. (40)]. Such an approach can incorporate
stiffness control in which the stiffness (or compliance) of the
prosthetic limb is made to match that of the limb model for a
given client.
MODELING OF EMG SIGNALS; MODEL-BASED

INTERPRETATION OF ARRAY SIGNALS
Many anatomical and physiological parameters of a muscle
are not accessible and cannot be measured directly. However,
they are accessible in a model simulating EMG signals and
EMG variables and may be changed until the simulated observable EMG variables and parameters match the experimental ones. When the matching is obtained, it is likely that
the parameters of the model have values similar to those that
cannot be measured directly from the real system. This conclusion must always be taken with caution since (a) a model
always implies approximations and simplifications that may
affect the results and (b) there may be more than one set of
model parameters that provide a good fit of the experimental
data. Motor unit action potential models have been developed
by many researchers among which are P. Rosenfalck (41) and
N. Dimitrova (42). The model described in Fig. 14(a) is based
on the work of Gootzen et al. (43) and has been used to investigate and explain some experimental findings. An application
example is provided by Fig. 14(b), which shows 10 firings of
the same MU of a healthy biceps brachii during a low-level
voluntary contraction. The signals are detected bipolarly from
a 16-contact linear array.
538
ELECTROMYOGRAPHY
2.5
MES amplitude (V 5000)
2.5
2.5
2.5
2.5
2.5
2.5
2.5
(a)
2.5
2.5
(b)
Figure 13. Initial 300 ms of EMG obtained from bipolar differential measurement with electrodes over biceps and triceps during elbow flexion. (a) Four 300 ms records and (b) the ensemble
average of sixty 300 ms records demonstrating the deterministic component of the initial phase
[from Hudgins et al. (35)].
The 10 firings are selected during a time interval of 1.5 s,

are aligned and superimposed, and are similar enough to justify the assumption that they belong to the same MU. The
results of the simulation (open circles) are superimposed and
the indicated model parameters provide an estimate for ana-
tomical features of the MU, conduction velocity, and anisotropy of the tissue. Future research might lead to the development of systems for the automatic identification of the most
likely set of parameters for individual MUs and make them
available to the neurologist for diagnostic evaluation.
ELECTROMYOGRAPHY
Current sources
DD1
2 DD signals
3 DD signals
4 Monopolar signals
I1
a
I2
I2
SD1
VA
SD2
SD3
VB
e
VC
e
I3 I3
Right tripole
I1
Left tripole
DD2
I1 + I2 + I3 = 0
I 1a + I 3 b = 0
539
VD
e
ZE
LL
LR
h
R
N
CV
WTL
CV
WI
WTR
Conductive medium: y z; y = x
(a)
: exp. data
14
: simulation
Model parameters:
12
e = 10 mm
z / r = 6
Electrode pair
10
b = 7 mm
8
a/b = 1/3
h = 4.5 mm
R = 2 mm
WI = 5 mm
WTR = 20 mm
2
WTL = 10 mm
LR = 56 mm
LL = 73 mm
0
10
15
Time (ms)
20
25
CV = m/s
(b)
BIBLIOGRAPHY
1. B. Saltin and P. Gollnick, Skeletal muscle adaptability: Significance for metabolism and performance, Handbook of Physiology,
Skeletal Muscle, Sec. 10, Chap. 19, American Physiological Society, pp. 555631, 1983.
2. M. Smorto and J. V. Basmajian, Clinical Electroneurography,
Baltimore: Williams and Wilkins, 1979.
3. J. Duchene and F. Goubel, Surface electromyogram during voluntary contraction: Processing tools and relation to physiological
events, CRC Crit. Rev. Biomed. Eng., 21: 313397, 1993.
4. R. Merletti, M. Knaflitz, and C. J. DeLuca, Electrically evoked
myoelectric signals, CRC Crit. Rev. Biomed. Eng., 19: 293340,
1992.
5. E. Kupa et al., Effects of muscle fiber type and size on EMG median frequency and conduction velocity, J. Appl. Physiol., 79:
2332, 1995.
Figure 14. (a) Model for the simulation of surface

EMG signals and of their variables. Schematic
structure of the model of a single motor unit. The
motor unit has N fibers uniformly distributed in a
cylinder of radius R at depth h. The axis of this
cylinder may present an angle with respect to the
skin plane and with respect to the z axis. The neuromuscular junctions are uniformly distributed in
a region WI, and the fibertendon terminations
are uniformly distributed in two regions WTR and
WTL. A right and a left current tripole originate
from each neuromuscular junction and propagate
to the fibertendon termination, where they become extinguished. The conduction velocity is the
same in both directions and for all fibers of a motor unit but may be different in different motor
units. Each of the voltages VA, VB, VC, and VD is
the summation of the contributions of each tripole. (b) Example of simulation of 10 superimposed firings of a motor unit detected during a
low-level contraction of a healthy biceps brachii
muscle with a linear 16 contacts array. Pair 14 is
proximal, pair 0 is distal.
6. C. J. DeLuca, The use of the surface EMG signal for performance
evaluation of back muscles. Muscle Nerve, 16: 210216, 1993.
7. J. Perry, Gait Analysis, Thorofare, NJ: Slack, 1992.
8. R. Kleissen and G. Zilvold, Estimation uncertainty in ensemble
average surface EMG profiles during gait. J. Electromy. Kinesiol.,
4: 8394, 1994.
9. B. Hudgins, P. Parker, and R. Scott, Control of artificial limbs
using myoelectric pattern recognition, Med. Life Sci., 12: 2138,
1994.
10. J. Basmajan and C. J. De Luca, Muscles Alive, Baltimore: Williams and Wilkins, 1995.
11. Various authors, Selected Topics in Surface Electromyography for
Use in the Occupational Setting ; Expert Perspectives, US Dept. of
Health and Human Services, National Institute of Occupational
Safety and Health, NIOSH Publication Number 91-100, March
1992.
540
ELECTRON AND HOLE MOBILITY IN SEMICONDUCTOR DEVICES
12. J. Desmedt (ed.), Computer Aided Electromyography and Expert

Systems, Amsterdam: Elsevier, 1989.
13. H. Broman, G. Bilotto, and C. DeLuca, A note on the non-invasive
estimation of muscle fiber conduction velocity, IEEE Trans. Biomed. Eng., 32: 341344, 1985.
14. G. Rau and C. Disselhorst-Klug, Principles of high spatial resolution single motor unit detection and the application to the diagnosis of neuromuscular disorders, J. Electromyogr. Kinesiol. (in
press).
15. R. Merletti and L. R. Lo Conte, Surface EMG signal processing
during isometric contractions, J. Electromyogr. Kinesiol. (in
press).
16. J. K-F. Ng and C. A. Richardson, Reliability of electromyographic
power spectral analysis of back muscle endurance in healthy subjects, Arch. Phys. Med. Rehabil., 77: 259264, 1996.
17. S. Roy et al., Classification of back muscle impairment based on
the surface electromyographic signal, J. Rehabil. Res. Dev., 34:
405414, 1997.
18. L. Oddsson et al., Development of new protocols and analysis procedures for the assessment of LBP by surface EMG techniques,
J. Rehabil. Res. Dev. 34: 405414, 1997.
19. E. Stalberg, Propagation velocity in human muscle fibers in situ,
Acta Physiol Scand., 287: 3112, 1966.
20. W. Troni, R. Cantello, and I. Rainero, Conduction velocity along
human muscle fibers in situ, Neurology, 33: 14531459, 1983.
21. M. Naeije and H. Zorn, Estimation of the action potential conduction velocity in human skeletal muscle using the surface EMG
cross-correlation technique, Electromyogr. Clin. Neurophysiol.,
23: 7380, 1983.
22. I. W. Hunter, R. E. Kearney, and L. A. Jones, Estimation of the
conduction velocity of muscle action potentials using phase and
impulse response function techniques, Med. Biol. Eng. Comput.,
25: 121126, 1987.
23. L. Arendt-Nielsen and M. Zwarts, Measurement of muscle fiber
conduction velocity in humans: Techniques and applications, J.
Clin. Neurophysiol., 6: 173190, 1989.
24. S. Davies and P. Parker, Estimation of myoelectric conduction
velocity distribution, IEEE Trans. Biomed. Eng., 34: 365374,
1987.
25. T. Masuda, H. Miyano, and T. Sadoyama, A surface electrode
array for detecting action potential trains of single motor units,
Electroencephalogr. Clin. Neurophysiol., 60: 435443, 1985.
26. W. Li and K. Sakamoto, Distribution of muscle fiber conduction
velocity of Biceps Brachii during voluntary isometric contraction
with use of surface array electrodes, J. Physiol. Anthropol., 15:
4153, 1996.
27. N. Rababy, R. Kearney, and I. Hunter, Method for EMG conduction velocity estimation which accounts for input and output
noise, Med. Biol. Eng. Comput., 27: 125129, 1989.
28. C. J. De Luca and R. Merletti, Surface myoelectric signal crosstalk among muscles of the leg, Electroencephalogr. Clin. Neurophysiol., 69: 568575, 1988.
29. M. Solomonow et al., Surface and wire EMG crosstalk in neighboring muscles, J. Electromyogr. Kinesiol., 4, 131142, 1994.
30. K. Roeleveld et al., Volume conduction models for surface EMG;
confrontation with measurement, J. Electromy. Kinesiol., 7: 221
232, 1997.
31. H. Hermens, K. Boom, and G. Zilvold, The clinical use of surface
EMG, J. Electromyogr. Clin. Neurophysiol., 24: 243265, 1984.
32. R. Shiavi, J. Bourne, and A. Hooland, Automated extraction of
activity features in linear envelopes of locomotor electromyographic patterns, IEEE Trans. BME, 33: 594599, 1986.
33. T. DAlessio, Analysis of digital EMG signal processor in dynamic
conditions, IEEE Trans. BME, 32: 7882, 1984.
34. S. E. Mathiassen, J. Winkel, and G. Hagg, Normalization of surface EMG amplitude from the upper trapezius muscle in ergonomic studies: A review, J. Electromy. Kinesiol., 5: 197226, 1995.
35. B. Hudgins, P. Parker, and R. Scott, Control of artificial limbs
using myoelectric pattern recognition, Med. Life Sci. Eng., 13:
2138, 1994.
36. P. Parker and R. Scott, Myoelectric control of prostheses, CRC
Crit. Rev. Biomed. Eng., 13: 283310, 1986.
37. D. Graupe, J. Salahi, and K. Kohn, Multifunction prosthesis and
orthosis control via microcomputer identification of temporal pattern differences in single-site myoelectric signals, J. Biomed.
Eng., 4: 1722, 1982.
38. M. Kelly, P. Parker, and R. Scott, The application of neural networks to myoelectric signal analysis: A preliminary study, IEEE
Trans. Biomed. Eng., 37: 221230, 1990.
39. B. Hudgins, P. Parker, and R. Scott, A new strategy for multifunction myoelectric control, IEEE Trans. Biomed. Eng., 40: 82
94, 1993.
40. J. Wood, S. Meek, and S. Jacobson, Quantitation of human shoulder anatomy for prosthetic arm control, J. Biomechanic., 22: 273
292, 1989.
41. P. Rosenfalck, Intra- and extracellular potential fields of active
nerve and muscle fibers. A physico-mathematical analysis of different models, Acta Physiol Scand. Suppl. 321: 1168, 1969.
42. N. Dimitrova, Model of the extracellular potential field of a single
striated muscle fiber, Electromy. Clin. Neurophysiol., 14: 5366,
1974.
43. T. Gootzen, D. Stegeman, and A. Van Oosterom, Finite limb dimensions and finite muscle length in a model for the generation
of electromyographic signals, Electroencephalogr. Clin. Neurophysiol., 81: 152162, 1991.
ROBERTO MERLETTI
Politecnico di Torino
224
LASER APPLICATIONS IN MEDICINE

Lasers have been investigated for possible use in medical applications ever since their discovery in the 1960s. Early applications of lasers in medicine were reported in ophthalmology
(13), surgery (4), neurology (5,6), and dermatology (7). In recent years, numerous additional applications of lasers in medicine were the direct result of an improved understanding
about how laser radiation interacts with biological tissues,
the rapid progress in the manufacturing of laser sources at
selective wavelengths that are absorbed by different biological
tissues, and improved optical fiber and lens technologies for
the efficient delivery of laser radiation.
Clinical applications of lasers can be divided into two major categories: therapeutic and diagnostic. Because of the
many therapeutic uses of lasers in medicine, it is further possible to group the applications into surgical and nonsurgical
procedures. Surgical applications involve direct removal of
tissues. Nonsurgical applications include alternative methods
to stimulate peripheral nerves (laser biostimulation) and to
elevate local tissue temperature (hyperthermia), as well as
photodynamic therapy.
225
LASER PHYSICS
The primary components of a laser (which is an acronym for
Light Amplification by Stimulated Emission of Radiation) are
(Fig. 1): (1) a lasing medium, which may be in a solid, liquid,
or gas phase, capable of undergoing stimulated emission; (2)
an excitation mechanism, which causes the atoms or molecules of the lasing medium to ionize and rise to a higher electronic energy state by absorbing either electrical, thermal, or
optical energy (this process results in a condition known as a
population inversion where more atoms have electrons at a
higher energy state than at a lower energy level); and (3) a
positive feedback system that consist of a highly reflective
curved mirror and a partially transmitting flat mirror causing
the spontaneous photon emission from the active medium to
bounce back and forth between the two reflecting mirrors.
The collision between the spontaneous emission and the
atoms or molecules in the excited state stimulate additional
emission (stimulated emission) inside the active laser cavity
or optical resonator. If the frequencies are properly chosen,
the light will be amplified and emitted along the axis of the
resonator as an intense narrow beam. The result of the stimulated emission is two electromagnetic waves of the same
wavelength traveling parallel and in phase (spatial and temporal coherence) with one another. In contrast to gas-filled
lasers, solid-state lasers, such as the Nd:YAG and ruby lasers,
require an external optical light source as an excitation
source to pump the atoms in the solid-state crystal.
Lasers produce a beam of nonionizing radiation (spot size
on the order of 1 m) that is highly coherent (directional),
generally monochromatic (single wavelength) and collimated
(the beam remains almost parallel along its trajectory with
minimum loss of power due to divergence). In addition, a laser
beam has a high power density, or irradiance, usually expressed in watts per square centimeters (W/cm2). If the radiation is delivered as a pulse, the pulse duration becomes an
additional factor in determining the effect on tissue because
the energy applied during the exposure, which is expressed in
Joules per square centimeter (J/cm2), is equal to the power
density times the pulse duration (also called the fluence).
When the laser is used as a cutting tool, the spot size must
be very small in order to concentrate the power into a tiny
spot. On the contrary, when a laser is used for coagulation,
the beam is defocused to allow for an increased spot size to
spread the laser beam over a large area (Fig. 2). Power densities above 1 kW/cm2 tend to be used for incisions, whereas
Resonator
or
optical cavity
Laser
beam
output
Active medium
Totally reflecting
mirror
Energy input
Partially
transmitting
mirror
Figure 1. The common components of a laser system. Energy input

can be supplied in the form of an electrical discharge current or from
a flash lamp.
Handpiece
Laser beam
Focal zone
Tissue
Defocused zone
Figure 2. A laser beam can be used for cutting (vaporization) when
the focal zone is near the surface of the tissue. For coagulation, the
laser beam is defocused causing the energy to be distributed over a
wider area.
power densities below 500 mW/cm2 are generally used for coagulation.
The transverse electromagnetic mode (TEM) is a term used
to describe the power distribution of a laser beam over the
spot area. For example, a TEM01 mode refers to a multimode
distribution and indicates that the spot has a cool region in
the center of the beam. A TEM00 mode, on the other hand,
produces a uniform Gaussian power distribution with most of
the power concentrated in the center of the beam and the rest
decreasing in intensity toward the periphery of the beam.
Q-Switched Lasers
When a laser is Q-switched, the energy that is normally
stored in the inverted atomic population is suddenly transferred to the oscillating laser cavity. The term Q is used to
convey the fact that a laser cavity is essentially a resonator
with a certain quality factor Q, similar to any conventional
electronic resonant circuit. Normally, the two highly reflecting mirrors inside a conventional laser tube bounce the
energy back and forth internally, thus leading to a high resonant Q factor. By inserting a lossy attenuator into the resonating laser cavity, it is possible to change the Q factor of
the laser considerably to a point where the energy build-up is
insufficiently high for laser oscillation to occur. Conversely, if
the attenuation is suddenly switched off, the energy builds up
inside the laser cavity so that the excess excitation can be
released in a controllable manner as a very high burst of
short-duration energy (usually nanoseconds).
Pulsed and CW Lasers
Depending on how the excitation energy inside the laser cavity is applied, the output beam could be either in the form of
a continuous wave (CW) or a pulsed wave (PW). The output
of PW lasers can vary widely depending on the duration, repetition rate, and energy of the individual pulses.
The effect of the laser on the tissue can be enhanced with
PW rather than CW delivery lasers because pulsed radiation
minimizes thermal diffusion of the energy deposited in the

tissue away from the heated zone. Short pulses of intense laser radiation can produce nonlinear absorption effects such as
optical breakdown and plasma formation. These effects can
cause significant damage to optical components or optical fibers used to deliver the laser radiation.
INTERACTION OF ELECTROMAGNETIC
RADIATION WITH BIOLOGICAL TISSUES
It is essential that the operator knows both the optical characteristics of the particular tissue being radiated and the
properties of the laser used. Accordingly, different lasers
must be used depending on the particular medical application. When a laser beam enters the tissue, it may be partially
reflected from, transmitted through, scattered within, or absorbed by the tissue (Fig. 3). The interaction of laser radiation
with biological tissue also depends on the incident angle of
the beam, the wavelength of the laser, and tissue composition
being treated.
Reflection
Light can be reflected from a target either by specular or by
diffuse reflection. Specular reflection occurs from polished
surfaces when the angle of reflection is equal to the angle of
the incident beam. On the other hand, diffuse reflectance occurs from rough unpolished surfaces as a result of wide-angle
backscattering of the light.
Scattering
Biological tissues are considered highly heterogeneous both
at the microscopic and macroscopic scales. Therefore, electromagnetic radiation entering a biological medium will deviate
from its incident direction mostly at the boundaries between
regions having different indices of refraction. The scattering
effect in turbid biological media depends on the size and
shape of the scattering particles. When particle size is smaller
than the wavelength of the incident light, we refer to this
scattering as Rayleigh scattering. The result will be a homogeneous zone of light intensity surrounding the scattering
Incident laser
radiation
Diffuse
reflection
Scattering
Tissue
Absorption
Diffuse
Transmission
Figure 3. Lasertissue interaction.
105
Absorption coefficient (cm1)
226
104
103
Melanin
102
101
Hemoglobin
100
101
102
Protein
Water
103
104
0.1
1.0
Wavelength ( m)
10
Figure 4. Laser absorption in biological tissue.
particle. On the other hand, when the scattering particles are

larger than the wavelength of radiation, the resulting radiation, also referred to as Mie scattering, generally continues in
the forward orientation along the direction of the incident
beam. Because scattering spreads the laser beam laterally
where it is ultimately converted to heat, it can limit the precision with which certain tissue structures can be destroyed
without causing damage to adjacent tissues. Therefore, potentially wide-spread tissue damage can result unless the target
tissue has specific pigments that absorb the laser light before
it reaches the surrounding tissue structures. From a practical
point of view, obtaining a precise spot size for cutting or coagulation can sometimes be difficult because scattering in tissue
also causes laser radiation to spread in different directions
around the point of contact or below the tissue surface before
it can be noticed by the surgeon.
Absorption
For surgical applications, the monochromatic property of laser radiation is particular useful when tissue absorbs only
certain wavelengths. In order for the laser light to affect a
biological medium, it must be absorbed by the tissue and either converted to heat or, in the case of ultraviolet (UV) radiation, initiating a photochemical decomposition process without generating heat.
The absorption of electromagnetic radiation by biological
tissues arises from the wavelength-dependent resonant absorption by the tissue components (Fig. 4). For example, hemoglobin and red blood cells reflect the red light produced by
a ruby laser but absorb the blue/green light from an argon
laser. Likewise, melanosomes, the granular pigments in the
skin, act as absorbing chromophores for wavelengths in the
UV and visible range.
In the absence of scattering, absorption results in an exponential decrease in light intensity according to LambertBeers Law. With appreciable scattering, this decrease in incident intensity is not monotonic. The relative contributions of
absorption and scattering will stipulate the depth in a given
biological tissue at which the resulting tissue effects will be
present.
EFFECT OF LASER RADIATION ON BIOLOGICAL TISSUES
The effect of laser radiation on biological tissues depends on
the wavelength of the beam, the power density, the duration
Coagulation/necrosis
zone
Laser beam
Hyperthermia
zone
227
structures in tissues, typically to an average depth of about 2

to 5 mm. Therefore, it is able to coagulate a larger volume of
blood vessels at the vicinity of an incision, thereby providing
better homeostasis.
Ablation/vaporization
zone
LASER DELIVERY SYSTEMS

The most common methods to guide a laser beam to the biological target site typically fall into three categories: (1) a
fixed optical delivery system, (2) a flexible fiber optic-based
delivery system, and (3) a semirigid mechanical system based
on an articulated arm (Fig. 6).
Figure 5. Thermal zones inside biological tissue.
of the exposure, and the amount of energy absorbed and then

released by the tissue. The common way to control power density during surgery is by adjusting the spot size of the laser
beam. Spot size can be varied by adjusting the distance between the focusing lens in the hand piece and the treated
tissue. A 2-fold decrease in the spot size will produce a 4-fold
decrease in the power density.
Energy release from water, hemoglobin, and melanin pigments in tissue is by molecular vibrations resulting in a local
temperature increase. Between 50 and 100C, this heating energy causes proteins, enzymes, and collagens to denature or
coagulate within a few seconds. Above 100C, photothermal
vaporization or ablation takes place. At this temperature, water boiling and evaporation generates gaseous decomposition
products, followed by carbonization of the dry tissue. Tissue
evaporation can take place when sufficient energy is generated either to disrupt carbon bonding or when the local temperature causes the water in tissue to reach a boiling point
leading to localized micro explosions and consequently rupturing of the fine tissue structure. Because tissue rupturing
and protein coagulation occur simultaneously, the tissue can
be cut and coagulated at the same time leading to nearly
blood-free dissections (Fig. 5). When diseased tissue is destroyed and removed, the surrounding tissues remain unaffected. Bleeding is minimized because the laser beam seals
off the small blood vessels in the surrounding tissues. In addition, microorganisms are destroyed by the intense laser radiation, so the risk of postoperative infections is minimized considerably.
The penetration depth of laser radiation in biological tissues depends on the incident wavelength. For wavelengths in
the mid-infrared (IR) region above 2 m, most of the energy is
absorbed within a very short distance from the tissue surface;
therefore, scattering becomes insignificant. For example, 10.6
m IR radiation produced by a CO2 laser passes only a short
distance of about 0.1 mm before it is completely absorbed by
soft tissues and converted to heat. Because this radiation is
concentrated in a relatively small tissue volume near the
point of impact, the CO2 laser is widely used to produce sharp
and rapid incisions.
Near-IR and visible range laser radiation penetrate deeper
into soft tissue compared to mid-IR radiation. For comparison, radiation produced by an argon laser, which is heavily
absorbed by the hemoglobin in blood, typically penetrates an
average distance of about 0.5 mm. Energy from a Nd:YAG
laser, on the other hand, can readily penetrate into deeper
Fixed Delivery Systems

Fixed laser delivery systems are used to deliver the energy to
the target by a joystick or similar control mechanism over a
relatively limited movement range. They are typically found
in ophthalmic laser systems that are connected to a slit lamp
to permit the surgeon a clear view of the operating site.
Fiber Optics Delivery Systems
The energy generated by a laser can be delivered to the operating site via a flexible fiber optic cable. The laser beam is
Mirror joints
;;
Handpiece
Mirror
Laser console
Hollow Laser
tube beam
(a)
Cladding
Fiber optic
cable
Laser console
Handpiece
(b)
Figure 6. Typical medical laser delivery systems. (a) Articulated
arm. (b) Fiber optic coupling.
228
(a)
(b)
Figure 7. Different laser probe geometries can generate different

power densities. (a) A general-purpose elongated tip can be used for
precise cutting with maximal lateral coagulation. (b) A round probe
is used for wider vaporization including coagulation.
used technique to deliver the radiation and position the laser

beam onto the desired tissue spot. Efficient coupling and direction of the energy from the laser source to the distal end
is accomplished by high reflectivity multilayered mirrors and
lenses that are positioned near the distal end of the arm to
concentrate the beam into a small spot. These mirrors are
mounted inside tubular sections and are precisely aligned to
help guide the laser beam during the manipulation of the arm
by the surgeon. Some articulated arms can also be attached
to operating microscopes that are used to visualize the operating field in delicate surgical procedures.
Articulated arms are used for wavelengths above 2.1 m
and for high laser output powers capable of damaging delicate
optical fibers. A HeNe laser beam is usually used as a coaxial aiming (aligning) guide alongside an invisible beam (e.g.,
from a CO2 laser) to direct IR radiation to the target tissue.
TYPES OF LASERS USED IN MEDICAL APPLICATIONS
introduced into the proximal end of the fiber and is transmitted along the fiber length through a process known as total
internal reflection. Flexible fiber optic guides for laser beam
delivery are usually made of silica (SiO2) glass and some additives. These fibers are commonly used for visible wavelengths
as well as near-infrared wavelengths below 2.1 m and can
deliver a relatively low output power. For example, argon and
YAG laser energy can be transmitted through glass optical
fibers with very little loss. Wavelengths above 2.1 m are
usually absorbed by the fused silica or by water impurities
inside the fiber core. High-power UV radiation produced for
example by excimer lasers is difficult to transmit by optical
glass fibers. Therefore, regular glass fibers are used only for
light transmission in the visible region of the spectrum.
The advantages of fiber optic guides, compared to the more
traditional articulating arms or fixed delivery systems, is
their small size, flexibility, improved manipulation, low
weight, and reduced cost. The ability to launch high optical
power into small optical fibers extends the clinical application
of lasers considerably. It permits precisely controlled energy
delivery through flexible endoscopes to remote intravascular
or intracavitary regions of the body where conventional surgical procedures would otherwise be very invasive to perform.
Recently, new polycrystalline fiber optic materials are being evaluated for short-length delivery of the longer IR wavelengths. Examples include Al2O3 fibers for transmission of
CO2 laser energy and silver halide fibers for use with
Er:YAG lasers.
Fiber optic systems used to deliver laser light culminate in
either hot tip (contact) interaction or free beam (noncontact) interaction, depending on what happens to the beam at
the end of the fiber. If the beam is absorbed by the fiber tip
or a tip affixed to the fiber distal end, it is called a hot tip; if
the beam is directed out of the fiber tip and travels a short
distance before it interacts with the tissue, it is considered a
free beam. Tips come in a wide variety of designs. Some tips
are intended to focus the beam, other are shaped to spread
the beam over a wider area (Fig. 7).
Articulated Arms
The multisegmented adjustable articulated arm, coupled to a
detachable lensed handpiece or endoscope, is the most widely
Many specialities in medicine are using or exploring the use

of lasers as alternative treatments to conventional surgical
procedures. The use of each laser depends on the specific medical application. Some widely used methods to classify medical lasers are based on their active medium, emission wavelength, or intended applications (Table 1).
Gas Lasers
Gas lasers are available from the far UV (excimer) through
the visible (HeNe) to the infrared (CO2). The basic types of
gas mixtures used in lasers include HeNe, Ar, Kr, CO2, and
the rare-gas excimer lasers. Gas lasers are relatively inexpensive but are bulky and can be fragile.
Table 1. Classification Criteria for Medical Lasers

Laser
Excimer:
Arf
KrCl
KrF
XeCl
XeF
Heliumcadmium
Argon
Copper vapor
Gold vapor
Heliumneon
Krypton
Ruby
Alexandrite
Diode
Thulium : YAG
Holmium : YAG
Erbium/YAG
Neodimium : YAG
Carbon dioxide
Wavelength
(nm)
Color
193
222
248
308
351
325
488
514
511
578
627
633
531
568
647
694
720800
6601500
2010
2120
2940
1064
1318
10,600
Ultraviolet
Ultraviolet
Ultraviolet
Ultraviolet
Ultraviolet
Ultraviolet
Blue
Green
Green
Yellow
Red
Red
Green
Yellow
Red
Red
Near-IR
Near-IR
Near-IR
Near-IR
Near-IR
Near-IR
Near-IR
Mid-IR
229
Dye Lasers
Neodymium-Doped Lasers
Dye lasers are more complex than gas lasers and require optical pumping either by an intense flash lamp or by another
laser because the laser action is not very efficient. They typically use flowing organic dyes and operate in the visible optical spectrum. Dye lasers operate over a broad wavelength
range and have relatively low efficiency. Dye lasers are useful
mostly in photodynamic therapy (PDT) and in retinal photocoagulation.
Neodymium-doped YAG (Nd:YAG) solid-state lasers, introduced in 1961 (11), produce 1.064 and 1.32 m (near infrared)
radiation. These lasers are widely employed when heavy coagulation is desired or when the use of a fiber optic-based delivery system is preferred.
Carbon Dioxide Lasers

Carbon dioxide (CO2) gas-filled infrared lasers, introduced by
Patel et al. in 1964 (8), produce radiation at 10.6 m. This
radiation is highly absorbed by water, making it useful for
precise surface ablation in soft tissue applications. The CO2
laser is the most common type of laser and can be found in
almost every medical specialty except ophthalmology and
PDT. It is used extensively in general surgery to remove polyps, warts, cysts, and tumors and to cut through heavily vascularized tissues where fast coagulation is desired.
Helium-Neon Lasers
Helium-neon (HeNe) lasers, first developed in 1961 (9), produce a dominant spectral line at a red wavelength of 632.8
nm. Standard HeNe lasers offer powers between 1 and 50
mW. HeNe lasers are used primarily for positioning patients
in radiation therapy, for aiming invisible IR laser beams in
surgery, and in laser Doppler flowmetry.
Excimer Lasers
Excimer lasers are based on the ionization of inert gas molecules, such as xenon, argon, or krypton combined with halogen molecules, such as fluorine or chlorine, to generate a
source of high-power UV radiation. Examples of excimer lasers include XeF, which provides a source of 351 nm radiation; XeCl, which is lasing at 308 nm; KrF, which lases at 248
nm; and ArF, which is lasing at 193 nm. These lasers have a
relatively shallow depth of penetration into soft tissues, making them available for very delicate surgical procedures like
the removal of occluding atherosclerotic plaques (thrombus)
inside the vascular system.
An excimer argon fluoride (ArF) laser produces radiation
at 193 nm. This radiation predominantly causes tissue ablation by a photochemical process because short wavelength radiation have sufficient energy to break molecular bonds.
Pulsed ArF lasers enable the removal of approximately 0.2
m-thick tissue layers and are therefore useful in ophthalmology for the correction of near-sighted vision.
MEDICAL APPLICATIONS OF LASERS
Argon Lasers
Therapeutic Applications
Argon (Ar) ion gas-filled lasers, introduced by Bennett in 1962

(10), produce wavelengths in the 476.5 nm (blue) to 514.5 nm
(green) region of the spectrum. Ar lasers are used in gynecological applications for ablation of pigmented vascular tissues,
in otolaryngology for coagulation of vascular lesions in dermatologic applications, and in ophthalmology for retinal coagulation.
Ophthalmology. The application of lasers in ophthalmology

was among the earliest uses of lasers in surgery. The main
application was as photocoagulating devices that replaced the
more traditional and difficult to use xenon-arc light sources.
For example, argon lasers are widely used to treat glaucoma
(a disease leading to elevated intraocular pressure) by drilling
tiny holes around the trabecular network at the corners of the
eye (a procedure known as trabeculoplasty) to allow drainage
of the aqueous fluid (12). Likewise, argon and Nd:YAG lasers
are widely used for coagulation of the retina in diabetic retinopathy and the treatment of chronic simple glaucoma because the radiation can pass through the semitransparent
medium of the eye with little attenuation (13).
The cornea at the front of the eye absorbs both far UV (200
to 300 nm) and far IR (above 1.4 m) radiation. Near-UV (300
to 400 nm) radiation, on the other hand, will penetrate
through the cornea and will be almost totally absorbed by the
normal crystalline lens. Visible (400 to 700 nm) and near-IR
(700 nm to 1.4 m) radiation will penetrate all the way to the
back of the eye where the retina is located. This unique property makes the retina in the eye available for treatment but,
at the same time, poses a safety concern because it is also
highly vulnerable to damage from sufficiently intense laser
radiation (Fig. 8).
Other uses of lasers in ophthalmology involve reshaping
the corneal surface to correct refractive abnormalities and
treatment of dry eyes, which is a condition of inadequate production of tears to moisten the cornea. Excimer lasers are also
being used to correct mild to moderate near-sightness (myopia) in patients with minimal astigmatism. In this surgical
Erbium-Yttrium-Aluminum-Garnet Lasers
Erbium (Er) has been used to dope a yttrium-aluminum-garnet (a crystal composed of aluminum and yttrium oxides) and
form a Er:YAG laser. This solid-state laser emits a 2.94 m
and can be used for ablation of tooth enamel and dentin, for
corneal ablation, and more recently also in cosmetic skin resurfacing surgery.
Holmium-Doped Lasers
Holmium (a rare earth element)-doped YAG (Ho:YAG) lasers
emit pulses of 2.1 m wavelength typically with energies below 4 J. They are being used by orthopedic surgeons for soft
tissue ablation in arthroscopic joint surgery and in some urologic applications because it can be used in fluid environments.
Ruby Lasers
Q-switched ruby (Cr:Al2O3) lasers, which emit pulses of 694.3
nm (red) wavelength and can deliver up to 2 J in energy, are
used in dermatological applications to disperse different tattoo pigments and various nonmalignant pigmented lesions.
230

Sclera
Lens
Retina
Iris
Fovea
Visible and
near-infrared
(4001400 nm)
radiation
Macula
Vitreous
body
Cornea
Optic
nerve
Aqueous
humor
Sclera
Lens
Retina
Iris
Fovea
Mid-infrared and
far-infrared
(1400 nm to 1nm)
radiation
Macula
Vitreous
body
Cornea
Optic
nerve
Aqueous
humor
Sclera
Lens
Retina
Iris
Fovea
Near-ultraviolet
(315390 nm)
radiation
Macula
Vitreous
body
Cornea
Aqueous
humor
Optic
nerve
Figure 8. Laser absorption and penetration in different parts of the

human eye.
procedure, termed photorefractive keratectomy (PRK), the

laster ablates a predetermined curvature in the cornea and
thus alters the refraction of the eye. Directing laser light into
the eye requires delicate delivery devices that must provide a
means for illuminating and visualizing the treatment site and
protecting the surgeon from potential harmful reflected laser
light. Modified slit-lamps equipped with micromanipulator
positioners are commonly used in laser ophthalmology because they provide improved illumination and more stable
aiming of the laser beam.
Ophthalmologists have focused the radiation from Qswitched YAG lasers onto the posterior part of the lens in the
eye and cut away the thin membrane coating of the lens,
which becomes cloudy with age and can cause partial obstruction of vision.
HeNe lasers have also found diagnostic values in ophthalmic applications. One application of this laser is as a computerized scanning ophthalmoscope to produce very rapid images of the retina.
Urology. Nd:YAG, CO2, Ho:YAG, and Ar lasers have been

used successfully in urological applications to remove bladder
tumors and to treat urethral strictures. Other applications of
lasers beyond the more traditional usescutting, coagulation, and vaporizationinclude fragmenting certain urinary
stones (calculi) either through a small diameter flexible or
semirigid ureteroscope (14,15). In this procedure, also called
intracorporeal lasertripsy, the photoacoustic effect of the ultrashort (1 to 2 s duration at 10 Hz) bursts of laser energy
first causes a portion of the stone surface to heat up and then
creates a plasma (a microscopic cloud of rapidly expanding
electrons) that expands and collapses, generating a mechanical shock wave to fragment the yellow stones without damaging the surrounding mucosal walls of inside the ureter. Some
of the small stone fragments are passed spontaneously
through the urine.
Another common surgical laser procedure involves the
treatment of bilateral prostatic hypertrophy (BPH), a condition causing the prostate gland that encircles the neck of the
bladder and proximal urethral outflow tract to enlarge leading to voiding dysfunction in men. The laser procedure to
treat BPH usually involves the use of a noncontact flexible
fiber or a specially designed side-firing (normally a rightangled) optical fiber introduced through a cystoscope that delivers the laser energy to the hypertrophied prostate gland.
Gynecology. The CO2 gas lasers were among the first lasers used in a number of gynecological procedures. The laser
may be hand-held or used in combination with a hysteroscope, colposcope, or laparoscope. Example applications include the treatment of gynecological conditions such as cervical intraepithelial neoplasia (CIN), which is an abnormal
growth of cells in the epithelium of the cervix; vaginal and
vulval intraepithelial neoplasia (VAIN and VIN) to superficially ablate diseased areas of vaginal and vulvar tissues; and
endometriosis (bleeding from the endometrium), a painful
condition common in young women especially during menstruation (16,17).
Dermatology. Dermatologists have been using different
types of lasers to treat a number of skin conditions including
the irradiation of disfiguring birthmarks, the removal of tattoos, and the treatment of skin cancer (18). Among the most
common types of lasers used in dermatology are the argon
and ND:YAG lasers.
One application of laser radiation involves the removal of
port-wine spots, a red-purple skin mark that often occurs on
the face due to abnormal skin vasculature. A large area of
stains can be treated at a time with very little pain (19). Another application includes the effective removal of tattoos and
other pigmented lesions in the skin using a Q-switched ruby
laser operating at 694.3 nm.
Endoscopic Applications. Endoscopic applications of lasers
is common in surgery of the larynx and the gastrointestinal
tract (20,21). The two most common types of endoscopic lasers
are the argon and the Nd:YAG lasers because the wavelengths generated by these lasers can be readily transmitted
through flexible fiber optic endoscopes. Among the uses of lasers in gastrointestinal applications is to control bleeding ulcers in the stomach, colon, and esophagus.
Photodynamic Therapy. Photodynamic therapy is a photochemotherapy technique of photoactivation exogenous photosensitized drugs at specific target sites and the subsequent
selective destruction of certain tumors. Typically, a photosensitizing dye (e.g., hematoporphyrin derivative, or HpD) is first
introduced into malignant cells, which retain the dye, perhaps because of impaired lymphatic drainage or abnormal tumor vasculature. The tissue is then exposed to the incident
laser radiation. Dye lasers in the 630 nm range (e.g., argon)
are generally employed in PDT because this wavelength is
most effective in activating the photosensitizing dye. Even
though the exact mechanism involved in this process is not
fully understood, it is generally believed that this photosensitizer produces a side effect in the target cells that arises from
the highly toxic level of oxygen-free radicals that damage intracellular organelles in the malignant cells, leading to cell
death. The technique has been used mostly to treat superficial
tumors in the skin (22,23). Among the side effects of PDT is
excessive skin photosensitivity, which requires that patients
avoid direct exposure to sunlight for several weeks thereby
preventing potential sunburns.
Diagnostic Applications
Laser Doppler Velocimetry
Basic Principle. Laser Doppler velocimetry (LDV) is a relatively new clinical method for assessing cutaneous blood flow.
This real-time measurement technique is based on the Doppler shift of light backscattered from moving red blood cells
and is used to provide a continuous measurement of blood
flow through the microcirculation in the skin. Although LDV
provides a relative rather than an absolute measure of blood
flow, empirical observations have shown good correlation between this technique and other independent methods to measure skin blood flow.
According to the fundamental Doppler principle, the frequency of sound, or any other type of monochromatic and coherent electromagnetic radiation such as laser light, that is
emitted by a moving object is shifted in proportion to the velocity of the moving object relative to a stationary observer.
Accordingly, when the object is moving away from an observer, the observer will detect a lower wave frequency. Likewise, when the object moves toward the observer, the frequency of the wave will appear higher. By knowing the
difference between the frequencies of both the emitted and
the detected waves, the Doppler shift, it is possible to calculate the velocity of the moving object according to the following equation:
f = 20f cos /c
(1)
where f is the Doppler shift frequency, is the mean angle

that the incident light makes with the moving red blood cells,
c is the speed of light, f 0 is the frequency of the incident light,
and is the average velocity of the moving red blood cells.
Because the red blood cells do not move through the microcirculation at a constant velocity and light scattering leads to a
wide distribution of angles , the Doppler-shifted light contains a spectrum of different frequency components.
The Doppler shift of laser light caused by the average blood
velocity in the capillaries (around 103 m/s) is very small and
difficult to measure directly. Therefore, the frequency shifted
231
and unshifted backscattered light components from the skin

are mixed on the surface of a nonlinear photodiode. The output from the photodiode, an average dc offset voltage and a
small superimposed ac component, is amplified and band pass
filtered to eliminate low-frequency components in the range
between 10 and 50 Hz. These frequencies are attributed to
noise resulting from motion artifact and high-frequency noise
components (typically in the kilohertz range) resulting from
nonbiological noise. As the average red blood cells (RBC) velocity is increased, the frequency content of the ac signal
changes proportionally.
Assuming a constant blood flow geometry, Stern (24) proposed the following empirical relationship between the amplitude of the Doppler-shifted spectrum and the velocity of the
blood flow:

F=
2 P() d
(2)
where F is the root-mean-square (rms) bandwidth of the

Doppler power spectrum signal, is the angular frequency,
and P() is the power spectral density of the Doppler signal.
To compensate for laser light intensity, skin pigmentation,
and numerous other factors that affect the total amount of
light backscattered from the skin, the flow parameter is usually calculated by multiplying the percentage of light reflected
from the moving RBCs by the mean photodiode current,
which is a function of the average backscattered light intensity.
Instrumentation. The original light source used in LDV was
a HeNe laser (25,26). Newer systems use a much smaller and
less expensive single-mode semiconductor laser diode in the
near-infrared region around 750 to 850 nm as a light source.
These wavelengths are near the isosbestic wavelength of oxy
and deoxyhemoglobin (i.e., 810 nm) so that changes in blood
oxygenation have no effect on the measurement. Some LDV
systems are equipped with different light sources (e.g., green,
red, or near-infrared), which allow measurement from different tissue layer depths because light penetration depth is
wavelength-dependent. Typical output powers used in LDV
range from 1 to 15 mW.
In most LDV systems, the laser output is coupled through
a small focusing lens into the polished end of a flexible plastic
or silica optical fiber (25 to 1000 m core diameter), which
illuminates the blood directly in invasive measurements or
the surface of the skin in noninvasive applications. Light
backscattered from the biological media is collected either by
the same optical fiber used for illumination or by a separate
receiving fiber mounted in close proximity to the illuminating
fiber tip. A rigid probe helps to maintain the two optical fiber
tips parallel to each other and also perpendicular to the surface of the illuminated sample. Depending on the application,
a wide selection of probe geometries and sizes are available
commercially. In invasive applications, the optical fibers can
also be inserted through a catheter for measurement of flow
inside a blood vessel. In most noninvasive applications, the
flow probes are attached to the surface of the skin by a double-sided adhesive ring. Because blood perfusion is strongly
dependent on skin temperature, some LDV systems also have
probes with built-in heaters to control and monitor skin temperature.
232
Absolute calibration of an LDV instrument is inherently

difficult to obtain because blood flow in the skin is highly complex and variable. Because accurate calibration standards or
suitable physical models of blood flow through the skin do not
exist, instrument calibration is usually accomplished empirically either from an artificial tissue phantom, which is often
made out of a colloidal suspension of latex particles, or by
comparing the relative output from the laser Doppler instrument with other independent methods for measuring blood
flow.
In practice, most commercial systems express and display
the Doppler-shifted quantity measured by the instrument either in terms of blood flow (in units of milliliters per minute
per 100 g of tissue), blood volume (in milliliters of blood per
100 g of tissue), or blood velocity (in centimeters per second).
The clinical and medical research applications of LDV
range from cutaneous studies of ischemia in the legs (27) to
general subcutaneous physiological investigations related to
the response of various organs to physical (temperature, pressure) and chemical (pharmacological agents) perturbations
that can alter local blood perfusion. LDV has been used extensively in dermatology to assess cutaneous microvascular disease (28,29), arteriosclerosis, or diabetic microangiopathy; in
plastic surgery to determine the postoperative survival of skin
grafts; in ophthalmology to evaluate retinal blood flow (30,31);
and in evaluating skeletal muscles (32). To date, LDV remains mainly an experimental method. Although it has been
widely used as a research and clinical tool since the mid
1970s, LDV has not reached the stage of routine clinical application.
Fluorescence Spectroscopy
Many dyes that absorb energy can reemit some of this energy
as fluorescence. Laser-induced fluorescence emission is currently being investigated for the early detection, localization,
and imaging of normal and abnormal tissues, determining
whether a tumor is malignant or benign and identifying excessive areas of atherosclerotic plaque. One of the future goals
is to incorporate this technique into special fiber optic based
guidance systems used during ablation or laser angioplasty
particularly inside the coronary arteries.
Diffuse Reflectance and Transillumination Spectroscopy
Several methods are being developed to measure the absorption spectra of tissues illuminated by laser light. In a relatively new technique known as photon time-of-flight spectroscopy, researchers are trying to measure the temporal
spreading of very short pulses of laser light as photons undergo multiple scattering in the tissue. By measuring the
time it takes the light to travel through the tissue it is possible to estimate how much light scattering and absorption occurs. Some of these time-resolved or photon migration methods are being evaluated clinically as a potential alternative
to ionizing radiation used in X-ray mammography for early
noninvasive diagnosis of breast cancer.
LASER SAFETY
It is essential that all medical personnel using lasers have
a thorough knowledge of the potential hazards involved and
become familiar with proper methods to minimize the chances

of accidental injuries.
To ensure safety and protect the personnel in the operating
room, everyone must wear protective goggles and clothing to
protect the eyes or skin from dangerous exposure to laser radiation. The eyes are especially vulnerable because the collimated laser beam incident on the cornea will be focused to a
small and highly intense spot on the retina (33). For example,
a visible, 10 mW/cm2 laser beam could result in a 1000
W/cm2 radiant exposure to the retina. This power density is
more than enough to cause permanent damage to the retina.
Laser emission in the UV and IR spectral regions produce
ocular effects mainly at the cornea. Laser eye protection goggles must be matched with the type of laser being used because the protective properties of different goggle materials
depend on wavelength and laser output intensity. To keep
stray laser radiation at a minimum and to protect the surgical
staff from potential exposure to misdirected laser beams or
stray laser light, special nonreflecting matte surfaces should
be used. CO2 wavelengths can be attenuated considerably by
ordinary clear glass or plastic goggles. For other types of lasers, such as argon or YAG, a colored blocking material must
be used to protect the eyes. This could also become a limitation to the surgeon because the operating site must be viewed
through the same color-tinted goggles that are used to protect
the eyes.
In addition to optical safety, using high-energy lasers near
combustible materials or flammable anesthetic gases may
cause ignition and potential explosions. Several incidences
have been reported where lasers have caused inadvertent explosions inside endotracheal tubes by igniting the inspired oxygen gas used during anesthesia.
Laser safety standards for health care facilities are available from the American National Standards Institute (ANSI)
(34). The hazard standards published in 1993 by ANSI (e.g.,
series Z136.1 laser safety standards) list the maximum permissible exposure limits for different medical lasers and provide a detailed description of control measures to protect
against potential hazards. These safety standards are also
referenced by the Occupational Safety and Health Administration (OSHA) and other U.S. safety agencies as the basis of
evaluating occupational safety hazards from the use of medical lasers. In addition, medical lasers are also grouped and
classified according to four major hazard categories. In the
United States, manufacturers must certify that all medical
lasers meet the regulatory standards issued by the Food and
Drug Administration (FDA). Each laser must bear a clearly
marked label indicating compliance with the appropriate
standard and denoting the laser hazard according to one of
four general classifications. The higher the classification number, the greater the potential hazard.
Besides the risks associated with the exposure to harmful
laser light, it is important to recognize also that many surgical lasers are equipped with a localized exhaust ventilation
system to remove potentially hazardous airborne contaminants and effluent fumes generated during surgery. This risk
can result for example from CO2 lasers used for incision, ablation, and vaporization of tissues.
Additional hazards with some medical lasers are related to
the use of toxic gases and dyes. For example, excimer lasers
employ a toxic halogen gas, which must be kept in well-ventilated cabinets.
Most laser systems are equipped with a standby mode that

prevents the beam from being emitted accidentally. A master
key is often used to lock the control panel of most laser systems for added safety. Foot pedals are widely used to activate
the laser delivery system. In most systems, a shutter mechanism is used to control the laser energy output that is delivered to the final destination.
233
7. H. Solomon et al., Histopathology of the laser treatment of port

wine lesions, J. Invest. Dermatol., 50(2): 141146, 1968.
8. C. K. N. Patel, R. A. McFarlane, and W. L. Faust, Selective excitation through vibrational energy transfer and optical maser action in N2CO2, Physiol. Rev., 13: 617619, 1964.
9. A. Javan, W. R. Bennett, and D. R. Herriott, Population inversion
and continuous optical maser oscillation in a gas discharge containing a HeNe mixture. (Letter), Physiol. Rev., 6: 106110, 1961.
LASER CLASSIFICATIONS
10. W. R. Bennett et al., Dissociative excitation transfer and optical

maser oscillation in NeO2 and ArO2 rf discharges (Letter), Phys.
Rev. Lett., 8: 470473, 1962.
Lasers are classified according to their hazard potential,

which depends on their optical wavelength and output power.
A brief description of each classification follows.
11. L. F. Johnson, Optical maser characteristics of rare-earth ions in

crystals, J. Appl. Physiol., 34: 897909, 1961.
Class 1 Lasers
Class 1 laser products produce very low power (e.g., semiconductor diode lasers used in video disc players); they pose no
known hazard under normal operating conditions.
Class 2 Lasers
Class 2 laser products produce low-power visible light, normally used for brief periods during alignments (e.g., in radiology). These lasers are therefore considered safe for momentary viewing (0.25 s or less) unless an individual stares
directly into the laser beam.
Class 3 Lasers
Class 3 laser products generally consist of medium-power lasers that pose potential hazards to a person during instantaneous exposure of the eyes. This classification is further subdivided into two subclassifications: Class 3A and Class 3B.
Subclassification 3A lasers emit visible light with an average
output power between 1 and 5 mW. Lasers that deliver visible
light with an average output powers between 5 and 500 mW
are subclassified in Class 3B.
Class 4 Lasers
Class 4 lasers include most surgical lasers that have an average output power in excess of 500 mW. Very stringent control
measures are required for this class of lasers. These lasers
produce a hazard not only from direct or specular reflection,
but they may also be hazardous with diffuse reflection.
BIBLIOGRAPHY
1. H. C. Zweng and M. Flocks, Clinical experience with laser photocoagulation. Fed. Proc., Fed. Am. Soc. Exp. Biol., 24 (Suppl. 14):
6570, 1965.
2. H. C. Zweng, L. L. Hunter, and R. R. Peabody, Laser Photocoagulation and Retinal Angiography. St. Louis: Mosby, 1969.
3. H. Beckman et al., Transscleral ruby laser irradiation of the ciliary body in the treatment of intractable glaucoma, Trans. Am.
Acad. Ophthalmol. Otolaryngol., 76: 423436, 1972.
4. L. Goldman et al., Laser radiation of malignancy in man, Cancer
(Philadelphia), 18: 533545, 1965.
5. H. L. Rosomoff and F. Carroll, Reaction of neoplasm and brain to
laser, Arch. Neurol. (Chicago), 14: 143148, 1966.
6. T. E. Brown et al., Laser radiation I: Acute effects of laser radiation on the cerebral cortex, Neurology, 16: 783, 1966.
12. A. Tuulonen et al., Laser trabeculoplasty versus medication

treatment as primary therapy for glaucoma, Acta Ophthal., 67:
275280, 1989.
13. P. Sullivan et al., Long term outcome after photocoagulation for
proliferative diabetic retinopathy, Diabetic Medicine, 7: 788
794, 1990.
14. E. M. Beck, E. D. Vaughan, and R. E. Sosa, The pulsed dye laser
in the treatment of ureteral calculi, Semin. Urol., 7: 2529, 1989.
15. S. P. Dretler, Laser lithotripsy: A review of 20 years of research
and clinical applications, Lasers Surg. Med., 8: 341356, 1988.
16. E. Bostofte et al., Conization by carbon dioxide laser or cold knife
in the treatment of cervical intraepithelial neoplasia, Acta Obstetrica et Gynecologica Scandinavica, 65: 199203, 1986.
17. J. M. Lomano, Nd:YAG laser applications in gynecology. In Advances in Nd:YAG Laser Surgery. New York: Springer-Verlag,
1988.
18. P. Andre et al., Lasers in dermatology, Ann. Dermatol. Venereol.,
117: 377395, 1990.
19. S. J. Adams et al., The effect of wavelength, power and treatment
pattern on the outcome of laser treatment of port-wine stains,
Brit. J. Dermatol., 117: 487494, 1987.
20. J. G. Hunter, Endoscopic laser applications in the gastrointestinal tract, Surg. Clin. N. Amer., 69: 11471166, 1989.
21. D. M. Crockett and B. N. Reynolds, Laryngeal laser surgery, Otolaryngol. Clin. N. Amer., 23: 4966, 1990.
22. T. J. Dougherty, Photodynamic therapy: Status and potential,
Oncology, 3: 6773, 1989.
23. S. L. Marcus, Photodynamic therapy of human cancer: Clinical
status, potential and needs, Proc. SPIE, IS6: 556, 1990.
24. M. D. Stern, In vivo evaluation of microcirculation by coherent
light scattering. Nature, 254: 5658, 1975.
25. D. Kilpatrick, J. V. Tyberg, and W. W. Parmley, Blood velocity
measurement by fiber optic Doppler anemometry, IEEE Trans.
Biomed. Eng., BME-29: 142145, 1982.
26. T. Tanaka and G. B. Benedek, Measurement of the velocity of
blood (in vivo) using a fiber optic catheter and optical mixing
spectroscopy, Appl. Optics, 25: 14171419, 1986.
27. H. Stricker et al., Acute and long-term effects of prostaglandin
E1 assessed by clinical and microcirculatory parameters in critical limb ischemia: A pilot study, Int. J. Microcirc., 16: 5763,
1996.
28. E. G. Salerud et al., Rhythmical variations in human skin blood
flow, Int. J. Microcirc. Clin. Exp., 2: 91, 1983.
29. I. M. Braverman et al., Topographic mapping of the cutaneous
microcirculation using two outputs of laser Doppler flowmetry:
Flux and the concentration of moving blood cells, Microvasc. Res.,
44: 3348, 1992.
30. C. E. Riva, H. E. Grunwald, and S. H. Sinclair, Laser Doppler
velocimetry study of the effect of pure oxygen breathing on retinal blood flow, Invest. Ophthal. Vis. Sci., 24: 4751, 1983.
234
LASER BEAM MACHINING
31. T. Tanaka, C. Riva, and I. Ben-Sira, Velocity measurement in

human retinal vessels, Science, 186: 830, 1974.
. O
berg et al., Use of a new laser Doppler flowmeter for mea32. P. A
surements of capillary blood flow in skeletal muscle after bullet
wounding, Acta Chir. Scand., Suppl., 489: 145, 1979.
33. Guide for the selection of laser eye protection, 1993. Laser Institute of America, 12424 Research Parkway, Suite 125, Orlando,
FL 32826.
34. American National Standards Institute, Safe use of lasers, ANSI
Standard Z136.1, 1993, available from Laser Institute of
America, 12424 Research Parkway, Suite 125, Orlando, FL
32826.
Reading List
Books
M. H. Niemz, Laser-Tissue Interactions: Fundamentals and Applications, New York: Springer-Verlag, 1996.
J. A. S. Carruth and A. L. McKenzie, Medical Lasers: Science and
Clinical Practice, Bristol, UK: Adam Hilger, 1986.
D. H. Sliney and S. L. Trokel, Medical Lasers and Their Safe Use,
New York: Springer-Verlag, 1993.
K. A. Ball, Lasers: The Perioperative Challenge, St. Louis: Mosby,
1995.
C. A. Puliafito, Lasers in Surgery and Medicine: Principles and Practice, New York: Wiley-Liss, 1996.
G. T. Absten and S. N. Joffe, Lasers in Medicine: An Introductory
Guide, Cambridge: Cambridge University Press, 1985.
J. Wilson and J. F. B. Hawkes, Lasers: Principles and Applications,
London: Prentice Hall, 1987.
O
berg, Laser-Doppler blood flowmetry, BosA. P. Shepherd and P. A
ton: Kluwer, 1990.
Review Articles
M. M. Judy, Biomedical lasers. In J. D. Bronzino, (ed.), The Biomedical Engineering Handbook, Boca Rato, FL: CRC/IEEE Press, 1995.
K. F. Gibson and W. G. Kernohan, Lasers in medicinea review. J.
Med. Eng. Tech., 17 (2): 5157, 1993.
Chapters on laser scalpel and laser surgery. In J. G. Webster, (ed.),
Encyclopedia of Medical Devices and Instrumentation. New York:
Wiley, 1988.
J. A. Parrish and B. C. Wilson, Current and future trends in laser
medicine. Photochemistry Photobiology, 53 (6): 731738, 1991.
A. N. Obeid et al., A critical review of laser Doppler flowmetry. J.
Med. Eng. Tech., 14: 178181, 1990.
O
berg, Laser-Doppler flowmetry. Biomed. Eng., 18 (2): 125
P. A
163, 1990.
G. A. Holloway, Jr., Laser Doppler measurements of cutaneous blood
flow. In P. Rolfe (ed.), Non-Invasive Physiological Measurements,
vol. 2. London: Academic Press, 1983.
Periodicals
Research articles on current applications of lasers in medicine are
published in the following journals: Lasers in Medical Science, Lasers in Surgery and Medicine, Lasers in the Life Sciences, Lasers
in Ophthalmology, Photochemistry Photobiology, and Biophotonics International.
YITZHAK MENDELSON
Worcester Polytechnic Institute
MEDICAL COMPUTING
485
MEDICAL COMPUTING
A computer in one form or another is present in almost every
instrument used for making measurements or delivering
486
MEDICAL COMPUTING
therapeutic or experimental interventions in clinical practice,

clinical research, or in any of the fields of the biomedical sciences. In some cases, the computer is a chip that is embedded
somewhere in the instrument, and the user has little indication that he or she is actually interacting with a central processing unit (CPU). In other cases, the form of the computer
is an engineering workstation or a personal computer with
keyboard, mouse, and the other accoutrements that are normally associated with the act of computing. In either case,
the computer allows the accomplishment of tasks that were
impossible or overwhelmingly difficult without the use of this
ubiquitous technology.
Many of the concepts that were implemented in early computers are still routinely used in modern devices, but with
vastly different technological bases. Most digital computers
have traditional architectures, with separate memories, central processing units, and input/output and storage devices.
A major difference is the miniaturization and increased efficiency of electronics devices, including computers, which have
allowed the development of sophisticated medical therapeutic
options that can be totally implanted in the body. The increased sophistication of computers has also led to vastly improved user interfaces and quality control in all kinds of medical instruments.
This article reviews some of the historical and contemporary computer applications in medicine and biomedical research, from real-time applications to medical informatics to
virtual reality. These applications are as diverse as the physical form of the computer on which they are implemented.
They include: real-time signal acquisition and subsequent
processing; efficient storage and manipulation of enormous
databases for large-scale clinical trials and other clinical and
basic research; assistance in clinical decision-making; acquisition, processing, and transmission of diagnostic images from
a wide range of technologies; simulation and display of twoand three- dimensional structures and function; animated
displays of physiological processes; and access to educational
and promotional medical information. Of course, any discussion of the use of computers in biomedicine would be incomplete without reference to the explosion in the use of the Internet for electronic mail, news, and World Wide Web access.
Many of these topics will be briefly mentioned; specific applications will be more completely described as examples of how
computers can be used to improve medical practice through
broadening the scientific underpinnings and making new
techniques available to practitioners. Many applications that
are implemented on computers will be more fully developed
in other articles of this encyclopedia.
MEDICAL INFORMATICS
The study of the use of computers in medicine is often called
medical informatics. Medical informatics has been termed
an emerging discipline that has been defined as the study,
invention, and implementation of structures and algorithms
to improve communication, understanding, and management
of medical information (1). (Further information can be
found on the newsgroup sci.med.informatics). This broad area
comprises many of the subjects included in medical computing, and there is a vast literature on ways in which computers
improve health care delivery through aids in organizing, ac-
cessing, and presenting knowledge. Medical informatics is

concerned with the flow of medical information, and how data
that can be used in the effective diagnosis and treatment of
disease can be made available in a timely fashion and in a
format that will provide the most usefulness (2). The integration of medical information systems into routine clinical activities has not been fully achieved (3).
Efficient organization and delivery of biological and medical information have assumed new importance with the explosion in knowledge about the genetic and molecular bases of
disease. One of the grand challenges of computing is the deciphering of the human genome, and sophisticated computer
algorithms have been developed for identifying genes embedded in long DNA sequences (4). A recent issue of the journal
Science described some of the opportunities for the application of computers in several areas of medical and biological
research. Articles in the aforementioned issue described new
approaches to searching the World Wide Web as a digital
medical library (5), the application of computing to mathematical ecological studies (6), and the use of a massive database for investigating the relationship between pharmacological agents and cancer (7). In general, the Internet, including
news groups and World Wide Web sites, has been a rich resource for all kinds of medical information (8,9); indeed, the
challenge is to develop methods for accessing the data in ways
that are intuitive and accurate. For example, a system has
been written to exploit effectively the digital images of human
anatomy that are stored on the Web (10). Figure 1 is an example of a workstation screen from this system which allows
flexible interrogation of three-dimensional anatomical databases, providing access to this wealth of information for users
without detailed knowledge of computer and database architectures.
The study of artificial intelligence and expert systems is
also typically considered a central theme of medical informatics. These systems use learning and advisory strategies to
assist in decisions related to diagnosis and therapy of human
disease. One of the earliest results in expert systems was the
computer program named MYCIN, which was designed to assist physicians in the selection of antimicrobial agents for hospital patients with bacterial infections (11). Artificial neural
networks have been used to assist in the detection of patterns
in clinical data and associate them with clinical conditions
and outcomes (12,13). Human beings are quite adept at incorporating uncertainties in their reasoning, but computers have
been viewed as unforgiving in the face of ambiguities. The
theories of fuzzy sets and fuzzy logic (14) have been developed
to address this discrepancy, and they have attracted attention
for their wide applicability (15). They have been used in classification of biomedical signals (16) and analysis of biomedical
images (17). The study of expert systems is an active area of
investigation in the application of computers to medicine.
SIGNAL ACQUISITION AND ANALYSIS

One of the earliest applications of computers in medicine and
medical science was their use for acquiring, analyzing, and
displaying waveforms that reflect physiological and pathological processes. Prominent examples include the electrocardiogram (ECG) and electroencephalogram (EEG), which record
the manifestations at the body surface of the electrical gener-
MEDICAL COMPUTING
487
in heart rhythms formerly thought to be completely disorganized (24,25). Figure 2 is an example of a signal processing
result that demonstrates that electrograms recorded from a
rectangular array during ventricular fibrillation have a great
deal of organization even when they are recorded from sites
separated by as much as 5 mm to 11 mm. Thus, computerbased multichannel acquisition and analysis of cardiac arrhythmias have revealed phenomena that might be crucial in
the improved prevention and treatment of these often fatal
derangements of rhythm.
Multichannel recording from the brain using computerbased systems has resulted in new insights into the way in
which the brains electrophysiological and psychological functions are organized (26). Similar systems have been developed
to study the electrical activity associated with the gastrointestinal (27), genitourinary (28), and reproductive (29) systems.
The development and implementation on computers of the
fast Fourier transform has allowed the examination of biosignals in the frequency domain, opening the doors for new insight into mechanisms of important clinical entities (30).
Spectral methods allow the elucidation of relationships between different physiological systems (31). Wavelet theory
has further extended the application of frequency domain
techniques by avoiding the limitations imposed by discrete
Fourier analysis (32). Wavelets have been applied to electrocardiography, to detect irregularities in heart rhythm; to pho-
R(dx, dy)
Figure 1. Computer screen from the Digital Anatomist Information

System (10), an interactive query system for access to three-dimensional anatomic databases using Internet browser principles. This
display allows the user to identify particular structures of the brain,
in this case the thalamus, by selecting the region of interest with the
computer input device. Reprinted from Ref. 10, with permission.
1
0.5
1
0.5
10
0
0.5
0 dy
(mm)
10
ators in the heart and brain, respectively. For computer analysis, it is necessary to sample the waveform and convert it
from analog to digital format (18). In electrocardiology, computers have been used for the analysis of clinical ECGs to
identify patterns in the waveform that reflected underlying
disorders of electrical activation or structural heart disease
(19,20). Similarly, patterns in EEGs have been correlated
with normal neurophysiology and with pathologies such as
epilepsy (21). Commercial systems are based on similar analyses and are widely used in hospitals and clinics for diagnosing cardiac and neurological abnormalities.
At the same time, the use of computers has expanded the
level of investigations that are possible in attempting to understand the scientific basis of normal and abnormal physiological function. Even though much was learned about cardiac
electrophysiology by using instruments like string galvanometers (22), improved technology and the application of digital
computers have allowed measurements to be made in situations that were not accessible to earlier investigators. For example, it is now possible to record from hundreds of sites on
the surface of the heart and within the cardiac tissue to assemble a high-resolution reconstruction of the intrinsic or externally generated bioelectric events in the myocardium (23).
Cardiac mapping studies that acquire data from many sites
simultaneously have shown that there is considerable order
0
(dx (mm)
0
0.5
10
10
0
0
10
(a)
(b)
1
0.5
0
0.5
10
10
0
0
10
10
10
10
10
1
0.5
0
10
0.5
10
0
0
10
10
(c)
(d)
Figure 2. Plots of correlation coefficient between two electrograms

recorded from a rectangular plaque array on the ventricular epicardium of a pig during ventricular fibrillation. Panels a to d are data
from 1 s to 2 s; 2 s to 3 s; 3 s to 4 s; and 60 s to 61 s after the induction
of ventricular fibrillation. The electrical activity in two electrodes appears to be well correlated, out to a distance between recording sites
of about 5 mm to 7 mm, but this distance changes as fibrillation continues. Whereas fibrillation has traditionally been considered to be
a completely disorganized rhythm, computer-based signal processing
techniques have revealed substantial levels of order. Reprinted from
Ref. 24, with permission.
488
MEDICAL COMPUTING
nocardiology, to analyze heart sounds in search of turbulence

associated with obstructions in coronary vessels; to examine
the EEG for evidence of fetal respiratory abnormalities; and
in image analysis and compression (32).
COMPUTER-BASED DEVICES
Many medical devices require the acquisition and analysis of
analog signals or variables that reflect underlying physiological processes. Typically, a method is developed to transduce
the variable of interest to a voltage which can then be converted to digital form and analyzed. The implementation of
these systems ranges from integrated chips with analog electronics, analog-to-digital converters, and some signal processing capabilities included to general-purpose computers
with specialized or general-purpose electronics designed to
carry out these functions. The devices are used in physiological and biomedical research and in instruments used for making clinical decisions. The computers are often used as control
systems as well as data acquisition devices, with or without
closed-loop feedback.
Embedded Computers
Figure 3 is an example of a highly specialized medical device,
the implantable cardioverter-defibrillator, with a fully func-
tional computer embedded in it. The function of this device

is to monitor continuously the electrical signal derived from
electrodes in contact with the heart, detect the onset of abnormalities in cardiac rhythm that are candidates for treatment,
and deliver appropriate therapy. To carry out this role effectively, it is necessary for the instrument to sample and digitize the cardiac electrograms with sufficient accuracy to allow
morphological analysis; identify the time at which the electrical wave passes the electrode in order to compute RR intervals and, thus, heart rate, often with analog processing (33);
distinguish between normal rhythms and categories of arrhythmias that require different levels of therapy (34); and control delivery of antitachycardia pacing or a defibrillation shock.
Figure 3(a) is a photograph of such a device. It is intended
for implantation beneath the clavicle in patients who are at
high risk for sudden cardiac death. In use, leads are directed
from the header through the venous system into the right
ventricle of the heart. There are electrodes on the catheters
for sensing the cardiac activity and delivering energy. Figure
3(b) is a photograph of the die of the microprocessor that is in
the defibrillator. The demands on this circuit are extraordinarily stringent, since a malfunction in either software or
hardware is likely to be fatal to the patient wearing it. The
validation of the operation of the computer is obviously of utmost concern to the manufacturer, the implanting physician,
regulatory agencies, and, especially, the patient.
Figure 3. (a) Photograph of a mockup of an implantable cardioverter-defibrillator. The device is

implanted under the clavicular region of a patient who is at high risk of sudden cardiac death.
The overall dimensions of the device are 4.5 7 1.25 cm. (b) A photograph of a die of a digital
computer system that is contained in the device shown in panel a. The dimension of the chip
that is fabricated from this die is 12.8 by 7.7 mm. The chip, with 2 interconnecting lines,
contains about 150,000 transistors. In addition to a typical microprocessor, the system contains
sense amplifiers, a digital control section for analyzing cardiac electrical activity and initiating
therapy when necessary, five banks of 1 Kb read-only memory, and circuits for telemetering data
to and from the external world. Photograph courtesy of CPI/Guidant Corporation.
MEDICAL COMPUTING
489
that provide users with scientifically and clinically useful information. Furthermore, the workstation and associated peripheral devices can be used as archival storage systems for
managing the overwhelming amounts of data that are generated by modern imaging modalities.
The computers associated with these clinical systems can
be networked using industry standard hardware and software
to provide convenient access to remote sites, either within a
hospital or medical center or outside the center to a referring
physician or a specialist who might have more experience in
interpreting certain imaging results. This is an example of
the interaction between medical imaging and telemedicine
(36), often thought of as one of the subdisciplines of medical
informatics.
COMPUTER GRAPHICS AND MEDICINE
Figure 4. A ray-traced, volume-rendered magnetic resonance image

of a canine heart with a myocardial infarction. The image was acquired from a postmortem, formalin-fixed heart after a study to determine whether necrotic tissue could be reliably detected by MRI. Computer-based magnetic resonance microscopy can be used for diagnosis
and evaluation of pathology, as well as for investigations into the
interplay of structure and function in clinical and experimental studies. Reprinted from Ref. 35, with permission.
Computers are often embedded in other medical instruments that do not play as acutely critical a role as do implanted devices, but are very important in assessing health
and disease. For example, a blood gas or electrolyte analyzer
might have a microprocessor that controls the user interface,
calibration procedures, and data acquisition, analysis, and
display. The widespread use of microcomputers in these analyzers provides convenient access to functions such as calibration and standardization that previously were time-consuming and labor-intensive.
Another early use of computers in medicine was the application of graphics systems for the acquisition and realistic display of anatomic structures in research and clinical situations. Much of the emphasis has been on two- and threedimensional reconstructions of data from medical imaging
modalities. Early work focused on the use of computers to estimate cardiac function from single- and dual-plane cineangiography (37), as well as estimate the reconstruction of coronary artery anatomy from coronary arteriograms (38,39). As
medical imaging technologies have advanced, the computational demands for extracting new information from image
analysis and displaying the data in realistic ways have increased. Substantial portions of the techniques developed for
nonmedical applications are not immediately applicable to biological and physiological systems because of the inherent
variability and irregularity that are not present in, for example, computer-aided design/computer-aided manufacturing
(CAD/CAM) structures (40). Another problem that is unique
to medical applications is the recent emphasis on reducing
health care costs, limiting the unfettered introduction of new
technologies (40). At times, there is a problem with the integration of creative, novel algorithms to a community that is
sometimes reluctant to modify procedures that have been established as effective, comfortable, and productive (3).
Image Processing
Stand-Alone Computers
Other medical instruments are based on general-purpose
computers, typically engineering workstations. Examples of
these are large imaging systems, such as magnetic resonance
imaging (MRI) or computed tomography (CT) systems. In
these cases, the computer fulfills a variety of roles. In MRI
systems, the computer can provide the user interface to the
highly specialized and complex hardware associated with the
magnet. The echo sequences that determine the imaging parameters and quality can be controlled through the computer
as a front end. These workstations often contain high-performance image processing and graphics hardware that can be
used for manipulation of the acquired images and display of
the results to the clinician in intuitive, usable formats. Figure
4 is a volume-rendered, ray-traced magnetic resonance image
of a canine heart with an experimentally induced myocardial
infarction (35). This image was generated on a high-performance workstation, and it demonstrates the kinds of displays
Most of the images from modern techniques, including digital

radiography, magnetic resonance imaging, computed axial tomography, and ultrasound imaging, require similar procedures for the production of usable graphics displays. Often,
the first step is the segmentation of the images, or the identification of different structures. Much work has been done on
computer-based segmentation, and currently most approaches use basically automatic systems with varying
amounts of manual editing of the results. There is a wide variety of segmentation algorithms, and many of them have
been implemented in readily available software packages. In
general, the imaging devices yield two-dimensional images,
with M N picture elements, or pixels, where M and N are
the number of pixels in the horizontal and vertical dimensions, respectively. Typical image sizes are around 256 256.
Image resolution, the number of pixels per length, is then determined by the size of the field of view of the device. The
computational demands of image processing algorithms in-
490
MEDICAL COMPUTING
Figure 5. Sagittal (a) and coronal (b)

magnetic resonance images of the head.
The left two panels are the original images. The middle panels show the outline
of the brain as detected by an automatic
three dimensional segmentation algorithm. The right panels are the portions
of the images that are within the detected
brain outline. Reprinted from Ref. 42,
with permission.
crease drastically with increasing resolution, since processing

generally increases with the square of the resolution for twodimensional images or as the cube of the resolution for volume analyses.
Each pixel is usually comprised of eight or more bits, each
combination of which represents a gray level. The value of the
gray level is determined by the method by which the image is
formed. For example, in CT scans and radiographs, the intensity of the pixels reflect whether the photons have passed
through bone or soft tissue. Magnetic resonance images detect
differences in water content in organs and can provide different kinds of information from CT scans. In any case, segmentation algorithms must manually, automatically, or semiautomatically detect the transition from one level of intensity to
another (41). Some are based on region growing methods, in
which a seed is provided and a region of similar intensities is
expanded until the level of the pixel intensities changes beyond a preset limit. Others are based on discontinuities in the
pixel intensities, often computing the gradients of the intensities and changing tissue classifications based on the magnitude of the gradient. Figure 5 is a two-dimensional projection
of a three-dimensional magnetic resonance image of a head.
The brain tissue has been separated from non- brain using a
three-dimensional seed growing algorithm (42). The automatic segmentation of images from the various medical imaging technologies remains an active area of research (41).
Modern imaging technologies routinely provide three-dimensional anatomic information (35), and the demands on
image processing programs have increased correspondingly.
The underlying principles are similar, but the computational
constraints become more severe. In addition, even though
storage and networking technologies are progressing quite
rapidly, image processing software must often incorporate
data compression and decompression capabilities to accommodate extension of images from two to three dimensions, the
increased image resolutions achievable with modern devices,
and the need to transmit large datasets over networks.
Computer Graphics
Intimately related to the issues of image processing are the
techniques by which medical and biological images are displayed with enough realism to achieve the intended results
but with enough efficiency to be used in actual clinical situations. Algorithms and programs for accurately portraying
anatomy and, to some extent, function have improved steadily, sometimes exceeding the ability of the hardware to meet
the demands. Fortunately, the well-known advances in performance and cost of advanced graphics hardware, including
general-purpose computers as well as special-purpose graphics processors, have provided the platforms necessary for implementation of state-of-the-art graphics techniques.
The display of two-dimensional images is, in principle,
straightforward on a computer output screen with multiple
colors or gray levels per pixel. The display programs provide
an interface between the user, the image, and the graphics
hardware and software of the computer so that one pixel of
the image is translated to one pixel of the video screen. Complications arise when there is a mismatch between the image
and the screen, so that image pixels must be removed or display pixels must be interpolated. A further complication for
the developer of either two- or three-dimensional graphics
software is the plethora of data file formats that exist (43).
Fortunately, many public domain or proprietary software
packages provide excellent format conversion tools, but some
experimentation is frequently required to use them properly.
MEDICAL COMPUTING
The development of methods for efficient and realistic rendering of three-dimensional images continues to be an area of
ongoing research. Early work reduced anatomic structures to
wire frame models (44), and that technique is still sometimes
used for previewing and rapid manipulation on hardware that
is not sufficiently powerful for handling full images in real
time or near real time. Several methods require the identification of surfaces through image segmentation, as described
above. The surfaces can be triangulated and displayed as essentially two-dimensional structures in three dimensions
(45). After initial processing, this is a rather efficient display
method, but much of the three-dimensional information is
lost. Alternately, the image can be reduced to a series of volumetric structures that can be rendered by hardware specialized for their reproduction (46). One of the most realistic, but
computationally expensive, three-dimensional rendering
methods is ray tracing, in which an imaginary ray of light is
sent through the structures and is attenuated by the opacity
of the anatomic structures that it encounters along the way
(47). Different effects can be emphasized by modifying the dynamic range of the pixels in the imagethat is, by changing
the relationship between the opacity of the image and the
pixel value to be displayed on the screen.
Medical computer graphics are at their most useful when
it is possible to superimpose images from more than one modality into a single display or to superimpose functional information acquired from biochemical, electrical, thermal, or
other devices onto anatomical renderings. As an example of
the former, images from positron emission tomography (PET)
scans, which reflect metabolic activity, can be displayed on
anatomy acquired by magnetic resonance imaging. The combination provides a powerful correlation between structure
491
and function, but the technical challenges of registering images from two different devices or taken at different times are
significant (48). An example of the combination of functional
and anatomic data is the superposition of electrical activity,
either intrinsic or externally applied, of the heart onto realistic cardiac anatomy. This kind of technique can provide new
insights into the mechanisms and therapy of cardiac arrhythmias (49). Figure 6 is a sequence of still frames from a video
showing the progression of a wavefront of electrical activation
across a three-dimensional cardiac left ventricle after an unsuccessful defibrillation shock.
Computer graphics and image processing, along with advanced imaging technologies, are making a significant impact
in medical knowledge and practice and have the potential for
many more applications. A combination of traditional CAD/
CAM visualization and advanced imaging can be used for effective assessment of quality of fit of orthopedic prostheses
(50). Capabilities and functionality have increased dramatically with the advent of advanced graphics hardware and
commercial software packages aimed at scientists and clinicians who are not graphics experts. Full realization of the
benefits of these systems will require further advances in
these areas, along with adaptation to the needs of clinicians
and the constraints of the changing health care climate (51).
COMPUTER SIMULATIONS
Numerical and analytical simulations of physiological processes have intrigued investigators for many decades. The solution of inverse and forward problems in neurophysiology
and electrocardiology was considered to be an important exer-
Figure 6. A composite of eight magnetic resonance and isochronal surface images from the
second activation wavefront after an unsuccessful defibrillation shock. The electrical data were
acquired from about 60 plunge needles with endocardial and epicardial electrodes inserted
through the left and right ventricles of the heart of an experimental animal. Successive isochrones (left to right, top to bottom) are shown at 6 ms intervals. Visualization techniques that
allow the superposition of function and anatomy are very helpful in understanding the relationships between variables and how they affect physiological mechanisms, and they can potentially
lead to improved diagnosis and therapy. Reprinted from Ref. 49, with permission. Copyright CRC
Press, Boca Raton, FL.
492
MEDICAL COMPUTING
Figure 7. Display of a model of a femur

used to study adaptation of bone in response to a stimulus. (a) A finite element
mesh of proximal femur with stem of
prosthesis numerically implemented for
optimal fit. (b) Bone density distribution
in the femur at initial implant. (c) Applied
and muscle forces in the femoral head
with prosthesis in place. Reprinted from
Ref. 67, with permission. Copyright Gordon and Breach Publishers, Lausanne,
Switzerland.
cise for basic scientific reasons as well as for possible clinical

applications. The interpretation of the surface signals recorded in the context of approximate generators in the tissue
provided a basis for relating physiology to pathologic and clinical abnormalities in the electrocardiogram (52). The advances in the use of the vectorcardiogram as a diagnostic tool
is based on the approximation of the cardiac electrical generator as a current dipole (53). Similarly, patterns in the EEG
have been modeled as surface reflections of underlying electrical sources (54).
The recent introduction of minimally invasive procedures
for the treatment of diseases, including laparoscopic and thoracoscopic surgical procedures and radio-frequency ablation for
cardiac arrhythmias, has intensified the interest of forward
and inverse problems as a research area. For example, for
radio-frequency ablation of ventricular tachycardia, it would
be most helpful to localize, at least approximately, the origins
of abnormal cardiac electrical activity from sensors either on
the body surface or mounted on a catheter in the blood pool
(5557).
Computers have been used to model tissue at cellular and
fiber levels of resolution. It is possible to simulate the propagation of electrical activity either with finite-state automata
models (58) or by solving the differential equations that govern the current flow through the cell membrane (5962). The
electrophysiology of other organ systems has also been modeled effectively (63,64).
In addition to electrophysiological simulations, mechanical
models have been applied to increase our understanding of
the mechanical properties of soft tissue (65) and bone (66).
Such models (Fig. 7) are important in simulating surgical procedures and implants (67) for training, planning, and evaluation of surgery.
VIRTUAL MEDICINE
Many of the computing techniques applied to medicine and
medical sciences come together in the development of applica-
tions of virtual reality to medical practice (68,69). Virtual reality has been applied to surgery planning (70), physical medicine and rehabilitation (71), parkinsonism (72), and
psychological disorders (73,74).
Computers have been used in a great many ways to assist
in surgical procedures (70). Surgeons can be trained in surgical techniques by using advanced computer graphics and virtual reality methods (75,76); similar techniques can be used
for surgical planning (7779) and for improving the safety
and efficacy of the surgical procedure. Computers are used
during complex brain surgery as interactive tools for guiding
and measuring the progress of the procedure, with the hope
that resection of lesions could be performed with less damage
to surrounding tissue (80). It is possible to use high-resolution
graphics to traverse internal organs virtually, yielding much
of the same information that is available from standard endoscopic techniques, as shown in the image in Fig. 8 acquired
at the Mayo Clinic (68).
Figure 8. Virtual colonoscopy, with an internal view of the transverse colon. The image was acquired by a helical CT scan, segmented,
and reconstructed. Virtual procedures can replace or augment actual
endoscopic examinations, reducing or eliminating the attendant risk
and discomfort. The image was acquired at the Mayo Clinic. Reprinted from Ref. 68, with permission. Copyright 1998, IEEE.
MEDICAL COMPUTING
Computer technology has also made it possible for medical

experts to use their knowledge at remote locations. This
allows state-of-the-art medical diagnosis and treatment at
sites where advanced technology would not normally be available (36). At another level, similar technologies have provided
the necessary tools for minimally invasive surgery and microsurgery, using computer graphics and simulated tactile responses to extend the capabilities of the surgeon and to provide simulations for realistic rehearsals of the surgical
experience (81,82).
The most sophisticated applications of computer graphics
to remote and minimally invasive surgery have depended on
the availability of high-resolution, high-quality graphical representations of the anatomy under treatment, either on an
individualized basis (83) or as a global atlas of human anatomy (84).
REGULATION, RELIABILITY, AND PRIVACY
Computer software that controls devices used in the diagnosis
and treatment of disease is of great interest to a regulatory
agency of the US government, namely, the Center for Devices
and Radiological Health (CDRH) of the Food and Drug Administration (FDA). This agency maintains a World Wide
Web site (http://www.fda.gov/cdrh/swpolpg.html) which provides guidance to device manufacturers in understanding regulatory policy enacted by the Federal Food, Drug and Cosmetic Act through a review of the FDA Software Policy
Workshop. The CDRH also sponsors biannual workshops on
software policy, the proceedings of which are included at this
site. The Food, Drug and Cosmetic Act defines a medical device as any instrument, apparatus, implement, machine, contrivance, implant, in vitro reagent, or any other similar or
related article, including any component, part, or accessory,
which is . . . intended for use in the diagnosis of disease or
other conditions, or in the cure, mitigation, treatment, or prevention of disease . . . or intended to affect the structure or
any function of the body . . . Software that is excepted from
this includes general-purpose programs, such as spreadsheets, which are not intended solely for medical use, and a
few other categories.
Independent of the regulatory issues, the design of medical
software and systems requires the highest level of reliability
and accessibility. The difficulty of developing robust computer
programs and documenting their correctness are well known
(85,86). The field of software engineering has been instrumental in providing tools for developing medical software that has
the highest possible level of accuracy and reliability (87). Device manufacturers have recognized the importance and usefulness of software engineering principles in the design and
implementation of control programs for their products.
Finally, ethical use of computer databases and the Internet
for the transfer of medical information imposes a need for
methods for strict privacy and security in data transmission.
The rapid introduction of new technologies requires that the
issues related to the sharing of patient information over
openly accessible channels be continually evaluated and improved (88).
CARDIAC MAPPING
An area of medical practice and research that has benefited
greatly by the use of computers and many computer-based
493
techniques is cardiac mappingthat is, the use of technology

to determine the path of electrical activity in the heart during
normal and abnormal cardiac rhythms and to measure the
effects of external stimuli and shocks that are applied therapeutically or as test perturbations. The activation sequence
and response to interventions can be measured electrically
(89,90) or optically (91,92). Electrical mapping is widely applied clinically to guide catheter-based or surgical interventions, and both technologies are important in investigating
the mechanisms of electrophysiological phenomena. Both approaches are computer-intensive, requiring real time data acquisition, signal analysis, image acquisition, statistical analysis, data storage and manipulation, and visualization.
In electrical mapping, the first step is the transduction of
the ionic currents in the cardiac tissue to electrical signals
that can be input to the data acquisition system. The construction of electrodes is crucial to ensure that the appropriate variables are measured (93). Signals in optical mapping are generated by the application of fluorescent dyes that
are sensitive to the electrical potential across the membranes
of the cardiac cells. In either case, the waveforms are input
to analog-to-digital converters.
In one implementation of an electrical cardiac mapping
system (23), all of the parameters of the analog front end,
including gains, frequency settings, and input range, are controlled by a series of microprocessors (94). The front-end processors assemble the data from 528 independent input channels into a data stream and send it to a data bus for recording
on long- or short-term recording devices. Because of the complexity of the analog processing, the user interface has been
designed to provide a great deal of direction and intuitive interaction with the investigators. Other microprocessors control stimulators (95) and associated investigational tools, such
as defibrillators and waveform analyzers.
After the raw waveforms are acquired, digital signal processing algorithms are applied to them for several purposes.
In some cases, the electrograms are analyzed to detect and
locate in time local electrical activations, those events that
represent the passage of an electrical wavefront in proximity
to the electrode (96,97). The local activation times form the
basis for other analysis programs. In other cases, the potential generated in the cardiac tissue from an external shock is
measured in all electrodes (98) in order to understand the
relationship between the electrical potential and gradient (99)
distributions in the myocardium and the efficacy of the therapy. The results of a clinical mapping study are shown in Fig.
9. Figure 9(a) is a sequence of cardiac electrograms recorded
using a commercial mapping system in a patient undergoing
an electrophysiology study. The electrograms were recorded
by a halo catheter placed in the right atrium, and they demonstrate the progression of an atrial flutter wavefront in a
circular pattern. The activation is reentrantthat is, selfsustainingand continually traverses the same anatomic
pathway. Figure 9(b) is a plot of local activation times that
were defined by the intrinsic deflection of the electrogram.
Computer displays such as these guide the application of radio-frequency energy applied to the heart for ablation of arrhythmogenic tissue. Other useful parameters that emphasize other aspects of the electrophysiology can be derived from
data such as these (100103).
It is often necessary or, at least, helpful to know (1) the
three-dimensional anatomy of the heart in which measure-
494
MEDICAL COMPUTING
100 ms
400 mm/s
(a)
(b)
Figure 9. Results of a clinical electrical cardiac mapping study in a
patient undergoing ablation of atrial flutter. (a) Electrograms recorded during the arrhythmia from a catheter inserted into the right
atrium in a loop configuration. (b) Activations derived from the intrinsic deflections in the electrograms shown in panel a. The continuous
nature of the activity demonstrates the reentrant mechanism around
anatomical obstacles in the right atrium. Ablation can eliminate conductivity in part of the reentrant pathway, curing the atrial flutter.
though custom software is often used to produce contour

maps (107), most commercial visualization packages have
contour generation routines that are efficient and easy to use,
especially for regular geometries, as indicated in Fig. 10.
With adequate processing, more complex displays can be
produced. As discussed above, it can be most helpful to combine functional electrical information with anatomic data
(49). Another approach is the animation of the activation sequence (108). By animating the color-coded values of time
since last activation or potential or derivative of potential in
each electrode as a function of time, the activation sequence
can be effectively followed without explicitly defining times of
local activation (Fig. 11) (109). If the electrodes are sufficiently close together, the displays can be produced without
interpolation. Thus, animation has the potential of removing
two important sources of ambiguity in cardiac mapping (110).
Traditionally, the interpretation of isochronal maps and
activation sequences has been subjective and descriptive,
with little basis for statistical comparisons between episodes
of cardiac arrhythmias. Computer programs and algorithms
have recently been developed which describe in quantitative
terms the characteristics of supraventricular and ventricular
flutter, tachycardia, and fibrillation. Some of them are aimed
at inferring the level of organization of the arrhythmia, especially atrial and ventricular fibrillation (25,111). Others are
designed to extract and identify wavefronts objectively as
they course across the myocardial tissue and to quantitate
their characteristics so that the effect of different experimental conditions and interventions can be compared in a rigorous and reproducible manner (112114).
0.525
50
100
ments are made and (2) the location of the electrodes used to
make the measurements. These variables can then be used
for further computations or to make the visualization of the
results more compelling and useful (104). Imaging techniques
as described above can be used for this purpose, allowing the
application of standard image processing packages for better
understanding of the electrophysiology (105). Image processing algorithms can also be used to improve our knowledge
of the underlying pathology and its relation to abnormalities
in electrical phenomena (35).
A traditional way of viewing activation sequences or other
variables in the heart is through contour mapsthat is, lines
of equal values of activation time, potential, or other measured variable. The approach depends on whether the array
of recording electrodes is two- or three-dimensional and
whether the array is in a regular pattern or is irregularly
spaced over the tissue. The variable of interest is typically
interpolated over the region in which the measurements were
made for more pleasing visual effects (106). Figure 10 is a
simple isochronal map of the activation sequence beneath a
rectangular array of electrodes on the outer, or epicardial,
surface of the right ventricle of an experimental animal. Even
0.525
0.525
150
0.525
0.525
200
0.515
0.515
250
0.505
0.505
300
50
100
150
200
250
300
350
Figure 10. Isochronal map from activation times measured with a

21 24 rectangular array of electrodes placed on the right ventricular epicardium of a dog. The contours were interpolated to a 84 96
array, with linear interpolation. The lines represent times of equal
activation, and they are spaced at 10 ms intervals. The labels of the
isochrones are in seconds from an arbitrary reference time. The map
was taken from a sinus beat and shows activation spreading from the
apical region, at the bottom of the plot, to the base of the right ventricle. The loop in one of the isochrones demonstrates one of the problems with interpolation and isochrone production in data with noise.
MEDICAL COMPUTING
10
11
12
13
14
495
Institutes of Health, Bethesda, Maryland, and National Science Foundation Engineering Research Center Grant CDR8622201.
BIBLIOGRAPHY
15
16
17
18
19
20
21
Figure 11. A sequence of frames from an animation of the activation

sequence of the several cardiac cycles after an successful defibrillation shock. The circle represents an apical view of the heart, with the
apex at the center and the base around the periphery. The data were
recorded from a sock containing 510 electrodes which was pulled over
the ventricular epicardium. The left anterior descending coronary artery is the line from the top of the circle to near the center and the
posterior descending coronary artery is the line at the bottom. Each
dot represents a time at which the electrogram recorded by the corresponding electrode is active, meaning that the absolute value of the
derivative of the electrogram exceeded a preselected threshold. The
earliest postshock activation is in the apical region of the left ventricle. A secondary early site appears in panel 7. The two wavefronts
merge, and then they spread to the right ventricle. Even though the
data were sampled at 2 kHz, the sequential displays are at 5 ms
intervals. Animation of electrograms in this manner can provide information about the activation sequence without explicitly defining
times of local activation in each electrogram. Reprinted from 109,
with permission.
CONCLUSION
Computers are used in almost every aspect of clinical medicine and biomedical research. They are indispensable in advanced devices and instrumentation. They are widely used for
the collection and analysis of demographic and clinical data
which provide a basis for the improved understanding of the
causes and epidemiologies of disease. They can be very effectively used for the training and accreditation of physicians
and other health care providers. While obviously no substitute for human clinical and scientific judgement, computers
have assumed a critical role as facilitators of diagnostic and
therapeutic procedures. As the inevitable progress in computer software and hardware occurs, medical professionals
will become more dependent on them. There will be continuing improvement in our understanding of methods for increased reliability and safety of computer-based devices and
equipment, and regulatory agencies will develop procedures
for evaluating these resources routinely and objectively. Advanced imaging technologies, higher performance graphics
hardware and software, and new surgical techniques will expand the use of microsurgery and remotely applied surgical
and invasive procedures. These prospects will depend on investigators in computer science, biomedical and software engineering, clinical practice, and physiology, but the computer
has the potential to be a positive force in improving health
care delivery while decreasing the financial burden of the
health care system on society.
1. P. Zickler, Informatics: Transforming raw data into real medical

information, Biomed. Instrum. Technol., 32: 111115, 1998.
2. J. Gosbee and E. Ritchie, Human-computer interaction and
medical software development, Interactions, 4: 1318, 1997.
3. J. G. Anderson, Clearing the way for physicianss use of clinical
information systems, Commun. ACM, 40: 8390, 1997.
4. A. Kamb et al., Software trapping: A strategy for finding genes
in large genomic regions, Comput. Biomed. Res., 28: 140153,
1995.
5. B. R. Schatz, Information retrieval in digital libraries: Bringing
search to the net, Science, 275: 327334, 1997.
6. S. A. Levin et al., Mathematical and computational challenges
in population biology and ecosystems science, Science, 257: 334
343, 1997.
7. J. N. Weinstein et al., An information-intensive approach to the
molecular pharmacology of cancer, Science, 275: 343349, 1997.
8. W. H. Detmer and E. H. Shortliffe, Using the Internet to improve knowledge diffusion in medicine, Commun. ACM, 40: 101
108, 1997.
9. D. G. Kilman and D. W. Forslund, An international collaboratory based on virtual patient records, Commun. ACM, 40: 111
117, 1997.
10. J. F. Brinkley et al., The digital anatomist information system
and its use in the generation and delivery of web-based anatomy
atlases, Comput. Biomed. Res., 30: 472503, 1997.
11. E. H. Shortliffe et al., Computer-based consultations in clinical
therapeutics: Explanation and rule acquisition capabilities of
the MYCIN system, Comput. Biomed. Res., 8: 303320, 1975.
12. W. G. Baxt, Use of an artificial neural network for the diagnosis
of myocardial infarction, Ann. Intern. Med., 115: 843848, 1991.
13. M. A. Leon and F. L. Lorini, Ventilation mode recognition using
artificial neural networks, Comput. Biomed. Res., 30: 373378,
1997.
14. C. A. Kulikowski, History and Development of Artificial Intelligence Methods for Medical Decision Making, in J. D. Bronzino
CRC Press, 1995, pp. 26812698.
15. D. Kalmanson and H. F. Stegall, Cardiovascular investigations
and fuzzy sets theory, Amer. J. Cardiol., 35: 8084, 1975.
16. J. Piater et al., Fuzzy sets for feature identification in biomedical signals with self-assessment of reliability: An adaptable algorithm modeling human procedure in BAEP analysis, Comput.
Biomed. Res., 28: 335353, 1995.
17. J.-S. Lin, K.-S. Cheng, and C.-W. Mao, Segmentation of multispectral magnetic resonance image using penalized fuzzy competitive learning network, Comput. Biomed. Res., 29: 314326,
1996.
18. J. Macy, Jr., AnalogDigital Conversion Systems, in R. W. Stacy
and B. D. Waxman (eds.), Computers in Biomedical Research,
New York: Academic Press, 1965, pp. 334.
ACKNOWLEDGMENTS
19. R. E. Ideker et al., Evaluation of a QRS scoring system for estimating myocardial infarct size: II. Correlation with quantitative
anatomic findings for anterior infarcts, Amer. J. Cardiol., 49:
16041614, 1982.
This work was supported in part by research grant HL-33637

from the National Heart, Lung, and Blood Institute, National
20. G. S. Wagner et al., Evaluation of a QRS scoring system for

estimating myocardial infarct size: I. Specificity and observer
agreement, Circulation, 65: 342347, 1982.
496
MEDICAL COMPUTING
21. R. Restak, The Brain, New York: Bantam Books, 1984.

22. T. Lewis and M. A. Rothschild, The excitatory process in the
dogs heart. Part II. The ventricles, Philos. Trans. R. Soc. London, 206: 181, 1915.
23. P. D. Wolf et al., Design for a 512 Channel Cardiac Mapping
System, in D. C. Mikulecky and A. M. Clarke (eds.), Biomedical
Engineering: Opening New Doors, New York: New York Univ.
Press, 1990, pp. 513.
24. P. V. Bayly et al., A quantitative measurement of spatial order
in ventricular fibrillation, J. Cardiovasc. Electrophysiol., 4: 533
546, 1993.
25. K. M. Ropella et al., The coherence spectrum. A quantitative
discriminator of fibrillatory and nonfibrillatory cardiac rhythms,
Circulation, 80: 112119, 1989.
26. A. Riehle et al., Spike synchronization and rate modulation differentially involved in motor cortical function, Science, 278:
1950, 1997.
27. J. D. Z. Chen, B. D. Schirmer, and R. W. McCallum, Measurement of electrical activity of the human small intestine using
surface electrodes, IEEE Trans. Biomed. Eng., 40: 598602,
1993.
28. C. R. R. Gallegos and C. H. Fry, Alterations to the electrophysiology of isolated human detrusor smooth muscle cells in bladder
disease, J. Urol., 151: 754758, 1994.
29. D. Devedeux et al., Uterine electromyography: A critical review,
Amer. J. Obstet. Gynecol., 169: 16361653, 1993.
30. M. E. Cain et al., Fast-Fourier transform analysis of signal-averaged electrocardiograms for identification of patients prone to
sustained ventricular tachycardia, Circulation, 69: 711720,
1984.
31. A. Nakata et al., Spectral analysis of heart rate, arterial pressure, and muscle sympathetic nerve activity in normal humans,
Amer. J. Physiol., 274 (Heart Circ. Physiol., 43): H1211
H1217, 1998.
32. M. Akay, Wavelet applications in medicine, IEEE Spectrum, 34
(5): 5056, 1997.
33. D. A. Brumwell, K. Kroll, and M. H. Lehmann, The Amplifier:
Sensing the Depolarization, in M. W. Kroll and M. H. Lehmann
(eds.), Implantable Cardioverter Defibrillator Therapy: The EngineeringClinical Interface, Norwell, MA: Kluwer, 1996, pp.
275302.
34. S. M. Bach, Jr., M. H. Lehmann, and M. W. Kroll, Tachyarrhythmia Detection, in M. W. Kroll and M. H. Lehmann (eds.),
Implantable Cardioverter Defibrillator Therapy: The EngineeringClinical Interface, Norwell, MA: Kluwer, 1996, pp.
303323.
35. J. C. M. Hsu et al., Magnetic resonance imaging of chronic myocardial infarcts in formalin-fixed human autopsy hearts, Circulation, 89: 21332140, 1994.
36. J. W. Hill and J. F. Jensen, Telepresence technology in medicine: Principles and applications, Proc. IEEE, 86: 569580,
1998.
37. P. Virot et al., Comparison of 14 methods for analysing left ventricle segmental kinetics by cineangiography, Arch. Mal Coeur
Vaiss., 77: 433441, 1984.
38. B. G. Brown et al., Quantitative coronary arteriography: Estimation of dimensions, hemodynamic resistance, and atheroma
mass of coronary artery lesions using the arteriogram and digital computation, Circulation, 55: 329337, 1977.
39. C. F. Starmer and W. M. Smith, Problems in acquisition and
representation of coronary arterial trees, Computer, 8 (7): 36
41, 1975.
40. M. L. Rhodes, Computer graphics and medicine: A complex partnership, IEEE Comput. Graphics Appl., 17: 2228, 1997.
41. R. C. Gonzalez and R. E. Woods, Digital Image Processing, Reading, MA: Addison-Wesley, 1992.
42. R. K. Justice et al., Medical image segmentation using 3-D
seeded region growing, Proc. SPIE Med. Imag., Newport Beach,
CA, 1997, pp. 900910.
43. D. C. Kay and J. R. Levine, Graphics File Formats. Blue Ridge
Summit, PA: Windcrest/McGraw-Hill, 1992.
44. W. M. Newman and R. F. Sproull, Principles of Interactive Computer Graphics, New York: McGraw-Hill, 1979.
45. W. E. Lorensen and H. E. Cline, Marching cubes: A high resolution 3D surface construction algorithm, Comput. Graphics, 21:
163169, 1987.
46. E. V. Simpson et al., Three-dimensional visualization of electrical variables in the ventricular wall of the heart, Proc. 1st Conf.
Vis. Biomed. Comput., Atlanta, GA, 1990, pp. 190194.
47. K. S. Klimaszewski and T. W. Sederberg, Faster ray tracing using adaptive grids, IEEE Comput. Graphics Appl., 17: 4251,
1997.
48. K. R. Castleman, Digital Image Processing, Englewood Cliffs,
NJ: Prentice-Hall, 1979.
49. T. C. Palmer et al., Visualization of bioelectric phenomena, CRC
Crit. Rev. Biomed. Eng., 20: 355372, 1992.
50. M. W. Vannier et al., Visualization of prosthesis fit in lowerlimb amputees, IEEE Comput. Graphics Appl., 17: 1629, 1997.
51. D. P. Mahoney, The art and science of medical visualization,
Comput. Graphics World, 19: 2532, 1996.
52. L. G. Horan et al., On the possibility of directly relating the
pattern of ventricular surface activation to the pattern of body
surface potential distribution, IEEE Trans. Biomed. Eng., 34:
173179, 1987.
53. L. G. Horan and N. C. Flowers, The Relationship Between the
Vectorcardiogram and Actual Dipole Moment, in C. V. Nelson
and D. B. Geselowitz (eds.), The Theoretical Basis of Electrocardiology, London: Oxford Univ. Press, 1976, pp. 397412.
54. R. J. MacGregor and E. R. Lewis, Neural Modeling: Electrical
Signal Processing in the Nervous System, New York: Plenum,
1977.
55. B. He and R. J. Cohen, Body surface Laplacian ECG mapping,
IEEE Trans. Biomed. Eng., 39: 11791191, 1992.
56. G. Huiskamp and F. Greensite, A new method for myocardial
activation imaging, IEEE Trans. Biomed. Eng., 44: 433446,
1997.
57. D. S. Khoury and Y. Rudy, A model study of volume conductor
effects on endocardial and intracavity potentials, Circ. Res., 71:
511525,1992.
58. K. D. Bollacker et al., A cellular automata three-dimensional
model of ventricular cardiac activation, Proc. Annu. Int. Conf.
IEEE Eng. Med. Biol. Soc., Piscataway, NJ, 1991, pp. 627628.
59. G. W. Beeler, Jr. and H. Reuter, Reconstruction of the action
potential of ventricular myocardial fibers, J. Physiol. (London),
268: 177210, 1977.
60. A. L. Hodgkin and A. F. Huxley, A quantitative description of
membrane current and its application to conduction and excitation in nerve, J. Physiol. (London), 117: 500544, 1952.
61. C.-H. Luo and Y. Rudy, A dynamic model of the cardiac ventricular action potential. II. After depolarizations, triggered activity, and potentiation, Circ. Res., 74: 10971113, 1994.
62. C.-H. Luo and Y. Rudy, A dynamic model of the cardiac ventricular action potential. I. Simulations of ionic currents and concentration changes, Circ. Res., 74: 10711096, 1994.
63. E. E. Daniel et al., Relaxation oscillator and core conductor models are needed for understanding of GI electrical activities,
Amer. J. Physiol., 266 (Gastrointest. Liver Physiol. 29): G339
G349, 1994.
MEDICAL COMPUTING
497
64. B. O. Familoni, T. Abell, and K. L. Bowes, A model of gastric

electrical activity in health and disease, IEEE Trans. Biomed.
Eng., 42: 647657, 1995.
91. S. B. Knisley et al., Optical measurements of transmembrane

potential changes during electric field stimulation of ventricular
cells, Circ. Res., 72: 255270, 1993.
65. E. S. Almeida and R. L. Spilker, Mixed and penalty finite element models for the nonlinear behavior of biphasic soft tissues
in finite deformation: Part I. Alternate formulations, Comput.
Methods Biomech. Biomed. Eng., 1: 2546, 1997.
66. B. R. McCreadie and S. J. Hollister, Strain concentrations surrounding an ellipsoid model of lacunae and osteocytes, Comput.
92. G. Salama, R. Lombardi, and J. Elson, Maps of optical action

potentials and NADH fluorescence in intact working hearts,
Amer. J. Physiol., 252: H384H394, 1987.
93. M. R. Neuman, Biopotential electrodes, in J. G. Webster (ed.),
Medical Instrumentation Application and Design, 3rd ed., New
York: Wiley, 1998, pp. 183232.
67. A. Terrier et al., Adaptation models of anisotropic bone, Comput.

94. P. D. Wolf et al., A 528 channel system for the acquisition and
display of defibrillation and electrocardiographic potentials,
Proc. Comput. Cardiol., Los Alamitos, CA, 1993, pp. 125128.
68. R. M. Satava and S. B. Jones, Current and future applications

of virtual reality for medicine, Proc. IEEE, 86: 484489, 1998.
95. D. L. Rollins et al., A programmable cardiac stimulator, Proc

Comput. Cardiol., Los Alamitos, CA, 1992, pp. 507510.
69. W. M. Smith, Scanning the technology: Engineering and medical science chart fantastic voyage, Proc. IEEE, 86: 474478,
1998.
70. M. L. Rhodes and D. D. Robertson, Computers in surgery and
therapeutic procedures, Computer, 29 (1): 23, 1996.
71. W. J. Greenleaf, Applying VR to physical medicine and rehabilitation, Commun. ACM, 40: 4346, 1997.
72. S. Weghorst, Augmented reality and Parkinsons disease, Commun. ACM, 40: 4748, 1997.
73. G. Riva, L. Melis, and M. Bolzoni, Treating body-image disturbances, Commun. ACM, 40: 6971, 1997.
74. D. Strickland et al., Overcoming phobias by virtual exposure,
Commun. ACM, 40: 3539, 1997.
75. S. L. Dawson and J. A. Kaufman, The imperative for medical
simulation, Proc. IEEE, 86: 479483, 1998.
76. K. H. Hohne et al., A virtual body model for surgical education
and rehearsal, Computer, 29 (1): 2531, 1996.
77. M. Bro-Nielsen, Finite element modeling in surgery simulation,
Proc. IEEE, 86: 503, 1998.
78. E. K. Fishman et al., Surgical planning for liver resection, Computer, 29 (1): 6472, 1996.
79. R. A. Robb, D. P. Hanon, and J. J. Camp, Computer-aided surgery planning and rehearsal at Mayo Clinic, Computer, 29 (1):
3947, 1996.
80. L. Adams et al., An optical navigator for brain surgery, Computer, 29 (1): 4854, 1996.
81. E. Chen and B. Marcus, Force feedback for surgical simulation,
Proc. IEEE, 86: 524530, 1998.
82. H. Delingette, Toward realistic soft-tissue modeling in medical
simulation, Proc. IEEE, 86: 512523, 1998.
83. G. E. Christensen et al., Individualizing neuroanatomical
atlases using a massively parallel computer, Computer, 29 (1):
3238, 1996.
84. M. J. Ackerman, The visible human project, Proc. IEEE, 86:
504511, 1998.
85. W. W. Gibbs, Softwares chronic crisis, Sci. Amer., 271 (3): 86
95, 1994.
86. B. Littlewood and L. Strigini, The risks of software, Sci. Amer.,
267 (5): 6275, 1992.
87. W.-T. Tsai, R. Mojdehbakhsh, and S. Rayadurgam, Capturing
safety-critical medical requirements, Computer, 31: 4042,
1998.
96. K. P. Anderson et al., Determination of local myocardial electrical activation for activation sequence mapping: A statistical approach, Circ. Res., 69: 898917, 1991.
88. T. C. Rindfleisch, Privacy, information technology, and health

care, Commun. ACM, 40: 93100, 1997.
89. R. E. Ideker et al., Simultaneous multichannel cardiac mapping
systems, Pacing Clin. Electrophysiol., 10: 281292, 1987.
90. F. X. Witkowski and P. B. Corr, An automated simultaneous
transmural cardiac mapping system, Amer. J. Physiol., 247:
H661H668, 1984.
97. E. V. Simpson et al., Evaluation of an automatic cardiac activation detector for bipolar electrograms, Med. Biol. Eng. Comput.,
31: 118128, 1993.
98. A. S. L. Tang et al., Measurement of defibrillation shock potential distributions and activation sequences of the heart in threedimensions, Proc. IEEE, 76: 11761186, 1988.
99. W. Krassowska et al., Finite element approximation of potential
gradient in cardiac muscle undergoing stimulation, Math. Comput. Modelling, 11: 801806, 1988.
100. P. V. Bayly et al., Estimation of conduction velocity vector fields
from 504-channel epicardial mapping data, Proc. Comput.
Cardiol., Indianapolis, IN, 1996, pp. 133140.
101. A. H. Kadish et al., Vector mapping of myocardial activation,
Circulation, 74: 603615, 1986.
102. D. S. Rosenbaum, B. He, and R. J. Cohen, New approaches for
evaluating cardiac electrical activity: Repolarization alternans
and body surface laplacian imaging, in D. P. Zipes and J. Jalife
(eds.), Cardiac Electrophysiology: From Cell to Bedside, Philadelphia: Saunders, 1995, pp. 11871198.
103. F. X. Witkowski et al., Significance of inwardly directed transmembrane current in determination of local myocardial electrical activation during ventricular fibrillation, Circ. Res., 74: 507
524, 1994.
104. E. V. Simpson, T. C. Palmer, and W. M. Smith, Visualization in
cardiac mapping of ventricular fibrillation and defibrillation,
Proc. Comput. Cardiol., Los Alamitos, CA, 1992, pp. 339342.
105. C. Laxer et al., An Interactive Graphics System for Locating
Plunge Electrodes in Cardiac MRI Images, in Y. Kim (ed.), Image Capture, Formatting and Display, Soc. Photo-Optical Instrum. Eng., 1991, pp. 190195.
106. E. V. Simpson et al., Discrete smooth interpolation as an aid to
visualizing electrical variables in the heart wall, Proc. Comput.
Cardiol., Venice, Italy, 1991, pp. 409412.
107. F. R. Bartram, R. E. Ideker, and W. M. Smith, A system for the
parametric description of the ventricular surface of the heart,
Comput. Biomed. Res., 14: 533541, 1981.
108. C. Laxer et al., The use of computer animation of mapped cardiac potentials in studying electrical conduction properties of arrhythmias, Proc. Comput. Cardiol., Los Alamitos, CA, 1991,
pp. 2326.
109. M. Usui et al., Epicardial shock mapping following monophasic
and biphaic shocks of equal voltage with an endocardial lead
system, J. Cardiovasc. Electrophysiol., 7: 322334, 1996.
110. E. J. Berbari et al., Ambiguities of epicardial mapping, J. Electrocardiol., 24 (Suppl.): 1620, 1991.
111. P. V. Bayly et al., Spatial organization, predictability, and determinism in ventricular fibrillation, Chaos, 8: 103115, 1998.
498
MEDICAL EXPERT SYSTEMS
112. J. Rogers et al., Recurrent wavefront morphologies: A method

for quantifying the complexity of epicardial activation patterns,
Ann. Biomed. Eng., 25: 761768, 1997.
113. J. M. Rogers et al., Quantitative characteristics of reentrant
pathways during ventricular fibrillation, Ann. Biomed. Eng., 25:
S-62, 1997.
114. J. M. Rogers et al., A quantitative framework for analyzing epicardial activation patterns during ventricular fibrillation, Ann.
Biomed. Eng., 25: 749760, 1997.
WILLIAM M. SMITH
The University of Alabama at
Birmingham
460
LIQUID INSULATION
LIQUID INSULATION
ARC SUPPRESSION PROPERTIES
Switchgear using the arc suppression properties of insulating
liquids (oils) was invented in the early 1880s. In the early
days, the structure of switchgear was simple: a pair of electrodes were placed in insulating oil. In such switchgear the
arc suppression mechanism is also simple: as the electrode
spacing increases so the arc length increases and the electric
arc is suppressed. This suppression results from the cooling
effect of hydrogen gas produced by the decomposition of the
insulating oil due to the arc. In arc suppression in insulating
oil, hydrogen gas produced by decomposition of the insulating
oil due to the arc plays an important role.
Arc Suppression by Hydrogen
The energy of the arc between a pair of electrodes in the insulating oil is dissipated by the electrodes, by conduction and
radiation, evaporation and decomposition of the insulating oil,
heating and expansion of gases produced by the decomposition of the insulating oil, and dissociation of hydrogen. Fifty
to seventy percent of produced gas is hydrogen, and the other
gases are acetylene, methane, and ethane. As shown in Table
1, the thermal conductivity of hydrogen at room temperature
is higher than that of other gases. At 4000C it is about 50
W/m K. This value is more than 5 times higher than for the
other gases. Therefore, the cooling effect is larger than that
of the other gases. By this cooling, the arc is suppressed at the
zero-current point of alternating current. Thus the current is
Table 1. Heat Conductivity of Gases

Gas
Hydrogen
Methane
Ethane
Acetylene
Heat conductivity (W/m K)

0.175
0.031
0.018
0.017
LIQUID INSULATION
cut off. Switches that utilize this arc suppression mechanism

are called plane-break oil circuit breakers.
As the current increases, it becomes more difficult to suppress the arc. Therefore, the breaking time (cutoff) of the current becomes longer. However, when the current exceeds a
certain magnitude, a large amount of hydrogenenough to
suppress the archas been produced. At this point the breaking time is again reduced. This means that the breaking time
shows a maximum value at a certain magnitude of current.
Arc Suppression by an Explosion Chamber
Cooling may not be sufficient to suppress a high-current arc
and to lessen the breaking time. In this case explosion chambers are used. In a simple explosion chamber a movable electrode of the circuit breaker (switchgear) acts as a stopper of
the chamber. In the early stages of separating the electrode,
the arc is enclosed in the limited space of the chamber. Therefore, the pressure in the chamber rises owing to gases produced by decomposition of the insulating oil. As this process
proceeds, the stopper is removed creating an exhaust hole.
Through this exhaust hole the gases in the chamber are
released abruptly. By this release, flows of gases and oil are
produced, and the arc is pressurized and blasted. These processes create efficient arc suppression. Furthermore, when
the lengthened arc contacts insulating solids while enclosed
in narrow gaps between them, more efficient arc suppression
results.
Because gases are abruptly exhausted through the hole,
adiabatic expansion occurs. Thus cooling is expected. In some
oil circuit breakers this is the main effect utilized. Some scientists maintain that in oil circuit breakers with an explosion
chamber, arc suppression can be entirely explained by the
cooling effect owing to the adiabatic expansion. In fact, however, insulating oils exhibit not only the arc suppression property resulting from the cooling effect of hydrogen, but also
the substantially different suppression properties of the oils
themselves. In high-current arc suppression, these two types
of suppression properties are combined.
In circuit breakers with an explosion chamber, a large
pressure rise is expected in the chamber in the case of highcurrent arc suppression, but not in the case of low-current
suppression. Thus in the latter case the breaking time is
longer, because the arc must be suppressed by the cooling effect of hydrogen alone. Therefore, the breaking time shows a
maximum at a certain magnitude of current (the critical current). But this breaking time is much shorter than that of
the plain-break oil circuit breaker. A circuit breaker with an
explosion chamber necessarily has plural arc suppression
mechanisms.
In some circuit breakers, in the region of the critical current, the auxiliary flow of the oil is forced by a piston to supplement the pressure rise and the conduction cooling effect.
By this means a constant breaking time is obtained over a
wide range of current.
Plane-break oil circuit breakers are used for low voltages
and low currents, such as 3.6 kV to 7.2 kV and 4 kA to 8 kA.
Oil circuit breakers with an explosion chamber are used for
high voltages and high currents. In the case of multibreak
circuit breakers, 700 kV with currents of several tens of kiloamperes have been achieved (1).
461
DISCHARGE RESISTANCE
The behavior of insulating liquids under highly stressed conditions and under conditions of partial discharge are among
the most important items in screening tests for newly developed insulating liquids and also in the routine testing of
liquids.
Gassing Rate
Methods of evaluating gas absorption and evolution of insulating oils under high stress after saturation with a gas are
described in IEC 628 and ASTM D 2330. The fundamental
approaches are similar to each other and amount to a modified Pirelli method.
The condition used in such methods differs from actual
field conditions, especially in the case of hermetically sealed
equipment such as power cables, capacitors, and many
power transformers.
Discharge Resistance
To evaluate the behavior of insulating liquids in a highly
stressed impregnated system and to obtain numerical results
for the recently developed impregnants with very high resistance to partial discharge, the above-mentioned methods are
not sufficient. As new liquids, especially with high aromaticity, are developed and applied voltage stresses are progressively increased, a new method is needed to characterize the
ability of such insulating liquids to prevent or suppress partial discharge under high stress. One of these methods, determination of the partial discharge inception voltage with a
needle and spherical ball oil gap, is described in IEC 61294.
The partial discharge inception voltage obtained by this
method is largely related to the chemical structure of the liquid and is correlative to partial discharge in impregnated insulating systems such as capacitor elements.
DIELECTRIC CONSTANT AND LOSS
Dielectric polarization occurs when an electric field is applied
to insulating oil. When there is a time delay in the formation
of polarization, dielectric loss arises from the phase delay of
polarization under an alternating electric field. The loss due
to this dielectric polarization is proportional to the dielectric
loss tangent tan , which is equal to the ratio of the dielectric
loss factor to the dielectric constant :
tan =
/
(1)
The temperature dependence of tan and in three mineral

insulating oils is shown in Fig. 1. At a frequency of 1 kHz,
the maximum of tan is observed at 40C to 50C. The
maximum value moves to higher temperatures as the measurement frequency is increased.
The maxima in tan and in the dielectric loss factor are
caused by asymmetry of the molecular structure leading to
dipole moments in, for example, aromatic compounds. These
maximum values increase as the insulating oil deteriorates.
The frequency f m (Hz) at which tan or are a maximum
has the following temperature dependence:
f m = f 0 exp(H/RT )
(2)
LIQUID INSULATION
1.4
2.40
tan (%)
1.2
2.20
1.0
Oil A
Oil B
Oil C
2.00
0.8
Oil A
1.80
Dielectric constant
462
0.6
0.4
Oil B
0.2
Oil C
Figure 1. Temperature dependence of dielectric constant and tan at 1 kHz in

mineral insulating oil.
0
80 60 40 20
f m0
C1 (T T0 )
=
fm
C2 + (T T0 )
(3)
where f m0 is the frequency that yields the maximum value of

tan at temperature T0, f m is the frequency that yields the
maximum value of tan at temperature T, and C1 and C2 are
the WLF constants. H is expressed by the following equation, deduced from Eqs. (2) and (3):
H =
2.303RC1C2 T 2
C2 + T T0
(4)
which applies over a wide temperature range. C1 and C2 are

obtained from the slope and intercept of 1/ln( fm0 /f m) plotted
against T T0, in view of Eq. (2).
There is no effect due to the orientation of dipoles above
room temperature at low frequencies, such as those of commercial power. However, the mobility of the charge carriers is
high, and electric conduction losses can be large. The electric
conduction loss is caused by the conduction due to ions that
arise for impurities and from the dissociation of the insulating oil itself. Therefore, tan can be related to the volume
resistivity that is obtained from dc measurements:
tan = 100/
0
20
40
60
80
100 120 140 160 180
Temperature (C)
where f 0 is a constant (Hz), H is the activation energy (cal/

mol), R is 1.987 cal/mole K, and T is the absolute temperature (K). The value of H is about 20 kcal/mol for insulating
oil with 15% of aromatic compounds, and about 40 kcal/mol
for insulating oil without aromatic compounds.
Equation (2) applies over a rather narrow temperature
range. On the other hand, it is known that the following equation, due to Williams, Landel, and Ferry (WLF), applies over
a wider temperature range:
ln
(5)
where tan is in percent, (2f, where f is the frequency

of the voltage) is the angular frequency, is the dielectric
constant, 0 is 8.85 1014 F/cm, and is in ohm-centimeters.
Dissipative current is observed on applying a dc voltage
to the electrode that measures the volume resistivity of the
insulating oil. The volume resistivity in Eq. (5) is calculated
from the current i0 that is obtained at time t 0.
In recent years, a measuring device that can read tan

directly has been marketed. In its absence, however, tan
can be calculated from Eq. (5).
Values of the dielectric constant and tan of various insulating oils are shown in Table 2.
THERMAL TRANSFER CHARACTERISTICS
Cooling Method
Insulating liquids (oils) have not only excellent dielectric
characteristics but also significant cooling effects. Insulating
oils are important coolants in apparatus in which much heat
is produced. Nevertheless, external cooling systems must
sometimes be added, depending on the capacity of the apparatus and the load.
In the case of transformers, there are several types with
different cooling methods, such as the oil-immersed selfcooled type, forced-oil self-cooled type, direct-oil-flow selfcooled type, forced-oil forced-air-cooled type, direct-oil-flow
forced-air-cooled type, oil-immersed forced-water-cooled type,
and forced-oil forced-water-cooled type. Oil-immersed selfcooling is the simplest method. In this method transformers
are cooled by natural convection. In other methods heated insulating oils are cooled by forced air, forced oil, or water using
coolers or heat exchangers. In high-voltage and high-power
transformers the latter methods are widely used.
Table 2. Dielectric Constant and tan in Various

Insulating Oils a
Insulating Oil
Mineral oil
5-Chlorobiphenyl
3-Chlorobiphenyl
Chloroalkylene
Trixylenyl phosphate
Diethylhexylphthalate
Castor oil
Alkylbenzene
Alkyl naphthalene
Alkylbiphenylethane
Silicone oil
a
60 Hz 80C.
Dielectric Constant
tan (%)
2.18
4.25
5.20
5.05
6.00
4.55
4.00
2.17
2.48
2.51
2.52
0.02
0.03
0.04
0.01
1.0
0.10
0.30
0.005
0.01
0.10
0.008
LIQUID INSULATION
Heat Transfer Characteristics of Insulating Oils

When insulating oil cools insulating solids that cover a heat
source, such as the paper on transformer windings, the heat
flux through the solid to the oil is expressed in (2)
q = KS(T1 T2 )
K=
1
1
d
+A
Ta Ti = T +
T = 1.22
(8)
Tm temperature of insulating material

is generally obtained from the Nusselt number Nu by
(9)
The Nusselt number is a function of the Reynolds number Re,

the Grashof number Gr, the Prandtl number Pr, the temperature difference between the insulating material and the oil,
and the thickness of the insulating material:
l
2
g 3 (Tm T2 )
2
Pr =
a
Gr =
(14)
where
where
Re = v
(13)
The cooling characteristics of a duct model of transformers

with a sufficiently large reservoir of insulating oil have been
investigated for the case that the oil flow is laminar (3). The
average temperature of a winding, Ta, is expressed by
The larger q is, the more effective the cooling. The amount
of heat flux is controlled by K, and thus by . The magnitude
of depends on physical properties of the insulating oil, and
the structure of the insulating material. The important physical properties of the oil are the density, kinetic viscosity, thermal expansion coefficient, heat conductivity, and velocity.
The heat flux is also expressed by
Nu = const Pr1/3 Re1/2
(7)
q heat flux through solid to oil

K overall heat transfer coefficient
S area of solid that is in contact with oil
T1,T2 temperatures of heat source and oil respectively
heat transfer coefficient
A constant
d thickness of solid
heat conductivity of solid (insulating material)
Nu =
In the expression for Nu, Re always appears in the form

Ren (n 0). Therefore, generally speaking, the higher the velocity is, the higher is . For instance, in the case of a flat
plane exposed to forced convection of liquid, Nu is expressed
by
(6)
where
q = S(Tm T2 )
463
(10)
(11)
106 WH 1/2
= 0.858
gCd 3
106 W 7 Hd3 1/8

gC 6
(15)
(16)
where
Ti temperature of insulating oil at inlet of duct
temperature rise of winding above average temperature of insulating oil
T average temperature rise of insulating oil in duct
W thermal loading into duct (W/m2)
H height of duct (m)
d half the duct depth (m)
C specific heat of insulating oil (W s/kg)
density of insulating oil
As shown in Eq. (15) and Eq. (16), it is desirable for cooling
that the kinematic viscosity be low and that the thermal
expansion coefficient, density, specific heat, and thermal conductivity be high.
In Table 3 physical properties of some mineral oils, silicone
liquids, and high-molecular-weight hydrocarbon oils are
shown. Using these data, T and of three oils were calculated. Ratios of T and of silicone liquids and high-molecular-weight hydrocarbon oils to those of mineral oils are shown
in Table 4. The physical properties of the three oils are not
very different, except for the kinematic viscosity. The high
kinematic viscosity reduces heat transfer, as in the case of
silicone and high-molecular-weight hydrocarbon oils.
(12)
RESISTANCE TO IGNITION
where
l length of insulating materials
heat conductivity of insulating oil
v velocity of insulating oil
kinematic viscosity of insulating oil
g acceleration of gravity
thermal expansion coefficient of insulating oil
a thermal diffusivity of insulating oil
Because of increasing environmental problems due to the bioaccumulative nature of polychlorinated biphenyls (PCBs), the
production and use of PCBs have been prohibited throughout
the world. Almost all substitutes for PCBs are more flammable than the PCBs, and the evaluation of flammability of
those liquids becomes very important.
There are many test methods for the evaluation of the resistance to ignition and fire propagation of insulating liquids.
464
LIQUID INSULATION
Table 3. Physical Properties of Insulating Oils

Property
Heat conductivity (W/m K)
25C
100C
Specific heat (kJ/kg K)
25C
100C
Density [kg/(0.1 m)3]
25C
Thermal expansion coefficient (K1):
25C
Kinematic viscosity (m2 /s)
25C
100C
Naphthenic Oil
Paraffinic Oil
Silicone Liquid
High-MolecularWeight Hydrocarbon
0.120
0.109
0.132
0.116
0.136
0.128
0.134
0.119
2.05
2.33
1.96
2.33
1.49
1.65
1.63
1.90
0.87
0.86
0.96
0.88
7.8 104
7.8 104
1.04 104
8.0 104
11 106
2.1 106
12 106
2.2 106
50 106
16 106
350 106
16 106
Flash Point and Fire Point

Determination of the flash and fire points of petroleum products by the Cleveland open cup method is described in ISO
2592, and that of the flash points of petroleum products and
lubricants by the PenskyMartens closed cup method is described in ISO 2719. These are normative and very easy
methods, now being used worldwide and specified by various
standards for the evaluation of flammability of insulation liquids of almost all types.
Oxygen Index
Determination of the oxygen index of insulating liquids is described in IEC 61144. This method is an adaptation for liquids of ISO 4589, which has the same principles and is applicable to solids. The oxygen index is defined as the minimum
concentration of oxygen by percentage volume in a mixture of
oxygen and nitrogen that will just support combustion of a
material. The smaller the index, the more combustible a
tested liquid is.
Net Calorific Value
Linear Flame Propagation Test

The linear flame propagation test method, using a glass fiber
tape, is described in IEC 61197. A glass fiber tape impregnated with the sample liquid is ignited at one end, and the
time for the flame to travel between two lines is measured.
This method is very easy to set up and useful for impregnated systems.
The method to be selected for the evaluation of fire hazard
depends on the kind of machine and conditions of use. No
single method is always effective, and in many cases some
combination of the methods mentioned above is necessary.
For example, classification of insulating liquids according to
fire point and net calorific value is given in IEC 61100.
MOISTURE EFFECT
Determination of the net calorific value or net heat of combustion of liquid hydrocarbons with a bomb calorimeter described
in ASTM D 240 is specified in ISO 1928. This quantity represents the rate of heat generation by the liquid during its combustion.
Other Fire Tests
The pool fire test (large scale and small scale), trough test,
spray mist test, and heat release test, developed by Factory
Mutual Research, are attractive and practical methods to
Table 4. Ratios of T and of Silicone Liquids and HighMolecular-Weight Hydrocarbon Oils to Those of Mineral Oils a
Property
Mineral Oil
Silicone Liquid
High-MolecularWeight Hydrocarbon
1.00
1.00
2.75
1.18
2.99
1.26
evaluate the ease of ignition and behavior of liquids during

combustion, but the required equipment is bulky and requires
ample space.
Except for the density and thermal expansion coefficient, characteristics at

100C were used for the calculations.
Moisture in Insulating Oil

Moisture in insulating oil leads to a decrease of volume resistivity and to dielectric breakdown. The moisture content is
related to the humidity in the atmosphere and also changes
with the oil temperature. Standards for it are given in JIS C
2101, BS 2511, IEC Publ. 733, and also ASTM D 1533. One
method of measuring it uses Karl Fischers reagent, which
reacts sensitively with very small quantities of moisture. The
reactions between Karl Fischers reagent and water are as
follows:
I2 + SO2 + 3C5 H5 N + H2 O 2C5 H5 N HI + C2 H5 N SO3
C2 H5 N SO3 + CH3 OH C2 H5 NH OSO2 OCH3
A liquid mixture of chloroform and methanol is used for the
titration solvent.
Dissolved water content in liquids in general is determined
by the relative humidity and the temperature of the atmo-
sphere. The relation is expressed by Henrys law,

X = Xmax = KPW
K=
(17)
Xmax
PWsat
(18)
where X is the moisture density (mol %), Xmax is the largest

possible dissolved water content (mol %), is the relative humidity (%) of the atmosphere, PW is the vapor pressure, and
K is Henrys constant, and PWsat is the saturated vapor pressure. Figure 2 shows the relation between dissolved water
content in the insulating oil and the relative humidity of the
atmosphere at various temperatures. Generally for synthetic
oils the dissolved water content is larger because their hydrophilic nature is stronger than that of mineral oil.
When moist insulating oil is cooled, the dissolved moisture
becomes supersaturated at a certain temperature and turns
to white mud. The dielectric breakdown voltage decreases
sharply when that happens. The usefulness of the insulating
oil depends both on the degree of saturation and on the absolute moisture content.
Moisture in the Insulating Oil and
Dielectric Breakdown Voltage
Moisture in the insulating oil not only decreases the dielectric
breakdown voltage and volume resistivity, but also corrodes
metals and degrades the insulating oil. Moisture in the insulating oil can be due to oxidative deterioration of the oil itself
as well as to hygroscopic absorption from the atmosphere.
The temperature dependence of the dielectric breakdown
voltage in transformer oil, alkylbenzen, and PCBs is de-
Moisture in insulating oil ( L/L)
100
465
12 L/L
80
33
60
55
40
106
20
0
40
20
0
20
Temperature (C)
40
60
Figure 3. Temperature dependence of dielectric breakdown voltage

in transformer oil. Nitrogen sealed-off transformer; transformer with rubber-bag conservator.
scribed by a U-shaped curve with a minimum at 10C to 0C

(4). The dependence for transformer oil is shown in Fig. 3.
For 30 L/L moisture in insulating oil, the white mud temperature is the same as the temperature at which the dielectric breakdown voltage shows a minimum value. The dielectric breakdown voltage at ordinary temperatures decreases
with increasing water content in the oil up to about 100 L/
L, and is essentially constant for greater water content.
Although fiber and dust in dehydrated oil decrease the dielectric breakdown voltage to some extent, in hygroscopic oil
they decrease it dramatically.
AGING EFFECTS
Oxidation Stability of Insulating Oil
400
300
80C
200
60C
50C
100
40C
25C
0C
Dielectric breakdown voltage (kV)
LIQUID INSULATION
20
40
60
80
100
Relative humidity of atmosphere (%)

Figure 2. Relation between moisture in insulating oil and relative
humidity of atmosphere.
The insulating oil reacts with oxygen and generates organic

acids, sludge, water, and other matter. These contaminants
significantly degrade the electrical characteristics of the insulating oil. The organic acid and the moisture advance corrosion and other deterioration of materials in contact with the
oil, and sludge lowers its cooling efficiency. Therefore, oxidation stability of the insulating oil is important for the life and
reliability of oil-filled electrical equipment. It is known that
the copper used for conductors in electrical equipment can
catalyze the oxidative deterioration of the insulating oil (5,6).
The oxidation stability of the insulating oil is evaluated
from the amount of sludge, total acid number, and electrical
characteristics after heating the insulating oil in contact with
excess oxygen and copper. Test methods are described in IEC
Publications 74, 474, 813; ASTM D 1313, 1934, 2112, 2440;
BS 148; DIN 51554; and JIS C 2101. Tests are carried out at
100C to 120C. Since ASTM D 1313 uses no catalyst, that
test is done at the highest temperature (140C).
The oxidation stability of mineral oil is influenced by its
degree of refinement. However, higher refinement does not
necessarily mean higher stability. Research on improved oxidation stability of insulating oil is often done from the viewpoint of optimum aromaticity.
tan of Insulating Oil in Relation to Oxidative Deterioration
The insulating oil of oil-immersed electrical devices that are
equipped with oxidation deterioration prevention devices does
466
LIQUID INSULATION
0
10
Operation (year)
10
15
20
when the number of positive ions is increased by the flow of

liquid. This static electrification is called streaming electrification.
25
tan (%) at 80C
Mechanisms of Streaming Electrification

The streaming electrification process arises from electric
charge motion, separation, and relaxation (Fig. 5). These processes happen simultaneously:
1
0.5
1. Electric Charge Motion. Certain ions in the liquid are

adsorbed on a solid in contact with it.
2. Electric Charge Separation. Other ions are carried off
by flow, and an imbalance of positive and negative ions
occurs.
3. Electric Charge Relaxation. Surplus ions are neutralized, and the imbalance is canceled.
0.1
tan (%) at 95C
0.05
1.0
Properties of Streaming Electrification

When the following three conditions coexist, large streaming
electrification is observed:
0.5
A: Field data
B: Laboratory Data
0
10
15
20
Aging time (h)
1. It is easy to move an electric charge, because a solid

surface is active.
2. Charges are easily separated.
3. Surplus electric charge is not easily canceled.
25
Figure 4. tan behavior for actual transformer oil in the field (A)
and insulating oil in laboratory data (B).
not come in direct contact with the atmosphere. Therefore the

oxidative deterioration of the insulating oil is slow. As shown
in Fig. 4, the temperature dependence of tan for insulating
oil shows peculiar behavior.
In Fig. 4, the band marked A shows the deterioration of
tan in the insulating oil in a nitrogen-enclosed transformer
and in a sealed transformer, at 80C, gathered from many
transformers over years of operation. The time dependence of
tan yields an N-shaped curve. A peak is observed at 5 to 7
years after the start of operation. This behavior is confirmed
in laboratory experiments. It implies that oxygen and copper
play an important role.
Curve B shows laboratory data on the deterioration of insulating oil where the copper surface area was 44.8 cm2 per
100 mL of oil, the oxygen volume was 5 mL per 100 mL of oil,
and the oil temperature was 95C (7).
From comparison of curves A and B, one hour of deterioration as accelerated in the laboratory is seen to be equivalent
to about one year in operation.
FLOW-INDUCED ELECTRIFICATION
Insulating oil, like petroleum system liquids such as gasoline,
toluene, and kerosene, has high volume resistivity. In such
insulating liquids, ionic compounds AB dissociate into A
and B, and the ions become charge carriers:
A+ B A+ + B
When any one of the three processes is hindered, streaming electrification can be prevented. The choice of insulating
oil can also affect streaming electrification.
Polarity of Streaming Electrification
In streaming electrification between insulating paper and insulating oil, the oil becomes positively charged and the insulating paper negatively charged. A possible reason is a peculiarity of the oxygen of the hydroxyl group (OH) in the
insulating paper (cellulose).
Oxygen, having high electronegativity (that is, ability to
attract electrons), attracts the electron of hydrogen. It thereby
becomes negatively charged, and the hydrogen becomes positively charged. The cellulose molecule surface is covered with
positively charged hydrogen, which adsorbs negative ions in
oil selectively. Therefore, the insulating oil becomes positively
charged, and the insulating paper is negatively charged.
Streaming Electrification and the Deterioration
of Insulating Paper
The hydroxyl group (OH) of cellulose is changed to the aldehyde group (CHO) or the carboxyl group (COOH) by oxidative deterioration. The extent of polarization due to electron
transfer from hydrogen to oxygen, mentioned above, is in the
following order:
hydroxyl group < aldehyde group < carboxyl group
Accordingly, streaming electrification increases as the insulating paper deteriorates.
Streaming Electrification of Transformers
There is no charging when the numbers of A and B are

equal. However, for example, static electrification is observed
Streaming electrification became a problem for the first time

with large-capacity transformers (8). Accidents (9,10) due to
LIQUID INSULATION
+
+
Adsorption
Flow
Relaxation
streaming electrification, and also direct observation of

streaming electrification (9) in transformers have been reported. A good measure of the streaming electrification in a
transformer is the coil leakage current.
Figure 6 shows the oil temperature dependence of the coil
leakage current, and reveals that it is influenced by differences in the insulating oil.
Coil leakage current is significant at all operating temperatures in large-capacity transformers. Therefore, measurement of the streaming electrification is essential in transformer design with respect to internal structure, oil flow rate,
and insulating material. One may also select an oil that resists charging and charge separation.
MINERAL OILS
Crude Oils
Mineral insulating oils have a long history and have been
used for transformers, cables, capacitors, and circuit breakers. They are manufactured by refining crude oils. The main
compounds in crude oils are naphthenic hydrocarbons, paraffinic hydrocarbons, and aromatic hydrocarbons. Small
16
Coil leakage current ( A)
Oil 2
12
8
Oil 1
0
0
20
Insulating oil
+
+
+ +
+
+ +
+
+
+
+
Insulating paper
Insulating paper
Insulating oil
Insulating paper
+
+
(2) Electric charge

relaxation
+
+
+
+ +
+ +
+ +
+
+ +
+
+ +
(2) Electric charge

separation
Insulating oil
(1) Electric charge

motion
467
40
60
80
Oil temperature (C)

Figure 6. Oil temperature dependence of coil leakage current.
Figure 5. Sketch of the mechanisms of

streaming electrification.
amounts of sulfur compounds, nitrogen compounds, and oxygen compounds also occur. The naphthenic hydrocarbons include dicyclic, tricyclic, and alkyl-substituted hydrocarbons;
the paraffinic hydrocarbons include normal paraffinic and isoparaffinic hydrocarbons; and the aromatic hydrocarbons include dicyclic, tricyclic, and alkyl-substituted hydrocarbons.
The composition of crude oils depends on the area where
they are produced. There are three kinds of crude oils: naphthenic, paraffinic, and mixed. Naphthenic crude oils contain a
large amount of naphthenic hydrocarbons, and paraffinic
crude oils contain a large amount of paraffinic hydrocarbons.
Mixed crude oils are intermediate between naphthenic and
paraffinic. Naphthenic crude oils are produced in South
America, North America, and southern Asia. Paraffinic crude
oils are produced in some areas of North America and northern Asia. Mixed crude oils are produced in the Middle East.
The composition of mineral oils depends on that of the
crude oils from which they are manufactured. There are two
kinds of mineral oils: naphthenic and paraffinic.
Refining Process
Mineral oils are manufactured from distillate of heavy light
oil and light lubricant oil by the process shown in Fig. 7.
Where naphthenic oils are refined, acid treatment followed by
clay filtration is also used. In the case of paraffinic oils dewaxing is part of the refining process. Examples of the composition of a naphthenic oil and a paraffinic oil are shown in Table
5. It is seen that in both paraffinic and naphthenic oils the
amount of paraffinic compounds is greater than the amount
of naphthenic compounds.
To obtain good dielectric characteristics, the amounts of
nitrogen compounds and sulfur compounds should be as small
as possible. However, excessive refining also decreases the
amount of aromatic hydrocarbons. Decrease of the amount of
aromatic hydrocarbons means a decrease in hydrogen absorption, and decrease of the amounts of both aromatic hydrocarbons and sulfur compounds means a decrease in oxidation
stability. It is known that hydrogen adsorption relates to the
partial discharge characteristics of oil (11,12) and that aromatic hydrocarbons have high hydrogen absorption. It is also
known that coexistence of aromatic hydrocarbons and sulfur
compounds is effective for oxidation stability. Therefore, refining must be performed so that the insulating oils maintain
balanced characteristics. The optimum amount of aromatic
hydrocarbons is 10 wt% to 20 wt%. In this case the amount
468
LIQUID INSULATION
Naphthenic oil
Atmospheric
distillation
Vacuum
distillation
Solvent
refining
Clay
filtration
Hydrogenation
Paraffinic oil
Atmospheric
distillation
Vacuum
distillation
Solvent
refining
Hydrogenation
Dewaxing
Clay
filtration
Figure 7. Refining of insulating oils.
of sulfur compounds is at least 0.5 wt%. Among them there

are some corrosive sulfur compounds, but the amount of those
is very small.
In some cases, oxidation stability can be corrected through
the addition of oxidation inhibitors.
Characteristics of Mineral Oils
In Table 6 characteristics of some naphthenic and paraffinic
transformer oils are shown. The higher pour point of the paraffinic oil is owing to the larger amount of paraffinic hydrocarbons in it. In the 1970s and 1980s the dielectric properties,
compatibility with insulating materials, thermal stability,
and other properties of paraffinic oils were investigated, and
it was proved that various properties of paraffinic oils are by
no means inferior to those of naphthenic oils, except for their
higher pour points. To lower the pour points of paraffinic oils,
pour point depressants are added. Mixing of alkylbenzenes
with paraffinic oils is also effective for this purpose. An example of mixture of alkylbenzene and paraffinic oil is shown in
Table 6. In this mixture a pour point depressant has also been
added. The lowering of the pour point is seen.
SYNTHETIC OILS
PCBs were among the best and most widely used synthetic
insulating liquids for electrical machines, such as power capacitors and transformers, due to their superb electrical characteristics and nonflammability, until a total ban on their use
and production was imposed, first in Japan in 1972, then in
the USA in 1976, and then in Europe in 1985.
In the 1960s alkylbenzenes were initially developed for
high-voltage cables in view of their superior gassing properties under high voltage stress, and especially for use with synthetic paper.
At the time PCBs were banned, other kinds of synthetic
aromatic hydrocarbons such as alkylnaphthalenes and alkyldiphenylethanes had been developed as candidates for improvements on mineral oils, but because of their higher cost,
they had not been put into practical use. PCBs were then replaced mainly by these new aromatic hydrocarbons.
Aromatic Hydrocarbons
Alkylbenzenes consist of a benzene ring and an alkyl group of
the straight-chain or branched-chain type. Alkylbenzenes are
Applications of Mineral Oils

Mineral oils are used for transformers, oil-filled (OF) cables,
pipe-type oil-filled (POF) cables, capacitors, and circuit breakers. The greatest use is in transformers. Specifications for
mineral oils are given in IEC 60296, 60465 and ASTM D3487,
D2297, D1818.
Table 5. Examples of Composition of Insulating Oils
Type of Oil
Paraffinic
Naphthenic
Proportion of C (%)
Sample
No.
Paraffinic
Naphthenic
Aromatic
1
2
3
60.1
59.9
61.8
29.7
27.5
29.7
10.2
12.6
8.3
1
2
3
45.1
49.0
50.7
36.3
39.0
40.8
18.6
12.0
8.5
Table 6. Characteristics of Mineral Transformer Oils

Property a
Flash point (C)
Kinematic viscosity
(m2 /s)
40C
100C
Pour point (C)
Permittivity
80C
tan (%)
80C
( m)
80C
a
b
Paraffinic
Oil 1
Paraffinic
Oil 2 b
Naphthenic
Oil
144
148
140
7.8 106
2.2 106
25
9.6 106
2.5 106
45
7.9 106
2.1 106
45
2.1
2.13
2.16
0.01
0.01
0.01
1013
1013
1013
tan dissipation factor; volume resistivity.

Containing azkylbenzene and pour-point depressant.
LIQUID INSULATION
469
mainly used for high-voltage cables, including cables with

synthetic paper, as they have excellent compatibility with
plastics. They can be blended with mineral oil to improve its
gassing properties and thermal stability.
In IEC 60867, alkylbenzenes are divided into three classes
according to their viscosity and flash point.
Alkyldiphenylethane (phenylxylylethane), alkylnaphthalene (diisopropylnaphthalene), and methylpolyarylmethanes
(blends of benzyl- and dibenzyltoluene and blends of benzyltoluene and diphenylethane) are mainly used for high-voltage
power capacitors and also used for instrument transformers.
Because all of these liquids consist of two benzene rings with
much shorter-chain alkyl groups than in alkylbenzene, their
aromatic contents are higher than those of alkylbenzene, and
their resistance to partial discharge is very high. They have
excellent dielectric properties and also good compatibility
with plastic film, especially with the polypropylene film currently used for capacitors; but all of them have lower permittivity (2.2 to 2.5) and flash point (130C to 150C) than
PCBs. In some cases, they are stabilized by epoxiside or antioxidant. Their properties are also specified in IEC 60867.
uids are now used mainly in hollow cables (pipe-type cables)

and to some extent in low-voltage capacitors. Polybutenes are
specified in IEC 963 and classified into three classes, mainly
depending on their viscosity. They are selected according to
the specific application.
Silicone Liquids
LIQUIDS FOR POWER TRANSFORMERS
The silicone liquids currently used for electrical machines are

polydimethylsiloxanes and have a variety of viscosities and
flash points. Properties of a silicone liquid with a kinematic
viscosity of 40 mm2 /s at 40C are specified in IEC 60836. They
are used mainly for special transformers, due to their good
thermal stability at higher temperature and better flow at
lower temperature than mineral oil and because they are not
very flammable. They are also sometimes used for capacitors
and cables.
Transformers were developed and began to be manufactured

in the mid 1880s in Hungary, the USA, the United Kingdom,
and France. In the years 1886 to 1891, manufacturers began
to use oils for insulation transformers. Such oils (transformer
oils) are specified in IEC 60296 and 60836 and in ASTM
D3487 and D4652.
Transformer oils must have the following properties:
Organic Esters
Dioctylphthalate (DOP) and diisononylphthalate (DINP) have
been used as substitutes for PCBs, especially for capacitors,
because they have higher permittivity (4.5 to 5.5) and flash
point (200C to 240C) than aromatic hydrocarbons. Di-2ethylhexyl orthophthalate (DOP) is specified in IEC 61099 as
a capacitor ester (type C1). As not easily flammable liquids,
phosphoric acid esters such as tricresyl phosphate (TCP) and
trixylenyl phosphate (TXP) are used as blends with aromatic
hydrocarbons. Generally speaking, these esters have high
permittivity and high inherent resistance to electrical stress,
but as manufactured they contain much water and impurities
and their dielectric dissipation factor is very high, so they
must be carefully dehydrated and purified before impregnation and often need an antioxidant or scavenger.
Recently, organic tetraester liquids have been introduced
in transformers because they are less flammable. Their fire
point is higher than 300C, but their viscosity is low compared
with that of currently used mineral oils. The same precautions should be followed as mentioned above, and additives
are effective as in the case of other organic esters.
Tetrahydric alcohol and a mixture of monocarboxylic acid
with suitable stabilizing additives are also specified in IEC
61099 (type T1).
Polybutenes
Polybutenes can have a large range of viscosity (1 mm2 /s to
105 mm2 /s at 40C), depending on polymerization. These liq-
VEGETABLE OILS
Vegetable oils (castor oil, rapeseed oil, etc.) are basically triglyceryl esters of fatty acids, and the fatty acids can be saturated or unsaturated. They were once used for cables and capacitors, and are now mostly used for the impregnation of dc
capacitors and especially energy storage capacitors, as they
have high permittivity. They have not been used for ac power
capacitors, as they have poor dielectric dissipation factors. Recently, however, they have been tried for use with metallized
polypropylene films, with which they have good compatibility,
and their dissipation factor and gas-absorbing ability have
been improved by blending them with aromatic hydrocarbon
liquids.
1.
2.
3.
4.
5.
6.
High dielectric strength and low dielectric losses

Good cooling power (mainly dependent on viscosity)
High chemical stability and high resistance to oxidation
Good compatibility with insulating materials
Low corrosive sulfur content
Low viscosity and good fluidity over a wide temperature
range (low pour point)
7. Sufficient source of supply
8. High flash point
9. Nontoxicity
Of these, properties 1 and 2 are most important from the
viewpoint of transformer performances.
Mineral Transformer Oils
Mineral oils have been used as transformer oils since the beginning of their manufacture. When properly refined, mineral
oils have various excellent properties mentioned above. At
present, mineral oils are used over wide range of transformer
capacity, from distribution transformers to ultrahigh-voltage
transformers.
Mineral oils are manufactured by refining crude oils. Depending on the composition of the crude oils, there are two
kinds of mineral oils: naphthenic and paraffinic. The pour
points of paraffinic oils are generally higher than those of
naphthenic oils.
Sometimes mineral oils are mixed with each other or with
other oils. The mixtures may be between oils of the same type,
470
LIQUID INSULATION
between naphthenic and paraffinic oils, or with nonmineral

oils. Some specifications can be found in IEC 60296. In some
countries, mixtures of mineral oils and alkylbenzenes are
used as transformer oils. Such oils have high resistance to
oxidation, low corrosion, and low pour point. In the case of
paraffinic oils, because of their relatively high pour points,
pour-point depressants are added.
Because oils are oxidized under air, small amounts of antioxidants are added to some mineral oils, especially in Europe
and North America. Such mineral oils are classified in IEC
60296 and ASTM D3487. However, in some countries mineral
oils with antioxidants are not used.
In 1970s some flashover faults were found in ultrahighvoltage transformers due to flow-induced electrification
(streaming electrification). Factors that affect this phenomenon are transformer design (especially the flow rate of the
oil), oil temperature, and properties of the insulating oils such
as the volume resistivity and electrostatic charging tendency.
Flow rates of oil have been controlled in some transformers to
suppress this phenomenon. It is said that 1,2,3-benzotriazol
(BTA), which has been known as a deactivator agent for metals, suppresses this phenomenon. In some countries a small
amount of BTA has been added to mineral oils for high-voltage and high-power transformers for that purpose.
Synthetic Transformer Oils
Ordinarily flash points of mineral transformer oils are around
150C. Therefore, mineral oils are not so desirable for transformers in trains and indoor substations. For those uses it is
desirable to use nonflammable or less-flammable transformer
oils. PCBs are nonflammable and are the most desirable oils
for such applications. However, PCBs are no longer environmentally acceptable. Since they were banned, no transformer
oils have been found that have the desired nonflammability.
Silicone (polydimethylsiloxiane) liquids have been put into

use. These liquids (described in IEC 60836) have high fire
points and good oxidation resistance, and are classified as
less-flammable liquids in the National Electrical Code in the
USA. They are often used for transformers of trains, and in
some countries they have been used for distribution transformers.
Some polyolester liquids (described in IEC 61099) are used
for transformer oils on account of their good thermal stability
and low hydrolysis in the presence of water. Midel 7131 (The
Micanite and Insulators Co.) and Enviro Temp 100 (RTE Co.)
are examples. Mixtures of flon 112 and tetrachloroethylene
such as Formel.NF (ISC Chemicals Ltd.) have been developed
for transformer use. This liquid has environmental problems
because of the flon 112. However, tetrachloroethylene is nonflammable, and it and its mixtures with mineral oils have
been classified as nonflammable by Factory Mutual.
As previously mentioned, high-molecular-weight hydrocarbons with fire point higher than 300C are classified as lessflammable oils in the National Electrical Code and are used
for transformer oils. In Table 7 properties of some transformer oils are shown.
LIQUIDS FOR POWER CABLES
Oil-immersed power cables were developed and put into use
in the 1880s, and a historic milestone in recent engineering
and industrial progress was established by the invention and
development of the oil-impregnated or oil-filled (OF) power
cable by Emanuelli in 1923. OF cables are impregnated with
oils without voids or moisture and then hermetically sealed
to avoid damage and harmful effects from the surroundings.
From the early stage of OF cables, naphthenic oils have
been mainly used because of their low pour point and high
Table 7. Properties of Some Insulating Oils

Property a
Flash point (C)
Fire point (C)
Ignition point (C)
Specific heat (J/kg K)
Specific gravity:
25C
Kinematic viscosity (m2 /s):
25C
100C
Pour point (C)
Permittivity:
25C
tan (%):
25C
100C
BDV (kV/2.5 mm):
25C
( m):
23C
25C
80C
Silicone Liquid
Ester Oil
150
160
332
209 103
314
360
1.42 103
242
304
415
1.76 103
0.89
0.96
0.98
1.62
1.62
11 106
2.5 106
50
50 106
16 106
55
76 106
31 106
45
0.88 106
33
0.8 106
22
2.2
2.7
3.2
2.36
2.37
0.01
0.09
0.01
0.009
0.24
(80C)
0.1
(23C)
0.5
(50C)
55
60
64
70
43
8 1012
6 1013
7.6 1012
1 1013
1.8 1011
tan dissipation factor; BDV breakdown voltage; volume resistivity.

Flon 112 tetrachloroethylene.
c
Nonflammable.
a
b
Mixture b
Mineral Oil
Nonc
Non
Non
Tetrachloroethylene
Non
Non
Non
LIQUID INSULATION
stability under high stress, but with the progressive improvement of process technology for refining crude oil, paraffinic
crude oils and mixtures of naphthenic and paraffinic oils have
also been used because of their wider availability.
Aromatic content in mineral oil is also important, and in
some cases synthetic aromatic hydrocarbons are added. Pure
synthetic aromatic hydrocarbons, mainly alkylbenzenes, are
also used, especially for ultrahigh-voltage power cables, because of their compatibility with synthetic papers, excellent
stability under high stress, and sufficient source of supply.
Polybutenes are used for hollow power cables because of
their wide range of viscosity.
Liquids for cables are specified in IEC 60465 (mineral oils),
60836 (silicone liquids), 60867 (aromatic hydrocarbons), and
60963 (polybutenes).
Cable oils must have the following properties.
1. High dielectric strength and high volume resistivity
2. Low dielectric losses and low dielectric constant
3. Low viscosity and good fluidity over a wide temperature
range (low pour point)
4. High chemical stability and high resistance to oxidation
5. Low temperature coefficient of expansion
7. Nontoxicity and environmental safety
Of these, properties 1, 2, and 3 are most important from the
viewpoint of power cable performance.
LIQUIDS FOR SWITCHGEARS

Liquids for switchgear (switchgear oils) must have arc suppression properties and high dielectric strength. Arc suppression properties are basically due to the high thermal conductivity of hydrogen gas produced by the decomposition of
switchgear oils. Thus it is desirable that liquids easily produce hydrogen gas and that the amount of free carbon produced by their decomposition be small. Good insulation, requires not only high dielectric strength, but also rapid
insulation recovery after interruption of electric arcs.
Besides these properties, it is desirable that switchgear
oils have chemical stability to maintain good dielectric properties, and that they be compatible with the solids used. Insulating oils that have the above-mentioned properties are mineral oils. Switchgear oils are specified in IEC 60296 and
ASTM D387. They are classified in the same classes as transformer oils.
The kinematic viscosities of insulating oils in these classes
are relatively low: for insulating oils at 40C classified in IEC
60296 as Class I and Class IA, Class II and Class IIA, and
Class III and Class IIIA they are 16.5 106 m2 /s, 11.0
106 m2 /s, and 3.5 106 m2 /s, respectively. The kinematic viscosities of insulating oils at 40C classified in ASTM
D3487sa Type I and Type II are 12.0 106 m2 /s. Low kinetic viscosity allows mechanical parts of switchgears to perform freely, and oil flows owing to hydrogen gas evolved by
decomposition of switchgear oils to be easily produced and facilitate arc suppression.
471
LIQUIDS FOR HIGH-VOLTAGE CAPACITORS

The capacitor is one of the oldest electrical components. An
oil-immersed capacitor was developed in the 1850s, but the
fundamental technologies of modern oil-impregnated and oilfilled high-voltage power capacitors originated with those of
high-voltage power cables.
After PCBs were developed around 1930, PCBs (mainly
trichlorobiphenyl for high-voltage power capacitors) were
used for can-type capacitors and mineral oils were used for
large tank-type capacitors, until PCBs were recognized as environmentally hazardous. PCBs were also used for a short period as the impregnant for plastic film (polypropylene) dielectrics of both paperfilm and all-film types. Just after the ban
on PCBs, those impregnants were replaced by aromatic hydrocarbons. As the aromatic contents of these hydrocarbons
are very high, they are very suitable for the impregnation of
high-voltage capacitors with sharp-edged foil electrodes.
Because aromatic hydrocarbons are more flammable than
PCBs, silicone or blended oils of aromatic hydrocarbons and
phosphoric acid esters have been used for high-voltage capacitors as less-flammable liquids for limited use; but recently dry
capacitors have been developed for use where fire-resistant
materials are strictly required.
To minimize the dielectric thickness, self-healing technology originally developed for low-voltage capacitors has recently been applied also for high-voltage capacitors. In this
case, as metallized paper or film is used, and therefore compatibility between the impregnant and the solid material is
very important, impregnants such as organic esters are used.
Liquids for high-voltage power capacitors are specified in
IEC 60836 (silicone liquids), 60867 (aromatic hydrocarbons),
and 61099 (organic esters).
Liquids for high-voltage capacitors must have the following properties:
1. High dielectric strength and high volume resistivity
2. Low dielectric losses and high dielectric constant
3. High stability under high voltage stresses and high partial discharge resistance
4. Good compatibility with film materials
5. High chemical stability and high resistance to oxidation
6. Low temperature coefficient of expansion
7. Nontoxicity and environmental safety
Of these, properties 1, 2, and 3 are most important from the
viewpoint of high-voltage capacitor performance.
BIBLIOGRAPHY
1. C. H. Fluschein (ed.), Power Circuit Breaker, Theory and Design,
IEE Power Engineering Series 1, Stevenage, UK: Peregrinus,
1982.
2. N. Ohoka and S. Maekawa, Transformers (in Japanese), Tokyo:
Tokyo Denki University Press, 1975.
3. R. K. Grubb et al., A transformer thermal duct study of various
insulation liquids, A-80 051-3, presented at IEEE PES Winter
Meeting, New York, 1980.
472
LIST PROCESSING
4. T. Ishii and M. Ueda, Effect of moisture on ac breakdown voltage

of insulating oils, Trans. Inst. Electr. Engl. Jpn. A, 92: 154158,
1972.
5. J. J. Melchiore and I. W. Mills, Factors affecting stability of electrical insulating oils, IEEE Trans. Electr. Insul., 2 (3): 150155,
1967.
6. A. K. Vijh, M. Duval, and J. P. Crine, Open-circuit corrosion as
the origin of the electrical instability arising from the dissolved
metal content in insulating oils in EHV current transformers,
Mater. Chem. Phys., 19 (4): 392, 1988.
7. T. Miyamoto, A. Kishi, and Y. Miura, Effect of insulating oils
streaming electrification in transformer, EPRI EI-6918, Project
149998, Proc. No. 3-1-1, 1990.
8. T. Takagi et al., Reliability Improvement of 500 kV Large Capacity
Power Transformer, CIGRE SC-12, Committee Report No. 1202, 1978.
9. T. Takagi et al., Reliability improvement of transformer, IEEE
Trans. Power Appar. Syst., PAS-99 (3): 1097, 1980.
10. M. Yasuda et al., Suppression of Static Electrification of Insulating
Oil for Large Power Transformer, 197-2, presented at IEEE PAS
Winter Meeting, 1982.
11. International Electrotechnical Commission, Gassing of insulating
liquids under stress and ionization, IEC Publ. 60628, 1985.
12. American Society for Testing and Materials, Test method for gassing of insulating oils under electrical stress and ionization,
ASTM D2300, 1991.
MINEAKI NISHIMATSU
Fukui University of Technology
TERUO MIYAMOTO
Mitsubishi Electric Corporation
TOSHIO SUZUKI
Aichi Electric Co., Ltd.
MEDICAL EXPERT SYSTEMS

The eld of medical informatics (also termed health informatics) concerns application of information science and information technology to health care, clinical care, education
and biomedical research. Most countries have national societies in this area, and some 40 of them are organized in
the International Medical Informatics Association (IMIA).
Articial intelligence (AI) methods specically refer to the
application of computer-based programs simulating human experts. Recent developments in medical informatics
benet from the availability of powerful personal computers (workstations), advanced information processing techniques such as the articial neural network (ANN), and
increased acceptance by the clinical community. The latter
seemingly trivial factor should not be underestimated by
engineers. In fact, wider acceptance is only partly due to
improved user interfaces, but largely by the gradual recognition that computers form a useful tool in the doctors ofce. In 1995, still fewer than 1% of the family practitioners
in the United States use a computerized patient record, but
enthusiasm is increasing. It has been shown that patient
satisfaction does not decrease when the computer is employed in the physicians examination room (1). Information, in general, requires that locally available knowledge
can be communicated. Indeed, facts are only meaningful if
they can be uniquely described and successfully transmitted from one location or person to another. Trivial examples from everyday life concern the combination of coding
of messages by writing news reports and the distribution of
newspapers, and the formulation of integrated weather reports and subsequent radio broadcasting. Similar lines of
communication apply to medical informatics, although the
implementation of advanced techniques started not earlier
than around 1975. To understand this delay that surprisingly impeded an important issue such as medical care,
and also to appreciate the potential progress that can be
realized, it is essential to indicate the circumstances that
make health care differ from other areas in the natural sciences. First, the primary information stems from humans
(or animals for the sake of veterinary informatics) inicted
with shortcomings regarding their functioning. Second, in
medicine it is difcult to dene what is normal. Normality does not refer to a single numerical value, but rather
to a certain range dened by reference values. Therefore
it seems almost impossible to dene a deviating process to
begin with, then to assess the severity of any abnormality,
to judge whether the defect is dangerous for your health,
next to evaluate the impact of therapeutic intervention,
and nally to determine the prognosis for each individual.
Clearly, a vast number of communication steps are to be
taken, thus limiting the efciency of the process. Moreover,
existing knowledge on a particular disease may not be immediately available to any physician, because it is an impossible task to scrutinize weekly or monthly all medical
journals published anywhere in the world. Further limitations regarding communication of medical knowledge
refer to clinical terminology and classication of health
data. With the knowledge that medical informatics deals
with enormous amounts of data, often located at widely dis-
tributed locations, it is not surprising that computer support in this area will be of great impact on efciency, accuracy, and advancement of health care. Many projects, often
concerted international efforts, address the issue of how to
handle an ever-increasing amount of medical information.
Universal classications have been designed and regularly
rened, while other approaches aim not only to collect, but
also to structure and disclose this exponentially growing
body of medical information. The following sections will be
devoted to a more general description of various topics of
relevance to the eld of medical informatics and may be of
interest for the average reader.
PATIENT DESCRIPTION AND THE ELECTRONIC
PATIENT FILE
The basic goals of the use of computers in medicine concern communication and clinically relevant combination of
data. This electronic medium is expected to enhance and
facilitate such interaction and data interpretation. Ideally,
every citizen should carry a patient data card, which in
an emergency case presents valuable information to the
physician. The Medical Records Institute is an instrumental force in the movement toward such an electronic patient
record. Locally, most hospitals have developed an information system [hospital information system (HIS)]. A patient
card may include information on medical history, familial traits, use of prescription drugs, allergies, lifestyle (including sports activities and use of alcohol and/or tobacco),
availability of x-ray pictures, electrocardiogram recording,
and blood chemistry (2). Obviously, these initiatives involve
delicate ethical issues, as well.
MEDICAL TERMINOLOGY AND EPONYMS
Knowledge obviously can be represented by symbols,
words, denitions, and their interrelations. Knowledge
may be expressed by spoken or written words, ow charts,
(mathematical) equations, tables, or gures. Aspects of language and text interpretation are central issues in AI. A
powerful abstraction of language also provides a powerful
representation of knowledge. Various strategies have been
explored: semantic networks offer a versatile tool for representing knowledge of virtually any type that can be captured in words by employing nodes (representing things)
and links (referring to meaningful relationships), thus expressing causal, temporal, taxonomic, and associational
connections. Other approaches (such as frame systems and
production rule systems) have also been investigated. Conceptual graphs (3) are an emerging standard for knowledge
representation, and the method is particularly suited to
the representation of natural language semantics. Freetext data have limitations due to spelling errors, ambiguity, and incompleteness. However, formalisms that collect
data in a structured and coded format are more likely to
increase the usefulness regarding biomedical research, decision support, quality assessment, and clinical care (4).
However, the lack of standardized medical language limits
the optimal use of computers in medicine. Incorporation of
knowledge bases containing equivalent expressions may
Medical Expert Systems
be required for the practical use of medical information

systems (5). The Generalized Architecture for Language
Encyclopaedias and Nomenclature in medicine (GALEN)
project, funded by the European Union, develops the architecture and prototypes for a terminology server (6). Indeed, medical language forms one of the greatest obstacles for the practical use of AI in the eld of medicine (5,
7). Natural language often has remote roots, for example,
adrenaline and epinephrine are the same chemical substances. Similarly, (pontine) angle tumor, acoustic (nerve)
neurinoma, and acoustic neurilemmoma all have the same
meaning. The Latin word os means both mouth and bone.
The terms heterogenous, heterogeneous, and heterogenic
look similar but all have a different meaning. Also, several
English words have a dual meaning, for example, apprehension, aromatic, attitude, auricle, bladder, capsule, cast,
cervical, cream, and cystectomy. Eponyms (8) further complicate descriptions. These examples illustrate the problem of translating medical phrases into concise computerstorable language. In addition to problems inherent to the
understanding of natural language, additional difculties
pertaining to medical terminology can be indicated, as follows.
American Versus British Spelling. Two standard differences are evident, namely the use of the digraph in British
spelling (e.g., anaemia versus anemia) and preference for
using c (e.g., in leucocyte) rather than k (as in the American word leukocyte). Interestingly, the British equivalent
of the American spelling of the word leukemia is spelled as
leukaemia.
Preferred Terminology. In radiology air means gas
within the body, regardless of its composition or site, but
the term should be reserved for inspired atmospheric gas.
Otherwise, the preferred term is gas. Sometimes the
preferred terminology refers to simplicity; the expression
lower extremity must be replaced by leg, for example. On other occasions the preferred terminology pertains to technical vocabulary that permits high precision
if the available information is exact; the word clumsiness
describes defective coordination of movement in general,
whereas dysdiadokokinesis refers to a defect in the ability to perform rapid movements of both hands in unison
(9).
Meaning Within a Certain Context. The quality blue primarily refers to a particular color. The actual meaning in
medical language may, however, relate to a specic noun,
for example, blue asphyxia, blue baby, blue bloater, blue diaper syndrome, blue dome, blue line, blue nevus, blue pus,
blue sclera, blue stone, and blue toe syndrome (7).
Implicit Information. A particular statement may imply
many relevant components, for example, if urinalysis is
normal, then this result implies the absence of proteinuria,
hematuria, glucosuria, and casts. Also antonyms may apply: leukopenia in particular implies no leukocytosis. This
mutual exclusion principle applies to all terms beginning
with hypo- or hyper-.

Imprecise Terminology. (10) Some terms may carry a
vague meaning, for example, tumor, swelling, mass, and
lump. To a large extent, however, the use of such terms
reects the uncertainty related to an observation. In that
respect it is justiable: indeed it would be incorrect to specify an observation in greater detail than the facts permit.
This notion has consequences for the selection of equivalent expressions.
Certainty Versus Uncertainty. Decision analysis itself
does not reduce our uncertainty about the true state of
nature, but as long as we must make some choice it does
enable us to make rational decisions in the light of our
uncertainty (11). Another aspect concerns subjective interpretation of percentage gures about prognosis (12). Outcomes perceived with certainty are overweighted relative
to uncertain outcomes. Thus, the formulation of information affects its interpretation by humans.
Limited Scope of a Thesaurus. Thus far, no agreement exists regarding directives for coding diseases. Major sources
are organized in different ways, for example, in the International Classication of Diseases1 (ICD) one nds
Bladder, see condition (e.g., Leukoplakia), whereas the
book Current Medical Information and Technology (CMIT)
(13) reads leukoplakia (of bladder), see bladder. Notably,
leukoplakia of the bladder as such is not listed in the book
Medical Subject Headings (MeSH) (14).
Knowledge Engineering. This type of engineering implies
various levels of translation. Thoughts by the human expert are formulated as precisely as possible, the engineer
provides feedback using his or her own phrases to ensure
an exact match between both minds, and subsequently the
resulting expression is translated into a format usable for
the computer program. These steps involve transformations of language while yet assuming that the ultimate user
of the program fully appreciates the scope of the original
thoughts of the expert.
Information Source Versus Actual Patient. (15) Current
medical information sources tend to adhere to preference
terminology to promote the use of uniform medical language. However, such standard vocabulary is not used by
the average patient to describe individual health problems
(16). Then it is left to the clinician to transpose, for example,
puffy face and moon face if appropriate. Indeed, better
health care can be realized by educating the patient about
the value of structured communication with the physician
(17).
Synonyms. For example, icterus is identical to jaundice. Thrombocytosis and thrombocythemia are two words
to indicate that the number of platelets in the peripheral
circulation is in excess of 350,000 per microliter.
Subspecialty Interpretation. When naming a hollow
space you may choose anything out of the following set:
cavity, crypt, pouch, gap, indentation, dell, burrow, crater,
concavity, excavation, gorge, pocket, cave, cavern, cistern,

or lacuna. However, every expression may exhibit a nuance
within a certain context. Then there is jargon: the terms
show, engagement, and station, for example, have a
particular meaning within the eld of obstetrics (7). The
term streaking has a different meaning for the microbiologist and the radiologist.
Eponyms. Many disease names refer to the rst author
(e.g., Boecks disease for sarcoidosis) who described that
particular disorder, to the rst patient analyzed in detail (e.g., Mortimers disease, again for sarcoidosis), or to
the geographical area (e.g., Lyme disease) where the illness was rst detected. But variations may occur: The
PlummerVinson syndrome (sideropenic dysphagia), as it
is known in the United States and Australia, is termed
PatersonKelly syndrome in the United Kingdom, but
WaldenstromKjellberg syndrome in Scandinavia (8).
Multilingual Approaches. The relation between a concept and the various corresponding terms in different languages is in general not unique. This implies that a multitude of different words from different syntactical categories may represent a single concept. Particularly the European countries are confronted with additional natural
language problems. The Commission of the European Communities supports research activities in this area through
the Advanced Informatics in Medicine (AIM) project, such
as EPILEX (a multilingual lexicon of epidemiological terms
in Catalan, Dutch, English, etc.) (18), and the development
of a multilingual natural language system (19).
Frequency of Occurrence. The meaning of semiquantitative indicators such as always and often is not transparent when screening a medical text. The intuitive interpretation of some quasinumerical determinants is summarized elsewhere (20).
Noise Terms in Patient Description. When analyzing 104
patient cases, we found (21) that the input consisted on
average of 75 terms; the required number of terms for establishing the primary diagnosis was only 15. This implies
that 80% of the input data consisted of noise terms that
may blur the process of hypothesis formation for humans
(22).
Illogical Terminology. Certain terms contain paradoxical details, for example, hayfever is usually not accompanied by fever, while acute rheumatic fever typically has a
chronic course.
CLASSIFICATION AND CODING SYSTEMS
With the exception of one British project, all classication or coding systems have been developed in the United
States. The following survey lists all projects along with
some of their characteristics.
The ICD system just entered its tenth version, although the ninth edition is still used. It is applied

worldwide for classifying diagnoses and also permits

diagnosis-related group (DRG) assignment employed
for billing and reimbursement purposes.
Another widely accepted system is called Systematized Nomenclature of Medicine (SNOMED) (23),
which offers a structured nomenclature and classication for use in human as well as in veterinary
medicine. It covers about 132,600 records, with a
printed and a CD-ROM version.
The Current Procedural Terminology volume (CPT)
(24) provides a uniform language for diagnostic as
well as surgical and other interventional services. The
system is distributed by the American Medical Association (AMA) and has been incorporated in the Medicare program.
MeSH (14) is a systematic terminology hierarchy that
is used to index the MEDLINE medical publications
system, with annual updates.
The National Library of Medicine2 (NLM) in 1986
started a project called the Unied Medical Language
System3 (UMLS) (25), aiming to address the fundamental information access problem caused by the variety of independently constructed vocabularies and
classications used in different sources of machinereadable biomedical information.
Gabrieli (26) constructed a computer-oriented medical nomenclature based on taxonomic principles. His
system covers 150,000 preferred terms and a similar
number of synonyms. The partitioning method employed for medical classication readily permits replacement of English names with terms of any other
language, thus creating the perspective of a worldwide standard.
Read from the United Kingdom designed a classication for various computer applications (27). Its design adheres to the following criteria: comprehensive,
hierarchical, coded, computerized, cross-referenced,
and dynamic. The system is closely connected to the
British National Health Service (NHS). Version 2 includes 100,000 preferred terms, 250,000 codes, and
150,000 synonyms.
MEDICAL KNOWLEDGE BASES

Ideally a medical knowledge base (KB) integrates text,
graphics, video, and sound. Furthermore, it should be accurate, veriable, and easily accessible where doctors see
patients, and the system should be adaptable to doctors
own preferred terms or abbreviations (28). Future developments will certainly include the use of ANNs (29), and
examples realized thus far include myocardial infarction,
diabetes mellitus, epilepsy, bone fracture healing, appendicitis, dermatology diagnosis, and electroencephalogram
(EEG) topography recognition. An overview will be given
of current KB systems. With the exception of the Oxford
System of Medicine (OSM) and Medwise, all projects originate in the United States. One approach (CONSULTANT)
addresses the eld of veterinary medicine. A survey referring to the year 1987 has been published before (15).
Obviously, the Internet, a rapidly expanding network for

computer-to-computer communication, nowadays offers a
convenient window to medical resources. A useful guide,
called Medical Matrix is the result of a project devoted to
posting, annotating, and continuously updating full content and unrestricted access to this medium. The system
can be reached at www.medmatrix.org and features a ranking system based on the utility for point-of-care clinical
application.
factors for disease prole matching.
The Framemed system (37) divides medical informa-

CMIT developed by the AMA (13) forms a reference
for the selection of preferred medical terms including

certain synonyms and generic terms with builtin arrangements to provide maximum convenience in usage, currency, and timely publication.
Blois was the rst to apply CMIT as a diagnostic tool
in his RECONSIDER project (30). The application was
released in 1981 and covered 3,262 disease entities,
while 21,415 search terms were listed in a directory
along with their frequency of occurrence. The program
is extensively described in his book (30).
DXplain (31) is also based on CMIT (13). The project
had close connections with the AMA, and information
is distributed using the World Wide Web. The KB contains information on 2,000 diseases and understands
over 4,700 terms, with 65,000 disease-term relationships.
QMR patient diagnostic software (32) covers some 600
disease proles, and is the personal computer version
of the INTERNIST-I prototype. Unfortunately the size
of its disease KB remained remarkably constant over
the last few years.
MEDITEL (33) addresses the issue of diagnosis in
adults. Over the last few years not much news was
reported in the literature, apart from a comparative
study (34).
ILIAD (35) is a software package designed to aid students and residents in their clinical decision logic. The
project stems from the health evaluation through logical processes (HELP) system developed at a major
hospital in Salt Lake City. Its KB covers 1,300 diseases and 5,600 manifestations, mainly subspecialties
of internal medicine.
The Oxford System of Medicine (OSM) project, initiated by the Imperial Cancer Research Fund for use in
primary care, helps general practitioners during routine work to support decision-making tasks such as diagnosing, planning investigations and patient treatment schedules, prescribing drugs, screening for disease, assessing the risk of a particular disease, and
determining referral to a specialist (36).
Medwise was founded in 1983 and now covers some
3,900 disease entities, with 29,000 different keywords
(21). It includes a separate KB with almost 500 equivalent terms that each refer, on average, to three related terms. Equivalent terms are automatically generated to assist the user during the process of data entry. The matrix structure of the Medwise KB permits
semantic differentiation, with corresponding weight
tion into 26 domains and arranges the items in a

hierarchical sequence, thus yielding a logical framework for a standardized terminology. The objective is
to achieve a standard coded terminology (including
synonyms) to which all existing systems can relate,
with obvious use as an electronic encyclopedia and
for differential diagnosis.
STAT!-Ref (38) offers the contents of a rst-choice
medical library (including several standard textbooks,
e.g., on primary care) as well as Medline on CD-ROM.
MD-Challenger (39) offers a clinical reference and
educational software for acute care and emergency
medicine (everything from abdominal pain to zygoapophyseal joint arthritis), with nearly 4,000 annotated questions and literature references. MDChallenger also includes continuing medical education (CME) credits.
Labsearch/286 is a differential diagnosis program allowing input of up to two abnormal laboratory ndings plus information on symptoms and signs. Laboratory data concentrate on body uids (blood, urine,
cerebrospinal, ascitic, synovial, and pleural uid) entered as high or low (40). The system includes 6,500
diseases and 9,800 different ndings.
CONSULTANT (41), a KB for veterinary medicine,
was developed at Cornell University. This database
for computer-assisted diagnosis and information
management is available on a fee-for-service basis in
North America.
Grifth and Dambro (42) compiled an annually updated book. The information is compiled by a group of
contributing authors whose names are listed in conjunction with each disease prole. The rst edition
appeared in 1993 and contains chartlike presented information on 1,000 topics, along with their ICD code.
The printed version shows similarity with CMIT [13].
Publication using electronic media recently became
available.
The Birth Defects Encyclopedia (43) is a comprehensive, systematic, illustrative reference source for the
diagnosis, delineation, etiology, biodynamics, occurrence, prevention, and treatment of human anomalies
of clinical relevance. A unique feature of the printed
edition is the Fax service for requesting a current,
daily updated version of any article in the KB. The related birth defects information system (BDIS) is a sophisticated computer-based prole-matching system
that helps research and diagnostic tasks associated
with complex syndromes.
DiagnosisPro by MedTech, U.S. (44), is a differential
diagnosis system including 8,500 diseases, designed
by the internist C. Meader and the clinical pathologist H. C. Pribor. It covers all major specialties and is
considered a useful tool for primary care professionals.
EXPERT SYSTEMS AND COMPUTER ASSISTED

DIAGNOSIS
The term expert system or knowledge-based system (KBS)
refers to a computer program that simulates the professional capabilities of a human expert (4549). In the eld
of medicine, the expression computer-assisted decision system is also used for an expert system. KB systems are used
for interpretation of actual data about a specic problem,
considering the knowledge represented in the domain of
the KB, to develop a problem-specic model and then to
construct plans for problem solution. KBs usually include
facts about the problem domain, and procedural knowledge
to manipulate facts. In production systems this procedural
knowledge adopts the form of IF-THEN or IF-THEN-ELSE
rules, where the IF part is the antecedent, the THEN part is
the conclusion, and the ELSE part, if exists, is the alternative conclusion. Candidate hypotheses are derived through
some pattern-matching system. A reasoning engine (also
termed inference machine) carries out the manipulation
specied to obtain an answer. An inference engine is no
more than a program, the function of which is to decide
what to do at any given moment, that is, it recognizes and
activates the appropriate rules. Generally, an inference engine should include an interpreter, which activates the relevant rules at any given moment, taking into account the
current state of the active memory; a search strategy, which
includes exploration heuristics; a self-knowledge mechanism, which permits the identication of the structures
being utilized, the state of the problem, and changes in
the active memory; and a termination mechanism for the
inferential processes.
The overall functioning of the inference engine occurs in
cycles called basic production system cycles. The nature of
these basic cycles is very different depending on whether
the search process is directed by the data or by the objectives. Given that the production systems are essentially
based on rules, it will be necessary to dene how the propagation of knowledge within the system can be affected.
Let us look at two basic propagation methods.
1. Forward chaining of rules, which is implemented as
a search process directed by the data, and where the
search initiates with the antecedents and leads to the
conclusions of the rule
2. Backward chaining, which is implemented as a search
process directed by the objectives, and where the search,
by means of an evocative process, initiates with the conclusions of the rules, established as hypotheses, and extends to the antecedents
Verication and validation are the two most important
stages in the evaluation of an expert systems behavior and
functioning (4853). Verication endeavors to ensure that
the system has been constructed correctly. This means that
it ensures that the software contains no errors and that the
nal product satises the initial design specications and
requirements. Validation, on the other hand, refers more
precisely to a detailed analysis of the quality of the expert system in the context of its work environment, which
permits determining whether or not the developed product

adequately meets the expectations deposited therein. Verication of an expert system necessarily involves a detailed
analysis of the knowledge contained within the system.
Particularly if we refer to production systems, the rules
may be the origin of many errors, among which the following can be identied:
Conictive knowledge
Circular knowledge
Redundant knowledge
Unnecessary knowledge
Rules included in or contained within others
Rules never executed
The preceding erroneous situations must be detected

and resolved, and in order to do this, the verication of
the expert system may be approached from two different
perspectives: rst, verication dependent on the domain
and second, verication independent of the domain. In the
preceding section we have described several systems with
KBs that are used for establishing a differential diagnosis. Recently, the diagnostic performance of four such commercially available programs (Dxplain, Iliad, Meditel, and
QMR) was evaluated (34). The fraction of correct diagnoses
by the computer ranged from 0.52 to 0.72, while half of
the candidate diagnoses proposed by the human experts
were not generated by the computer. However, on average
each program suggested two additional diagnoses per case
that the human expert did nd relevant but which the researchers failed to include in their original differential diagnosis list. An obscure limitation of the study design is
that the researchers themselves created a set of cases to be
analyzed by the computer programs. Also, they employed
the vocabulary provided by the programs developer. Our
written requests to receive a copy of the patient cases used
in this study for further testing of other programs were
not honored thus far. In his editorial, Kassirer (45) wrote
that the results of the study indicate substantial progress,
but he found them disappointing from a physician-skeptic
point of view. He concluded that the structure of the KBs,
the computational inference engine that integrates clinical data into a diagnosis, the methods of capturing clinical data from patients records, and the human-computer
interface are still in their infancy. However, several eld

prototype expert systems have been successfully validated
(5559), and some methodologies for expert systems validation have been proposed (6062).
ARTIFICIAL NEURAL NETWORKS (ANN)

In 1956, a group of researchers met in Dartmouth college in
order to discuss the possibility of constructing genuinely intelligent, technologically advanced machines. This meeting
laid the foundations for the science of AI. Principal among
the participating researchers of note were Samuel, McCarthy, Minsky, Newell, Shaw, and Simon. Following this
meeting two major breakaway groups were formed, both
of which continued working more or less independently of
each other. Thus, Newell and Simon formed a team with
the idea of developing human behavior models, whereas
McCarthy and Minsky formed another team dedicated to
the development of intelligent machines, not being particularly concerned with human behavior as such. The rst
approach entailed an emulation of cerebral activity and,
wherever possible, the copying of its structure. The second approach implied the construction of systems in which
the problem-solving procedures applied are such that, were
human beings to apply them, they would be considered intelligent.
However, in practice a combination of both approaches is
necessary in order to obtain results that may be considered
useful. Both approaches comply with the fundamental objectives of modern AI, namely, the understanding of human
intelligence and the use of intelligent machines to acquire
knowledge and to resolve complicated problems satisfactorily. Both approaches lead to AI programs, KB systems,
expert systems and, nally, to ANNs.
AI programs can be said to exhibit a certain intelligence
as a result of the skillful application or use of heuristics
in the broadest sense. Heuristic knowledge is considered
the fruit of experience, which is difcult to formulate, and
which is established implicitly in order to nd answers to a
specic problem, answers that may be more or less accurate
but that are nevertheless always valid. Although the use
of heuristic knowledge does not guarantee the nding of
optimal solutions, it does offer acceptable solutions, if they
exist, by means of so-called inferential processes.
The next epistemological level is that of knowledgebased and expert systems, for which knowledge of the
specic domain and knowledge of the control structures
used to manipulate said knowledge are physically separate. Therefore, it is necessary to dene and implement
architectures different from the ones we are accustomed
to, in which knowledge and control structures can be developed independently of one another in such a way that
one specic control structure can be applied to knowledge
from different domains. Expert systems can be considered
as specialized knowledge-based systems, in that they resolve real-life problems, which, although limited in terms
of size, are complex. The construction of an expert system
requires the employment of techniques developed to construct AI programs, in addition to architectures dened for
the development of knowledge-based systems. However, it
is absolutely essential to place more emphasis on differential aspects such as the acquisition of knowledge and
learning.
ANNs can be dened as massively parallel distributed
processors with a natural capacity not only for storing
experience-based, that is, heuristic, knowledge, but also
as a facility for making such knowledge available for use.
ANNs allow limitation of the brain in two ways. First,
knowledge is acquired by means of a learning process, and
second, the synaptic weights are used for storing the knowledge.
It is obvious that, in order to obtain acceptable results
at any of the levels of AI described above, we need to draw
from other elds such as mathematics (its language and
procedures), medicine (especially the neurophysiological
models), computer science (particularly software engineering and systems architecture), linguistics (especially syntax and semantics), psychology (which allows us to analyze
intelligent behavior models), and nally, even philosophy.
Advanced Aspects of AI
Dealing with Uncertainty. AI is not only concerned with
general mechanisms related to the search for solutions
within a given space or with how to represent and utilize the knowledge of a specic discourse domain. Another
aspect, up to now just mentioned in passing, is that concerning inferential mechanisms and/or processes, which
are considered as the starting point for the so-called reasoning models.
In any domain, the propagation of knowledge by means
of AI programs is always carried out by following a welldened reasoning model. These reasoning models contribute in a decisive way to the correct organization of the
search for solutions. Normally, the domain characteristics
and the characteristics of the problems to be solved determine the type of reasoning model to be employed. Thus,
there are domains of a markedly symbolic nature, in which
solutions can be established with absolute condence. In
these cases the use of categorical reasoning models is indicated (63). There are, on the other hand, domains that are
of a statistical nature, where unique solutions cannot be
obtained and where in addition, a decision must be made
as to which of the possible solutions arrived at is the most
probable. In these cases it is preferable to reason with statistical models of which, given the peculiarities of the inferential processes that AI deals with, the Bayesian scheme is
the most widely used (64, 65).
There are other domains in which the concept of uncertainty appears and which may be inherent to the data of the
problem and the facts of the domain, or to the inferential
mechanisms themselves. In such cases reasoning models
are chosen that are capable of correctly manipulating such
uncertainty (6668).
Finally, there are domains in which the inferential elements include nuances of a linguistic nature where hierarchies and classications can be established. Indicated in
these cases are reasoning models based on fuzzy sets (69,
70).
Obviously there are domains that manifest more than
one of the characteristics just mentioned, in which case the
reasoning model most appropriate to the characteristics of

the domain or a combination of different models can be
used.
basic probability theory, Dempster and Shafer dene the

following conditions for m:
The Dempster-Shafer Theory of Evidence

This reasoning scheme has a solid foundation in theory to
the extent, in fact, that the original reasoning model proposed by Dempster was subsequently formalized and converted into a genuine theory by Shafer (71). This scheme is
attractive, principally for the following reasons: (1) it permits the modeling of uncertainty associated with pieces of
evidence and hypotheses in a simplistic manner; (2) it permits the consideration of sets of hypotheses without the
condence in each set having to be distributed in any way
between each of the individual hypotheses of the set; (3) it
permits an elegant but precise representation of the lack
of knowledge so frequently associated with reasoning processes; (4) it deals with the probability theory as a special
case; (5) it contains some of the combinatory functions of
the certainty factors model.
But how is it possible to deal with both the inexact
knowledge and the lack of knowledge in the DempsterShafer model? In the rst place, and given any discourse
universe whatsoever, Dempster and Shafer introduce the
concept of a discernment frame that can be dened as the
nite set of all hypotheses that can be established in the
problem domain. The discernment frame should form a
complete and thus exhaustive set of hypotheses that are
mutually exclusive. On the other hand, the effect of a specic piece of evidence on the overall set of hypotheses is
not determined by the contribution of the condence deposited in the individual hypotheses. On the contrary, each
piece of evidence affects, generally speaking, a subset of hypotheses within the discernment frame. This approach is
consistent with the reality of almost all real-life routine
problems. In real-life problems, the reality is that the evidence e permits discrimination between groups or sets of
alternative hypotheses. However, at the same time, within
a set, uncertainty with respect to the alternative hypotheses is maintained. According to this argument:
Z is the discernment frame
A is any subset of the frame
h1 , . . . , hn are the hypotheses of the discernment frame
In this context, the appearance of specic evidence e will
favor a determined subset A within Z, in such a way that
the degree to which A is favored by e is represented by m(A),
where m is indicative of the condence that the evidence e
permits in A. The values of m are represented by the closed
interval [0, 1]. We will use the following notation:
The fact that the evidence e supports the subset A does

not imply, as already pointed out, that the individual hypotheses divide, explicitly, the condence deposited in A
itself. This fact diverges considerably from classical theories of probability. Each subset of the discernment frame,
for which given evidence e it is established or veried that
m(A) = 0, is called a focal element. Returning briey to the
Both conditions are a direct consequence of the restrictions

imposed on the discernment frame. Evidence theory provides us with a neat way to deal with the lack of knowledge
associated with reasoning processes. Let us take a discernment frame Z and evidence such that
The rst condition required for m establishes that A Z

m(A) = 1.
What happens to the rest of the condence that has not
been assigned to the focal element A? To answer to this
question, Dempster and Shafer postulate that if
then
This formula should be read as given that the evidence e

supposes the assignation of a given condence to a specic
focal element A within the discernment frame, then the rest
of the unassigned condence represents a lack of knowledge and therefore should be assigned to the discernment
frame itself. This situation leads us to reect as follows:
Unassigned condence is ignorance or lack of knowledge
with respect to the importance of the evidence in relation
to the focal element under consideration. In other words,
it is known that the evidence supports the focal element to
the extent of s. However, referring to the unassigned condence (1 s), we do not know if it contributes or not to A
(or to any other subset within the frame). The unassigned
condence (1 s) should be assigned to the frame since we
constructed the frame; what we do know as a fact is that
the solution is within the frame. The complete formulation
for the approach is as follows:
The discernment frame Z = [h1 , . . . , hn ]
The focal element A Z
Evidence e referring to A
A measure of the basic probability of A given e: m(A)
e: A m(A) = s
m(Z) = 1 s
m(B) = 0 B Z, B = 0, B = A
If the approach were probabilistic, the same evidence
would support the focal element A as well as the complement of the focal element. Thus,
This was precisely one of the major drawbacks of the probabilistic models. With this new theory, if
and
with
then
Generalizing broadly, it could be said that the way in which

the lack of knowledge in the evidential theory is managed
more than compensates for the defects in the probabilistic
models.
Fuzzy Systems. Uncertainty does not only occur as a consequence of an absence of information or of other circumstances that may be, to a greater or lesser extent, formalized. On the contrary, uncertainty could well be associated
with the very way in which humans express themselves.
In practice, most human statements are ambiguous and
this ambiguity is an essential characteristic not only of
language but also of the processes of classication, of the
establishment of taxonomies and hierarchies, and of the
reasoning process in itself. Hence, if we dene living things
as organized molecular structures that are born, grow, reproduce, and die, it is clear that even the humble beetroot
does precisely the same and therefore should also be considered a living thing. On the other hand, a sliver of a stone
does not behave in the same way, and thus should not be
considered a living thing. But what about a virus?
The difculty in dening a virus so as to include it in the
set of living things lies with the selfsame denition of the
concept living things. In other cases, however, the difculty
arises due to questions of a subjective nature. For example, characterizing the set of beautiful people is not easy,
since each person has a very personal idea of exactly what
attributes an ideal should have in order to be beautiful.
But it is particularly difcult to say if one specic object
is beautiful or not. In this particular situation, subjective
nuances appear that render impossible the very idea of
classication.
Furthermore, it is not only problems of denition or of
subjective nuances that may complicate categorical classication. In other cases, the context may modify the criteria. For example, the concept of a tall manwhich in itself
is intrinsically ambiguousdiffers notably depending on
whether one refers to a Scandinavian or to a Pygmy, and
to make matters worse, probably both are right!
So it is possible to conclude rapidly that ordinary sets,
in which an element of a determinate universe may or may
not belong, are not sufciently complete so as to represent
the knowledge normally utilized within the command of a
human being, not to speak of reasoning with that knowledge. The elds of mathematics and AI, at their outset
concerned with interesting problems of a cognitive nature,
were from an early stage intrigued by this concept. Finally,
in 1965 Lofti Zadeh published in his famous article Fuzzy
Sets the results of his investigations in this area (72).
Let us take any universe whatsoever, for example, the
universe formed by the set N of natural numbers. Let us
dene a subset of N called A, characterized by the following

description: A is the subset formed by natural even numbers of a value less than 10. Thus A, a subset of N, is perfectly dened as follows: A = [2, 4, 6, 8], and obviously 2
belongs to A, but 3 does not belong to A, and 10 does not
either. In this particular example, we have no difculty in
establishing the degree of belonging of an element of the
discourse universe to a particular subset.
Now let us look at the universe C characterized by the
following description: C is the set formed by all human beings; let B be a subset of C characterized by the description
B is a subset of C that includes tall, dark men. In this
situation it is difcult to establish the degree of belonging
of an element of the universe C to the subset B. Clearly,
a common set may be dened as a collection of elements
and if an element of the universe is represented in the collection, the element in question belongs to the said set. In
these cases, it can be said that the degree of belonging of
any particular element of the referential universe has a
Boolean value. Thus if the element belongs to the set, the
Boolean value is 1. If the element does not belong to the
set, the Boolean value is 0. In this way we can construct
a function f (which for common sets is a Boolean function)
such that, given an element x from the universe U, and
given (also) a subset of U, A, then
We will now extend the discussion to those special kinds of

sets that we have called fuzzy sets. In reference to these, we
said that linguistic, subjective, and other nuances impeded
a precise establishment of the degree of belonging to the
fuzzy set in question, of particular elements of the universe.
Thus there will be elements from the universe that clearly
belong to the set, others that clearly do not belong, and
yet others that belong to a certain extent, but not totally.
Following the approach previously described, the problem
is easy to resolve if we consider the function f to have the
following values, given an element x from the universe U,
and (from U) a fuzzy subset A (from U):
The function f quanties the degree of belonging of an element from the universe to the fuzzy set in question. Thus,
a fuzzy set is one for which there does not exist a clear dividing line between belonging and not belonging of determinate elements from the universe. In order to establish
the fuzzy limits of the corresponding set, we shall require
a criterion, which naturally will be arbitrary. Let us examine the universe of living persons along with the fuzzy
set A of U answering the description A is the set of young
living persons. a property we consider appropriate for the
characterization of the fuzzy subset A is the age of the
universe elements, but how should we dene age? We are
faced with the not insignicant problem of the denition
of criteria for the fuzzication of sets. In our example,
we will consider as young all those elements from the
universe whose age permits them to legally obtain Youth
Travel Cards, Inter-Rail tickets, etc. (i.e., elements of age

25 years old), and not young all those elements that can
legally obtain Pensioner Travel Cards (i.e., elements of age
>65 years old). Thus,
But what happens to all those elements from the universe

aged 26 to 64 years old? What exactly is their degree of
youth with respect to the criterion of age? We are faced with
yet another problem, which is the characterization of the
diffuse area or zone. In order to get out of the conundrum,
we will construct a linear function as follows:
In this way we can segment our numeric space [0 1] in

three zones, two of which are not fuzzy and which refer to
those elements of the universe that clearly belong or that
clearly do not belong, to the fuzzy subset in question, and
a third zone that is fuzzy and that corresponds to those
elements of the universe that belong, to a certain extent,
to the fuzzy subset in question.
We will now examine a situation in which Tom is 17
years old, Dick 31, and Harry 73. Conscious of the fact that
each one is a living person, what can we say about their
youth? In accordance with the established criteria, and
carrying out the appropriate substitutions, we can establish the values of their respective functions of belonging to
the fuzzy subset of young living persons as follows:
It is obvious that as the ages of our friends increase, their

degree of belonging to the fuzzy subset decreases. The approach is coherent but is not very natural in linguistic
terms. Just to illustrate our meaning, have a look at the
following dialogue: By the way, Sally, how old is Dick? I
think he is 31. Oh, so he is 0.85 young! Absurd! Nobody
talks in this way! We are faced again with a new problem, that of linguistic classication of fuzzy sets. The basic
idea is that once we have managed to segment the numeric
space (indicative of the degree of belonging of each element
to the fuzzy subset in question), we need to segment the linguistic space by means of labels containing information of
a semantic nature and then to match each linguistic label
to a specic numeric interval, on the basis of a (minimally)
reasonable criterion. Returning to our example, we dene
a linguistic scale to which we assign concrete values from
our function degree of belonging to the fuzzy subset in question. Thus
According to this scale, and using the facts from the example, we may now say that Tom is totally young (or simply
young), Dick is fairly young, and Harry is not at all
young (or simply not young). These expressions represent a natural, human way of expressing judgments with
respect to the ages of our friends.
Although an in-depth discussion of the problems deriving from knowledge representation and from fuzzy reasoning is way beyond the scope of this text, it is, however,
necessary to include a reference to both. Remember that,
from the perspective of AI, the fuzzy model permits us to
represent and manipulate expressions appropriate to the
language of human beings. In such expressions we come
across fuzzy predicates, fuzzy quantiers, and fuzzy probabilities. Other, more conventional approaches to the representation of knowledge lack the means for efciently representing the meaning of fuzzy concepts. Models based on
rst-order logic, or those based on classical probability theories, do not allow us to manipulate the inappropriately
named common-sense knowledge. The reasons for this are
as follows.
Knowledge derived from common sense is lexically imprecise.
Knowledge derived from common sense is of a noncategorical nature.
The characteristics of the fuzzy sets examined in the previous paragraphs give us clues as to the procedure to follow if what we require is the application of knowledge representation models and reasoning models, based on fuzzy
logic(s). Thus (73)
1. In fuzzy logic, categorical reasoning is a special case of
approximate reasoning.
2. In fuzzy logic, it is all a question of degree.
3. Any fuzzy system can be fuzzied.
4. In fuzzy logic, knowledge should be interpreted as a collection of fuzzy restrictions placed on a collection of
variables.
5. In fuzzy logic, reasoning problems and therefore inferential processes should be interpreted as propagations
of the fuzzy restrictions mentioned previously.
Although the theoretical bases for fuzzy logic are quite
clear, applications of the latter to systems of an inferential
nature is problematic. Even at the time of writing, these
difculties have not been entirely overcome. Nevertheless,
it appears that fuzzy-system theories applied to control
problems, in place of more conventional approaches, are
coming up with solutions that are both brilliant and elegant.
FURTHER READING
The elds of medical KBs and terminology are rapidly developing. There is no single comprehensive source of information available, but the professional reader is advised
to scrutinize the following journals and organizations for
10
updated information.
M.D. Computing, published by Springer Verlag (New
York and Berlin), reports on research in the eld of
medical informatics. Of special interest to the clinician is also the journal Experts Systems with Applications, published by Pergamon Press (New York).
IEEE Expert appears four times a year and presents
the latest on AI and expert systems. Contact P.O.
Box 3014, Los Alamitos, CA 90720. This journal is
for those interested in technical details on intelligent
systems and their applications. IEEE (P.O. Box 1331,
Piscataway NJ 08855-1331) also publishes a number
of related journals, for examples, on knowledge engineering, fuzzy logic, ANNs, and multimedia.
The NLM releases news bulletins and provides information on UMLS and contracts for cooperation.
Obviously, annual meetings form the forum for presentation of the latest developments:
IMIA conferences are well known, besides the world
congress organized every four years. IMIA publishes
a Yearbook of Medical Informatics. It offers the pearls
of medical informatics since it covers inuential papers from 100 journals in the eld. Contact Schattauer Publishers, P.O. Box 104545, 70040 Stuttgart,
Germany.
USABILITY OF EXPERT SYSTEMS
In the development of medical software systems in general, and in medical KBSs in particular, there is currently a
gradual shift in philosophy, from a system-directed one, in
which internal architecture and functions set the pace for
development, towards a user-centred philosophy (referred
to as user-centered design, or UCD), in which the user is
implicated in design aspects. However, one of the main obstacles to this new approach to design is the lack of suitable
tools. Consequently, greater effort is required from the software engineering community in the eld of manmachine
interaction or humancomputer Interaction (HCI). A wide
range of techniques are currently available for analyzing
the usability of computerised intelligent systems. (74) The
fact that so many techniques have been developed is due to
the fact that, to date, no single method will ensure that a
system is usable. In fact, the use of several approaches and
an overall analysis of results are generally recommended.
In an attempt to organize and facilitate the learning of
usability analysis techniques, a number of authors have
classied these in terms of hierarchical models.
Of particular interest are the classications drawn up
by Ivory and Hearst (75), Adelman and Riedel (76), and
Preece (77). The simplest of these is the Adelman and
Riedel classication, consisting of three main categories,
namely, heuristic methods, subjective methods, and empirical methods. These are described in turn as follows:
1. Heuristic methods are based on the opinions of usability
experts. These experts analyze the different system in-
terfaces and determine strengths and weaknesses from

an end-user perspective. Heuristic analysis techniques
can be classied in one of two main groups:
Analytical techniques, in which evaluators collect information on the usability of a system by constructing formal models that represent the system being
assessed for usability. One of the most relevant analytical techniques in our context is the GOMS analysis
(78).
Heuristic inspections, in which evaluators apply criteria of a heuristic nature to the identication of possible usability problems. Specically this consists of
a systematic analysis of the systems user interface
by usability experts. The ultimate aim of the inspection is to identify usability problems in relation to design that can be resolved in the iterative development
phase (79).
2. Subjective methods are based on the opinions of the system users. They analyze operational prototypes of the
product in the development phase and voice their opinions on the usability of these prototypes. These methods
can be classied in one of two main groups: survey methods and observational methods.
The aim of the survey methods is to compile data on
the opinions of users after these have used the system, by using interviews, questionnaires and/or user
feedback.
Observational methods, on the other hand, are based
on obtaining user opinions as these use the system
rather than after they use the system. The techniques
used in this case are direct observation, indirect observation, verbal protocols, post-event protocols, and
constructive interaction. Of the techniques described
above, the only approach that can be completely automated - in view of the current state of technology - is
the development of closed questionnaires. A particularly popular approach to evaluating closed questionnaires is MAUT (multi-attribute utility Theory) analysis (80). An alternative technique is AHP (Analytic
Hierarchy Process) (81).
3. Empirical methods are based on the actions of the system users. The approach is based on obtaining objective
data on practical hands-on use of the system. Empirical methods are based on an analysis of the actions of
users of a system. The process generally consists of logging a users interaction with the system while he/she is
performing a series of tasks. This log is then analyzed to
produce a series of measurements (such as number of errors, time required to perform a task, etc.). Conclusions
on system usability can then be drawn by performing a
detailed analysis of these measurements.
KBS HYBRID ARCHITECTURES
During the last years, there has been an increasing tendency in developing expert systems that are progressively
more complex. In order to be able to cope with the increasing complexity, expert systems have incorporated several
intelligent techniques that try to take advantage of the
complimentary characteristics of symbolic systems, neural
networks, fuzzy systems, or genetic algorithms, among others. These systems are called hybrid intelligent systems,
and combine intelligent techniques as well as conventional
computing techniques to achieve a higher level of machine
intelligence. Hybrid systems help to:
1. Improve the available techniques, integrating several
of them so as to conceal the problems that each of
them present. For example, neural networks are good
at learning, but can not do high level reasoning. On the
other hand, symbolic expert systems are good at high
level reasoning, but more limited in learning.
2. Find solutions for complex tasks. Most application domains present several subtasks with different characteristics. For example, the logic and static components
can be adequately managed by expert symbolic systems, while other components that are dynamic, fuzzy
or poorly understood could be managed, for example, by
neural networks.
3. Implementing multifunctional systems. In this case, the
goal is to create a system that can exhibit multiple capacities for information processing in a unique architecture. That is, there is only one system, but it tries to emulate the behaviour of different processing techniques.
One example of this is the use of neural networks for
symbolic processing.
Depending on factors such as functionality, processing
architecture and communication requirements, three basic types of architectures for hybrid systems can be distinguished (82):
1. Expert systems with function replacement, in which a
principal function of a given technique is replaced by
another intelligent processing technique. The aim for
replacement is either increasing the execution speed or
enhancing reliability. An example of this type of hybrid
system could be the replacement of the backpropagation
weight changing mechanism of a neural network with
genetic algorithm operators.
2. Intercommunicative hybrids, which are independent,
self-contained intelligent processing modules that exchange information and perform separate functions to
generate solutions. It is used when a problem can be
divided in sub-problems, each of which can be resolved
using a different technique, such as neural networks,
symbolic systems, etc. An example could be a diagnosis
system in which an expert system realizes inferences,
and calls neural networks when needed to analyze data
top obtain patterns (83).
3. Polymorphic hybrids, which are systems that use a single processing architecture to achieve the functionality
of different intelligent processing techniques. An example is a neural network that tries to perform symbolic
tasks such as step-wise inferencing (84).
METHODOLOGIES
Nowadays, the perspective of knowledge transfer (eliciting knowledge from the expert and translate it to a tool
11
using some kind of software methodology) has been substituted by the perspective of knowledge modelling. This has
been achieved using model-based methodologies that approach the complex problem of knowledge engineering by
constructing different aspect models of the human knowledge involved in some complex domain. There are several methodologies available, such as CommonKADS (85),
MIKE (86), Protege-II (87), etc. All these approach knowledge acquisition and modelling from a structural point of
view and try to palliate the knowledge acquisition bottleneck.
The last methodology, Protege, allows also for the definition of ontologies. Ontologies make possible knowledge
sharing and reuse, playing a major role in supporting information exchange across various networks. An ontology describes the concepts and relationships that are important
in a particular domain, providing a vocabulary for that domain as well as a computerized specication of the meaning of terms used in the vocabulary. Ontologies range from
taxonomies and classications, database schemas, to fully
axiomatized theories. In recent years, ontologies have been
adopted in many business and scientic communities as a
way to share, reuse and process domain knowledge. Ontologies are now central to many applications such as scientic
knowledge portals, information management and integration systems, electronic commerce, and semantic web services. Ontologies applied to the World Wide Web are creating the Semantic Web (88). There are several groups
that attempt to implement content management infrastructure and support the management of the vast amount
of knowledge encoded in clinical systems. These ontologies and rules are served up through applications and services to support guided observation capture, guided ordering, and guided interpretation of clinical data. Workow
portals leveraging this knowledge include the electronic
health record for care-givers and consumers, quality performance management, and clinical research.
BIBLIOGRAPHY
1. G. L. Solomon, M. Dechter, Are patients pleased with computer use in the examination room? J. Fam. Pract., 41:
241244, 1995.
2. P. L. M. Kerkhof, M. P. van Dieijen-Visser (eds.), Laboratory Data and Patient Care. New York: Plenum Publishing,
1988.
3. J. F. Sowa, Conceptual Graphs, Information Processing in
Mind and Machine. Reading, MA: Addison-Wesley, 1984.
The ICD is a widely accepted system that organizes all possible medical diagnoses. The tenth version has been translated for
worldwide application. Developed by the Commission on Professional and Hospital Activities, Ann Arbor, Michigan.
2 8600 Rockville Pike, Bethesda, MD 20894. The NLM releases
news bulletins and provides information on UMLS and contracts
for cooperation.
3 The UMLS is a project initiated by the NLM and distributed
on CD-ROM. Cooperation with parties to implement the system
within their own environment is encouraged but requires a contract (25).
12
4. P. W. Moorman et al., A model for structured data entry

based on explicit descriptional knowledge. Meth. Inf. Med.,
33: 454463, 1994.
5. P. L. M. Kerkhof, Knowledge base systems and medical terminology. Automedica, 14: 4754, 1992.
6. A. L. Rector et al., Medical-concept models andl medical
records: An approach based on GALEN and PEN&PAD J.
Am. Med. Informatics Assoc., 2: 1935, 1995.
7. P. L. M. Kerkhof, Woordenboek der geneeskunde N/E & E/N
(Dictionary of Medicine) 2nd printing. Houten, The Netherlands: Bohn Staeu Van Loghum, 1996.
8. B. G. Firkin, J. A. Whitworth, Eponyms. Park Ridge, NJ:
Parthenon Publication Group, 1987.
9. E. A. Murphy, The Logic of Medicine. Baltimore: Johns Hopkins University Press, 1976.
10. V. L. Yu, Conceptual obstacles in computerized medical diagnosis. J. Med. Philos., 8: 6783, 1983.
11. M. C. Weinstein, H. V. Fineberg, Clinical Decision Analysis.
Philadelphia: W. B. Saunders, 1980.
12. S. A. Eraker, P. E. Politser, How decisions are reached; physician and patient. Ann. Intern. Med., 97: 262268, 1982.
13. A. J. Finkel (ed.), Current Medical Information and Terminology, 5th ed. Chicago: American Medical Association, 1981.
14. MeSH, Medical Subject Headings. Bethesda: National Library of Medicine, 1993.
15. P. L. M. Kerkhof, Dreams and realities of computer-assisted
diagnosis systems in medicine (Editorial). Automedica, 8:
123134, 1987.
16. J. W. Hurst, The art and science of presenting a patients
problems. Arch. Intern. Med., 128: 463465, 1971.
17. J. Verby, J. Verby, How to Talk to Doctors. New York: Arco
Publ., 1977.
18. C. Du V Florey, European Commission EPILEX, ISBN 92826-6211-X, Luxembourg, 1993.
19. R. Baud et al., Modelling for natural language understanding. Symposium on Computer Applications in Medical Care,
Washington DC, in SCAMC, 93: 289293, 1994.
20. A. Kong et al., How medical professionals evaluate expressions of probability. N. Engl. J. Med., 315: 740744, 1986.
21. P. L. M. Kerkhof et al., The Medwise diagnostic module as a
consultant. Expert Syst. Applic., 6: 433440, 1993.
22. F. T. de Dombal, Diagnosis of Acute Abdominal Pain. Edinburgh: Churchill Livingstone, 1980.
23. SNOMED International, College of American Pathologists,
Northeld, IL, 1993.
24. Physicians Current Procedural Terminology (CPT), 4th ed.
Chicago: American Medical Association, 1992.
25. D. A. Lindberg, B. L. Humphreys, A. T. McCray, The unied
medical language system. Methods Inf. Med., 32: 281291,
1993.
26. E. R. Gabrieli, A new American electronic medical nomenclature. Automedica, 14: 2330, 1992.
27. J. Chisholm, Read clinical classication. Brit. Med. J. 300:
1092, 1990.
28. J. Wyatt, Computer-based knowledge systems. Lancet, 338:
14311436, 1991.
29. W. G. Baxt, Use of an articial neural network for the diagnosis of myocardial infarction. Ann. Intern. Med., 115: 483488,
1991.
30. M. S. Blois, Information and Medicine, Berkeley: University

of California Press, 1984.
31. G. O. Barnett et al., DXplain, an evolving diagnostic decisionsupport J. Am. Med. Assoc., 258: 6774, 1987.
32. B. Middleton et al., Probabilistic diagnosis using a reformulation of the INTERNIST-1/QMR knowledge base, part II
Evaluation of diagnostic performance. Meth. Inform. Med.,
30: 1522, 1991.
33. H. S. Waxman, W. E. Worley, Computer-assisted adult
medical diagnosis, subject review and evaluation of a
new microcomputer-based system. Medicine (Baltimore), 69:
125136, 1990.
34. E. S. Berner et al., Performance of four computer-based
diagnostic systems. N. Engl. J. Med., 330: 17921796,
1994.
35. B. Bergeron, ILIAD, a diagnostic consultant and patient simulator. M.D. Comput., 8: 4653, 1991.
36. P. Krause et al., Can we formally specify a medical decision
support system: IEEE Expert, 8: 5661, 1993.
37. C. W. Bishop, P. D. Ewing, Framemed, a prototypical medical knowledge base of unusual design. M.D. Comput., 10:
184192, 1993.
38. STAT!-Ref. Available from Teton Data Systems, P.O. Box
3082, Jackson, WY.
39. MD-Challenger, 70 Belle Meade Cove, Eads, TN 38028.
40. Labsearch/286, software available from REMIND/1 Corp.,
P.O. Box 752, Nashua, NH 03061.
41. M. E. White, J. Lewkowicz, The CONSULTANT database for
computer-assisted diagnosis and information management
in veterinary medicine. Automedica, 8: 135140, 1987.
42. H. W. Grifth, M. R. Dambro, The 5 Minute Clinical Consult.
Philadelphia: Lea & Febiger, 1997.
43. M. L. Buyse (ed.), Birth Defects Encyclopedia. Center for
Birth Defects Information Services, Inc., 30 Springdale Ave.,
Dover, MA 02030. BD Fax service (508) 7852347.
44. DiagnosisPro by MedTech USA, P.O. Box 67543, Los Angeles,
CA 90067.
45. D. A. Waterman, A Guide to Expert Systems. Reading, MA:
Addison-Wesley, 1986.
46. A. J. Gonzalez, D. Dankel, The Engineering of KnowledgeBased Systems: Theory and Practice. NJ: Prentice-Hall, 1993.
47. J. Giarratano, G. Riley, Expert Systems: Principles and Programming. PWS Publishing Co., 1994.
48. P. Jackson, Introduction to Expert Systems. Reading, MA:
49. S. Russell, P. Norvig, Articial Intelligence: A Modern Approach, Englewood Cliffs, NJ: Prentice-Hall, 1995.
50. U. Gupta (ed.), Validating and Verifying Knowledge-Based
Systems. NJ: IEEE Computer Society Press, 1991.
51. R. M. OKeefe, O. Balci, E. P. Smith,Validating Expert System
Performance. IEEE Expert, 2: 8189, 1987.
52. R. M. OKeefe, D. E. OLeary, Expert system verication and
validation: a survey and tutorial. Artif. Intell. Rev. 7: 342,
1993.
53. D. E. OLeary, Verifying and validating expert systems: A survey, in Expert Systems in Business and Finance: Issues and
Applications, P. R. Watkins and L. B. Eliot (eds.), Chichester,
UK: John Wiley & Sons Ltd., 1993, pp. 181208.
54. J. P. Kassirer, A report card on computer-assisted diagnosis
the grade: C (editorial). N. Engl. J. Med., 330: 18241825,
1994.

55. A. Alonso-Betanzos et al., FOETOS in clinical practice: A retrospective analysis of its performance, Artif. Intell. Med. 1:
9399, 1989.
56. V. Moret-Bonillo, et al., The PATRICIA project: A semanticbased methodology for intelligent monitoring in the ICU,
IEEE Eng. Med. Biol. 12: 5968, 1993.
57. A. Alonso-Betanzos, et al., Computerized antenatal assessment: The NST-EXPERT project, Automedica, 14: 322,
1992.
58. A. Alonso-Betanzos, et al., The NST-EXPERT project: the
need to evolve, Artif. Intell. Med., 7: 297314, 1995.
59. C. Hernandez, et al., Validation of the medical expert system
RENOIR, Comput. Biomed. Res. 27: 456471, 1994.
60. V. Moret-Bonillo, E. Mosqueira-Rey, A. Alonso-Betanzos, Information analysis and validation of intelligent monitoring
systems in intensive care units, IEEE Trans. Inf. Technol.
Biomed. 1: 8799, 1997.
61. G. Guida, G. Mauri, Evaluating performance and quality
of knowledge-based systems: Foundation and methodology,
IEEE Trans. Knowl. Data Eng. 5: 204224, 1993.
62. K. Clarke, et al., A methodology for evaluation of knowledgebased systems in medicine, Artif. Intell. Med., 6: 107121,
1994.
63. R. S. Ledley, L. B. Lusted, Reasoning foundations of medical
diagnosis, Science, 130: 921, 1959.
64. A. R. Feinstein, The haze of Bayes: The aerial palaces of
decision analysis and the computerized ouija board. Clin.
Pharma. Therap. 21: 482496, 1980.
65. R. O. Duda, P. E. Hart, N. J. Nilsson, Subjective Bayesian
methods for rule-based inference systems, Proc. Nat. Comput. Conf., pp. 10751082, 1976.
66. E. H. Shortliffe, B. G. Buchanan, A model of inexact reasoning
in medicine, Math. Biosci., 23: 351379, 1975.
67. J. F. Lemmer, Condence Factors, Empiricism and the
DempsterShafer Theory of Evidence, in Uncertainty in Articial Intelligence, L. N. Kanal and J. F. Lemmer (eds.), Elsevier, 1986, pp. 117126.
68. K. Ng, B. Abramson, Uncertainty management in expert systems, IEEE Expert, 5: 2948, 1990.
69. L. A. Zadeh, The role of fuzzy logic in the management of
uncertainty in expert systems, in Fuzzy Sets and Systems,
1983, pp. 199227.
70. H. J. Zimmerman, Fuzzy Set Theory and Its Applications,
Dordrecht, The Netherlands: Nijhoff, 1987.
71. G. Shafer, A Mathematical Theory of Evidence, Princeton
University Press, 1976.
72. L. A. Zadeh, Fuzzy sets, Inf. Contr. 81: 338353, 1965.
73. L. A. Zadeh, Knowledge representation in fuzzy logic, IEEE
Trans. Knowl. Data Eng. 1: 89100, 1989.
74. E. Mosqueira-Reyand V. Moret-Bonillo, Validation and Usability Analysis of Intelligent Systems: An Integrated Approach The IPSI-BGB Transactions on Advanced Research,
vol. 1,pp. 3747, 2005.
75. M. Ivory, M. Hearst, The State of the Art in Automated Usability Evaluation of User Interfaces. ACM Computing Surveys.Vol. 33, No. 4, 2001, pp. 173197.
76. L. Adelman, S. L. Riedel, Handbook for Evaluating
Knowledge-Based Systems, Kluwer Academic Publishers,
Boston, 1997.
77. J. Preece, A Guide to Usability: Human Factors in Computing, Addison-Wesley, Reading, MA, 1993.
13
78. S. K. Card, T. P. Moran, A. Newell, The Psychology of Human

Computer Interaction. Erlbaum, Hillsdale, NJ, 1983.
79. J. Nielsen, Heuristic Evaluation, in: Nielsen, J. Mack, R. L.
(Eds.). Usability Inspection Methods. John Wiley & Sons,
New York, NY, pp 2562 1994.
80. W. Edwards, J. R. Newman, Multiattribute Evaluation. Sage
University Paper series on Quantitative Applications in the
Social Sciences. Sage Publications, Newbury Park, CA, 1982.
81. T. L. Saaty, The Analytic Hierarchy Process, McGraw-Hill,
New York, 1980.
82. S. Goonatilake, S. Khebbal, Intelligent Hybrid Systems. John
Wiley and Sons, 1995.
83. B. Guijarro-Berdinas,
A. Alonso-Betanzos, O. FontenlaRomero. Intelligent analysis and pattern recognition in cardiotocographic signals using a tightly coupled hybrid system.
Articial Intelligence, Vol. 136, N 1, pp 127, 2002.
84. V. Aijanagadde, Venkat and L. Shastri. Rules and Variables
in Neural nets. Neural Computation, Vol. 3, N 1, pp. 121134,
1991.
85. G. Shreiber, H. Akkermans, A. Anjewierden, R. De Hoog, N.
Shadbolt, W. Van de Velde and B. Wielinga. Knowledge engineering and management. The CommonKADS methodology.
Third Edition. MIT Press, 2002.
86. J. Angele, S. Dekel, R. Perkuhn, and R. Struder. Developing
knowledge-based systems with MIKE. Journal of Automated
Software Engineering, Vol 5, n 4, pp. 326389, 1998.
87. Protege.
http://protege.stanford.edu/index.html.accessed
November, 20th- 2006.
88. IEEE Intelligent Systems. Special issue on The Semantic
Web: A brain for humankind. Vol 16, Issue 2, 2001.
89. M. S. Blois, Information and Medicine. Berkeley: Univ. California Press, 1984.
90. P. L. M. Kerkhof, M. P. van Dieijen-Visser (eds.), Laboratory
Data and Patient Care. New York: Plenum Press, 1988.
91. E. A. Murphy, The Logic of Medicine. Baltimore: Johns Hopkins Univ. Press, 1976.
92. J. A. Reggia, S. Tuhrim (eds.), Computer-Assisted Medical
Decision Making. New York: Springer, 1985.
93. E. H. Shortliffe, Computer programs to support clinical decision making. J. Am. Med. Assoc., 258: 6166, 1987.
94. E. H. Shortliffe, L. E. Perrault (eds.), Medical Informatics: Computer Applications in Healthcare. Reading, MA:
95. R. H. Taylor, S. Lavallee, G. C. Burdea, R. Mosges (eds.),
Computer-Intergrated Surgery. Cambridge, MA: MIT Press,
1995.
96. B. T. Williams, Computer Aids to Clinical Decisions. Boca
97. J. L. Bowen: Medical education; educational strategies to promote clinical diagnostic reasoning. New Engl. J. Med. 355(
2006) 22172225.
98. Alphabetical
list
of
disease
names
and
corresponding
ICD-9-CM:
http://www.dmi.
columbia.edu/hripcsak/icd9/2indexa.html
or:
http://icd9cm.chrisendres.com/index.php?action=contents
99. Public
Health
Image
Library:
http://phil.cdc.gov/
phil/whatsnew.asp
100. Software for PDA: http://www.epocrates.com/index.html
101. http://www.merckmedicus.com/pp/us/hcp/templates/tier2/
PDAtools.jsp
102. http://www.skyscape.com/index/home.aspx
14
103. Differential
Diagnosis:DXplain:
http://www.lcs.
mgh.harvard.edu/
104. Gideon: www.gideononline.com
105. Iliad: http://premieremedical.safeshopper.com/276/1904.htm
106. Scientic and Technical Acronyms, Symbols, and Abbreviations: http://www3.interscience.wiley.com/cgi-bin/mrwhome/
104554766/HOME
PETER L. M. KERKHOF
Medwise Working Group
Maarssen, The Netherlands
AMPARO ALONSO-BETANZOS
VICENTE MORET-BONILLO
University of Coruna
508
MEDICAL INFORMATION SYSTEMS

Medical information systems (MIS) support and enable the
diagnosis, therapy, monitoring, and management of patients
and also the management of health care organizations and
their resources delivering care to patients. Medical information systems are used in the same way as information systems in business and manufacturing. They support and automate certain tasks. They provide new ways to carry out
patient care. In business, information technology is a strategic tool in improving competitiveness. Similarly, in health
care it could empower the actors (clients/patients, care providers, and those paying for and organizing care delivery) to
achieve more health with less cost.
In the United States the application of information technology (IT) to health care started with the need to manage patient
admissions, discharges, and transfers (ADT) in order to bill the
financiers for the care provided to each patient. In European
countries, health systems generally have a different incentive
scheme with mostly publicly funded health systems. Consequently, costing and billing were not initially the focus. Instead
IT migrated into health care through the different departments
that benefited from IT. Microelectronics and digitization revolutionized medical devices. Devices today are embedded computers supporting such applications as computed tomography
and magnetic resonance imaging. Picture archiving and communication systems (PACS) grew as a solution to the need to
integrate the process of imaging and image interpretation. The
same was seen in all technology-intensive domains of health
care, such as intensive care, operating rooms, and clinical laboratories. Laboratory medicine was the first to utilize IT, both to
automate its processes and to create new improved services to
the departments requesting its services.
These opposite approaches have been merging for many
years, and currently these interests are highly integrated. At
the same time, the role of IT, and of information systems, in
particular, has changed. The importance of information (and
knowledge) to health care organizations is recognized. Today
many health care organizations employ a chief information
officer (CIO) and have created and maintained an information
management (IM) strategy. The IM concept has evolved to
cover the management of data, information, and processes
across the enterprise, including the management of the IT
and IS infrastructure. Although this trend is clear, its implementations in the United States and Europe are vastly different (at least for the time being). In the United States, health
care organizations investments in IT are on the order of 3%
to 4% of operating costs, whereas in the European countries
the figure is only about half that (between 1% and 2%).
Also, the medical information systems industry has undergone major changes. The current thinking of IT as a business
enabler, through integration of relevant information and
knowledge combined with new, improved services, is supported by the availability of distributed heterogeneous computing environments. These make it possible to network departments and health care organizations onto platforms
sharing patient data across a continuum of care. These platforms integrate care institutions and extend to the homes of
individuals, and through mobile communication anyone can
be connected. This trend requires standards of interoperability and common languages, encyclopedias, and nomenclatures, to share patient and other data in context.
A myriad of medical information systems exist today, from

home-grown local applications to off-the-shelf software products. The user has a choice in selecting which product to use.
The range of choice is also a drawback in that the market is
often nationally or regionally organized, with the result that
there are too many competing solutions with a limited installation base. Vendors therefore do not generate enough revenue to maintain and update their products, much less invest
in R&D for a new generation solutions. In such highly fragmented markets there is little incentive to push for standard
solutions. Consequently, the user has to be well informed to
understand all the options and consequences in selecting
products.
FRAME OF REFERENCE FOR MIS
Medical Domains and Architectures
A greatly simplified model describing the role of medical information systems in health care service delivery is given in
Fig. 1. It divides a health care unit into three parts: one containing the mission, vision, strategies, and goals of the organization; one comprising the operational system of processes,
resources, and knowledge; and one with the information systems supporting and enabling operations and achievement of
goals. The model illustrates two important concepts. First, it
emphasizes that any health care organization must have an
explicit strategy on how it intends to meet its goals and fulfill
its mission. Second, the strategy must also cover the ways
that IT and information systems can be used to further the
purposes of the organization.
An organization can also be represented as an architecture. Architectural views differ depending on the way activities are organized and managed. One way to organize them is
according to the hierarchical view of departments and clinical
units. Another way to represent a health care organization is
to present it as service lines supported by service units (e.g.,
clinical laboratories, operating theaters, imaging services,
etc.) and ancillary support services. This latter process view
also fits well with several current paradigms, like evidencebased medicine and clinical protocols (guidelines). In either
case, an organizational architecture implies an IS architecture that complies with the needs of the operational system.
The health care domain can best be characterized as federated. Federation means that parties have negotiated the extent to which they wish to share common resources and thus
surrender their exclusive authority over those resources. The
alternatives to federation are, at the one extreme, complete
Goals
Operational system
(processes + resources)
Information
systems
Information
systems
support
operations
Figure 1. Simplified model of a health care organization.
1960s and 1970s

IBM Other vendors
One vendor
for HW and SW
Strategic choices:
Hardware vendor
509
1970s and 1980s
1990s
Application program
manufacturers
System integrators
Installation projects
Installation projects
Applications
Applications
Servers,
workstations
Hardware
Networks
Strategic choices:
HW and SW platform
application
manufacturer
Strategic choices:
Architectures,
system tools,
industrial SW,
alliances
Figure 2. The strategic choices that users have to make have increased in number and complexity.
independence with no common agreements, and, at the other,

unification, where the parties are governed by one supreme
authority. In health care unification is, if not impossible, at
least a very difficult and tedious task to achieve. Similarly,
isolated IS islands are not any more practical. The concept of
federation is an accurate description of current health care
systems. Units act independently, but cooperate in patient
care by sharing patient data according to mutual interests
and agreements.
Consequently, the information systems supporting and enabling operations need not be fully integrated. It is enough if
their domains overlap on those areas where data need to be
shared. Overlapping of domains means that the organization
in question jointly agrees on this. It also means that they
agree to use common terminology and classifications in the
overlapping domain and that they furthermore agree on a
common communications protocol to exchange data. Federation is actually the solution to the dilemma that resulted from
the database-centered approach of the 1980s.
At that time, integration was achieved by a common database. As the number of users and needs grew it became gradually impossible to maintain and upgrade such large monolithic systems. Hence industry was forced to find new
solutions (Fig. 2). The notion of heterogeneous, federated domains emerged as a way to manage the complexity and to
migrate toward meeting the needs of the enterprises. IT vendors have established consortia to develop standards to cope
with this, for example, International Standardization
Organization/Open Distributed Processing (ISO/ODP), Advanced Networked Systems Architecture (ANSA), and Object
Management Group (OMG) (13). Distributed client-servers
and middleware describe the current approaches.
Current mainstream IT architectures subscribe to the federation principle. Consequently, ISs are otherwise independent, except that they have interfaces through which they
communicate with other IS applications. This also means that
health care organizations can use different information systems to support different operations, with the provision that
these can communicate in their mutual federated domain.
This gives users the freedom to select best-in-class applications, but at the same time requires that attention is paid to
510
streamlining of the care processes supported and enabled by

these ISs. It also means that a systems integrator must be
highly competent in people and project management.
General
strategy
Information
systems
Information
strategy
Classification
terms
Message
standards
Middleware
ITAn Agent of Change
Architecture
New IS
environment
System
integration
Network integrator
Workstations
Project
management
Servers
Process
redesign
Network OS
Physical
network
Figure 3. System integration comprises process and IT integration

and requires strong abilities in people and project management.
the interconnection of these systems. Interoperability means

that ISs can communicate among themselves using standard
messages and protocols.
The drawback to this is that users today face a much more
complex environment than in the past. Whereas in the early
days it was enough to select a vendor for the hardware, today
the user needs to make an educated choice on a number of
issues, such as what IS architecture to have, what software
tools to use, what applications to select, and finally with
whom to manage and develop this infrastructure (Fig. 2).
Systems integration is the function where information systems are made to work together in their federated domain
(Fig. 3). Systems integration comprises, in addition to the
technical integration of information systems, the redesign or
As in business and manufacturing industries, health care

uses information technology as a strategic change agent to
improve efficiency, quality, and effectiveness, and to contain
cost. This idea is based on the unique enabling capabilities of
IT to provide new ways to diagnose, treat, and monitor patients, and to allow patients to take a more active role in this
process (Fig. 4). IT is a vehicle that empowers all actors (citizens, patients, clinical staff, management, third parties,
health policy makers, etc.). However, there are no cookbook
recipes on how to do this. If there were, the health systems of
different nations would be similar. As it is, today most countries have recognized the need to reform their respective
health care systems. Although certain similarities exist overall, countries have selected different methods and tools to do
it. This also highlights the highly heterogeneous nature of
health care and underlines the need to model it as a federated
domain. Although there are no recipes on how to do this, the
first benefit that medical information systems have provided
is the ability to make better decisions: Physicians have an
integrated view of all data relating to a patient and can combine this with current medical knowledge to arrive at diagnostic and therapeutic decisions or to reevaluate current therapy and diagnosis. Similarly, aggregation of patient resource
use and cost data allows managers at all levels to evaluate
current practices and to design new ones.
Information Management
Information management (IM) facilitates the business mission of the enterprise by managing its information, processes,
and technology (4). An illustration of what IM contains in the
case of health care is given in Table 1. IM is not concerned
with the management of data and technology onlyprocesses
are also included. Information management has a broad focus
in seeking ways to improve the functioning of the enterprise
in question. Consequently, the role of information technology
Inventing new business

oportunities
How to support health care

processes with IT?
Potential of
IT
Figure 4. IT enables the creation of new

information system products that, in turn,
support and enable new ways for providing care to patients.
Redesign of
processes
Degree of business transformation
Improving efficiency
Business scope redefinition

Business network redesign
Business process redesign
Internal integration
Localized exploitation
Range of potential benefits
Revolution
Evolution
Table 1. What Information Management Is and Is Not
Table 3. IT Leaders Duties
Is . . .
Tactical planner
Application of management principles

to information
Whole enterprise
Data, process, and technology
Resource approach
Information stewardship
Management of a critical resource
Business enabler and driver for systems and technology deployment
All forms of information
Is Not . . .
Automation of existing
processes
Department
Just systems or computers
Technically driven
Just data management
Isolated islands of data
System integrator
Collaborator
Change agent
Customer advocate
is changing. Whereas in the past it was seen as an expense,

it is now an investment (Table 2). This entails the notion that
IT costs can be recovered as benefits in improved efficiency,
quality, or effectiveness, or in reduced operational costs. A
further corollary of this is that the IT leader, or chief information officer, is charged with making this all happen (Table
3). To succeed, the CIO must be a tactical planner, systems
integrator, collaborator, change agent, and customer advocate.
MIS Standards
Standardization in MIS is composed of two issues: The first
deals with medical practice and medical data; the second, the
exchange of medical data among MIS applications. In the heterogeneous changing environment of health care, standards
are greatly needed. However, due to its complexity, standards
are difficult to establish. This is seen on all levels of activity.
The medical profession is concerned with effective and efficient care. In the standardization context this has resulted in
a number of classifications and nomenclatures spanning the
whole medical domain [e.g., International Classification of
Diseases (ICD-10) and Systematized Nomenclature of Human
and Veterinary Medicine (SNOMED) (5,6)] and specialized
domains (e.g., primary care). In recent years the need for evidence-based medicine has been increasingly recognized (7).
This is partly due to the efforts in formalizing care through
care protocols, also called clinical pathways and clinical protocols. The results of these activities have been disappointingly
slow. This has led to the realization that local care practices
need to differ mostly for two reasons. First, the local circumstances need to be respected, as well as local demographics
and epidemiology. Second, as medicine is partly an art and
Table 2. Role of Information Technology Has Changed from

an Expense to an Investment Approach
From an Expense . . .
Small sums of money
Operational view of benefits
Must we spend this?
Discontinuous and discrete spending
Cost analysis
Self-financing by cost savings in other
areas
Cost management
Expense accounting
To Investment . . .
Large sums of money
Strategic view of benefits
We must spend this!
Continuous investing
Investment appraisal
Capital planning
Benefits management
Asset accounting
511
IT plans must align with the organizations

strategies and objectives
Initiate and develop an IT plan that facilitates the organizations objectives and has
the flexibility to adjust as marketplace
changes. Optimizing the ground between
best-of-breed and one system extremes
Integration of people and cultures. Using IT
to break down interdepartmental, interdisciplinary and interorganizational barriers
Facilitator and perpetrator for collaboration.
Success of quality improvement initiatives
depends on the capture of data.
Meeting both internal and external customer
requirements. Being responsive, timely,
flexible, and open to improvement
partly hard science, there is not enough evidence to warrant

standardizing care practices across all medical problems.
Countries have initiated health reforms in order to deliver
more and better quality health care with fewer resources,
that is, cost. Thus care management has become one of the
priorities. Means and indicators are needed to measure what
resources are needed and what outcomes are produced. Diagnostic related groups (DRGs) have, since their introduction,
been applied in many forms across health care systems (8).
Managed care and disease management are recent concepts
with the same goal.
The heterogeneity of the clinical domain affects medical information systems and ways to integrate them into an interoperable infrastructure. The basic notion is federation, that
is, do not try to agree on everything. It is enough to agree on
what data and context need to be shared. Communication
takes place with messages. Open systems is also a buzzword
in health care. Although quite a lot of energy has been invested in international standardization efforts in this area,
there is not much to show for it. Instead de facto standards
developed by user groups and industries dominate. Health
Level 7 (HL7) is the most notable example of data exchange
(9), together with Digital Imaging and Communications in
Medicine (DICOM) (10). Both have also gained wide acceptance outside United States and are well supported by vendors of medical information systems and imaging modalities
and networks. The success of these comes from two factors.
First, both allow a certain degree of freedom in implementation (which also means that implementations are not necessarily compatible). Second, both rely on integration engines
(message brokers) as the hub receiving and sending messages, thus reducing the number of connections needed between medical information systems in a complex environment. Additionally, some domain specific standards exist, for
instance, for exchanging laboratory orders and results (11).
Edifact is the third category of message standards that is
being used widely. Whereas HL7 and DICOM are meant
mainly for messaging between component systems of a health
care organization, Edifact is intended to be used in asynchronous messaging between organizations (12). In Europe several Edifact-based prestandards have been produced by the
Health Informatics committee of CEN, for example, for laboratories and drug prescriptions (13).
512
Middleware is a product of recent years. It is the result of

a migration from both the applications and physical layers of
services that are common across different environments. Its
need has emerged with the three-tier architectures of distributed client-servers and object orientation. A number of international consortia are competing in developing these common
services for health care. The main contenders are CORBAmed
and Microsoft, and in Europe project consortia STAR and
HANSA (1417). However, these are ongoing activities whose
outcomes are still uncertain. For users who need to decide
today, the only practical solution are the message brokers
based on HL7 and DICOM.
Today the network infrastructure can combine a number
of technologies from the TCP/IP based Local Area Network
(LAN) and Ethernet to Integrated Services Digital Network
(ISDN) and Asynchronous Transfer Mode (ATM) switches. In
addition to copper and fiber-optic cabling, connections can
also be provided by mobile networks. Consequently bandwidth, as such, is no longer a limiting factor.
As shown above, standardization efforts in the health care
domain can be divided into two broad categories: de jure and
de facto (industrial) standards. In the former category the
work has only begun. The slow progress is due to the nature
of medicine. It is evolving at a rapid pace and at the same
time a substantial part of it is based on experience rather
than hard evidence. This is also the reason why industry
standards such as HL7 and DICOM have been so successful.
It is also the reason why federation is today the only feasible
approach in the creation of an integrated information system
infrastructure for a health care enterprise. For middleware
and physical infrastructure there are no good reasons why
health care needs its own special solutions. Instead health
care should use mainstream technologies. This way, health
care can leverage the progress and price/performance ratios
in mainstream products rather than having to rely on niche
vendors specializing in health care.
Frame of Reference
The following are a few suggestions for users to stay competitive and survive in this changing environment:
Create an information management strategy supporting
and aligned with the goals of the health care organization.
Create an open architecture for the MIS, based on mainstream IT products, system integration, integration engines, HL7, and DICOM.
Create a plan to migrate from the current ad hoc (possibly monolithic) environment to this open architecture.
Buy best-of-breed medical information systems and use
the architecture and open interfaces to create the necessary interoperability between system components.
Use IT projects to trigger change in health care processes
and in the organization.
STATE-OF-THE-ART IN MIS
A complete description of available medical information systems is beyond the scope of this article. Also, such lists are
outdated as soon as they are done. Instead, a generic ap-
proach is attempted, where the question of MIS is approached

from the medical domain dimension of the health care information framework, that is, what is the medical domain supported by the application, and what capability is included?
The basic task in health care is patient management. This
comprises three activities: diagnosis, therapy, and monitoring. The complexity of the task depends on the individual
case. The patient may have a problem that can be treated
with a single visit (or contact). Or the problem may require
referrals to other specialists and care units leading to multiple visits. These extremes illustrate the problem of providing
effective, high-quality care efficiently at an acceptable cost to
citizens. A wide range of medical information systems has
been developed over the years to support patient care and
management. As the emphasis in health care has shifted, the
MIS have evolved.
The current care environment can be captured in the following phrases: patient-centered, process oriented, clinical
guidelines, evidence-based medicine, electronic patient record,
management information systems (including case management), structured data (including nomenclatures and encyclopedias), general practitioner/gate keeper, regional seamless
care delivery, citizen-centered care and health promotion, telemedicine and the Internet, data confidentiality and data security, and solutions for independent living and security. The
following provides a description and discussion of these
phrases.
Patient-Centered and Process Oriented
There are at least two ways to look at a health care enterprise: organizational and process oriented. Until recently,
hospitals were managed by dividing them into organizational
units, like admission/discharge/billing, policlinics, wards,
care service units (especially radiology and laboratory), operating rooms, and so forth. As medical information systems
are intended to support the organization, the kind of MIS applications that resulted were in accordance with the above
organizational structure. Management of the hospital was
done through these units, leading to a situation where each
unit is optimizing its performance at the cost of others. The
job of the clinicians treating patients is to coordinate these in
the best interest of the patients. At the same time hospital
management is faced with the need to be cost-effective and to
contain costs without compromising care.
As these demands are conflicting and cannot easily be reconciled in such an organizational structure, process orientation is winning ground. It has been pioneered with great success in business and industry. Total quality management
(TQM), continuous quality improvement (CQI), just-in-time
(JIT) delivery, flexible manufacturing, logistics, and others
tell of the different facets of the process approach.
A note of warning: Although care processes should be preplanned, they cannot be completely rigid. Each care process
is unique. The needs of the patient and the means of the
health care unit determine what can and will be done. Furthermore, plans may need to be changed as new evidence
emerges.
In health care, solving the problem of a patient can be seen
as a process: a chain of partly sequential and partly parallel
diagnostic and therapeutic actions. The challenge is to manage these events in order to optimize the outcome and the use
Clinical
service
line
Patient
+ problem Secondary
care unit
Care services
Policlinics
Wards
Surgery
X-ray
Laboratory
Intensive
care
Support services
Figure 5. Process-oriented organization of health care delivery.
of resources needed during that process. The process paradigm leads to a new organizational structure for care delivery
(Fig. 5). Resources are reorganized to serve the main activity
of problem-solving. The core activity is the clinical service
line, which uses the skills of service units such as laboratory,
radiology, surgery, and wards, according to need.
In the organization unit structure MISs support different
units, including laboratories, radiology, picture archiving and
communication, pharmacy, intensive care, anesthesia and operating rooms, administration, blood-banking, kitchen, maintenance, cleaning, clinical engineering, and so forth.
This creates a need for a glue that integrates the patient
data created in the different units and makes it available
where and when needed. This is supplied by the application
program interface (API) and the message brokering technology utilizing message standards, such as HL7 and DICOM.
The patient data store is the electronic patient record. Systems integration is the function where these MIS applications
are glued together into an interoperable environment.
The resulting MIS architecture has no order. All applications are equal. This equality has created a further need. The
clinicians need to have an overview of what is happening with
the patient, that is, how the care plan that they devised is
being implemented and with what success. Therefore applications like clinicians workstation have been created to provide an integrated picture of the care process. In fact, although the organization is a top-down structure, care is
administered in processes.
In the process paradigm, the MIS architecture centers on
service. The clinician responsible for a patient designs a care
plan. The care plan may contain orders for tests and procedures that are delivered by the care service units. The care
plan is reviewed at regular intervals and when new data become available it is adjusted/redesigned according to need.
The MIS paradigm for this approach is called order/entry
(O/E). O/E systems have lately been highly successful, as
they provide a way for the clinician to be in control of the
procedures performed on the patient and/or on samples taken
from the patient.
Common Services
As the understanding of care delivery and of ways to support
it with information technology has matured, the need for an
infrastructure architecture has emerged. An architecture de-
Medical
domain
Application layer
Middleware layer
(enabling services)
Controllability
Patient
+ problem
Dependability
Referring
unit
fines what the total system is, what function it provides, and
how the pieces that make up the whole system interact. Integration with message brokering is an architecture. Some MIS
applications are used by all, whereas some serve only one
function. In other words, there are common and specific services provided by MIS applications. Additionally, there are
services that are even more common and that are needed in
all IT environments, independent of whether or not that environment has a medical purpose.
Identifying these common and health-care specific common
services is attempted by a number of consortia made up of
industry and user organizations. The Object Management
Group (OMG) is defining a common object request broker architecture (CORBA) and is in the process of identifying these
common services (3). As a part of that activity, a health care
specific task force has been established with the name CORBAmed (14). Microsofts OLE/COM and its Healthcare Users
Group (HUG) compete with OMG, although some agreements
exist on how these can coexist (15). A third approach, known
as the Andover Group, combines the strengths of both, building on HL7 (18). HL7 itself may be also a contender as it is
moved toward full object orientation (9). In the European context, a prestandard on health-care specific common services
exists. It has been produced by the medical informatics committee (TC 251) of the European Standardization Committee
(13). The health care common components (HCC) identified
are: patient, health datum, activity, resource, authorization
and concept (19). The roots of this activity are in a stream of
European Union-funded projects that started nearly 10 years
ago (16).
Another prestandard by TC 251 presents a health care information framework (HIF) that can be used to view any
health care organization and the MIS it uses (20). The HIF
comprises three views: (1) health care domain, (2) technology,
and (3) performance requirements (Fig. 6). All MIS environments must have the required functionsbe dependable and
controllable. These requirements are met by technology in
three layers. Where they are located depends on the solution.
For instance, data privacy and protection can be an integrated feature of an MIS application, or there can be a common middleware service for this across all MIS applications.
The management of data privacy and protection in a MIS environment is certainly going to be easier if that function is
located in the middleware layer than if changes and updates
require manipulation of all MIS applications.
Functionality
Clinical service chain
513
Physical layer (network/physical

infrastructure)
Figure 6. Health care information framework of three views.
514
Clinical Guidelines and Structuring of Data

The process approach of organizing care into procedures ordered to be performed leads to protocol-based care. The activities needed in handling the problem of a patient can be seen
as an instantiation of a template containing that care package. Clinical protocols are used routinely in clinical research
to determine the efficacy and effectiveness of, for example,
new drugs and medical procedures. They are also commonly
used in cancer therapy and in other medical specialities.
As each patient is an individual, protocols cannot be applied in every detail. Instead, the user must have the freedom
to apply a template according to the needs of that particular
case. Consequently, what started as protocols and decision
trees in the knowledge-based system era of the 1980s has
gradually changed so that today they are called clinical guidelines and clinical pathways (21). Strict protocols have been
challenged as cookbook medicine, demonstrating that care
cannot be prescribed from a distance.
Guidelines are compiled from medical knowledge. While
medical knowledge is universal, clinical practice is local.
Therefore, protocols need to take into account the local circumstances. Unfortunately, a large part of medical knowledge
is based on experience rather than on hard facts. It has even
been claimed that only 40% of medical procedures are based
on evidence. Evidence-based medicine is one attempt to improve that ratio (7).
The challenge of a physician is to be up-to-date on what
are current best practices in medicine. Medical textbooks,
journals, and other reference materials are the stores of that
knowledge. The question is how to access the right source at
the right time. Today a lot of this is available on the Internet
and on CD-ROMs. Additional efforts are made to improve accessibility through indexing and by setting up knowledge
servers. The fame of MYCIN (22) and other favorable conditions led to a boom in medical knowledge-based systems research in the 1980s. As results usable for clinical practice
were slow to materialize, interest faded. In hindsight, the reasons for failure were twofold. First, patients are whole human
beings and can only in rare cases be treated within the limits
dictated by the knowledge-based system. Building systems
based on all medical knowledge was, and is still, impossible.
Second, clinicians are responsible for the diagnosis and treatment of patients. They cannot be replaced by a machine. A
number of approaches were developed to address this, like
critiquing and case-based reasoning. However, they did not
meet the expectations or the acceptance of clinicians. Today,
knowledge-based decision support is used only in embedded
systems like diagnostic ECG machines. Similarly, neural networks and fuzzy systems are finding applications into which
they can be embedded, thus hiding their existence from the
user.
In addition to making medical knowledge readily available
in a clinical setting, there is another challenge related to communication between clinicians. A major part of patient data
collected during a clinical episode is unstructured. Referrals
and discharge letters are summaries of the case accompanied
by structured elements like laboratory findings. Each clinician has an individual way of writing these and must include
the whole picture with context for them to be useful to other
clinicians. Several ways have been developed to summarize
this information. These include, among others, ICD-10,
SNOMED, Read codes, and diagnosis related groups (DRGs)

(5,6,8,23). Some years ago two major projects were launched
on both sides of the Atlantic to develop translators for communication between cliniciansUnified Medical Language
System (UMLS) and Generalised Architecture for Languages,
Encyclopaedias and Nomenclatures in Medicine (GALEN), respectively (24,25). The problem they are facing is the same
that all those working in medical informatics face. They cannot change medicine; they can only support its advancement.
Consequently the path to such translators is a long one and
requires among other things progress in the area of evidencebased medicine.
Electronic Patient Record
An integrated electronic patient record that is available anywhere and at any time to those authorized has been the (almost) holy grail in MIS research and development for years.
Having the relevant parts of patient data available at the
point of care is, of course, necessary. The questions, however,
are first, which comes first, the electronic patient record or
the interoperable MIS environment producing that integration and making data accessible, and second, what is needed
to make the integrated patient record clinically useful. Integration can, of course, always be accomplished by a data repository technique, where all data are stored in one database.
This is indeed the approach taken by several MIS vendors.
However, the usefulness of such a data repository presupposes that the clinical domains involved will have the same
interpretation of the data stored. Their domains need to overlap for those data to be useful. This, in turn, implies that such
agreements exist and are implemented in clinical practice.
The federation of the domain needs therefore to extend across
the whole data store. This has large implications for the
whole organization. In other words, the issue is not to procure
an electronic patient record system, but to achieve the necessary degree of federation among and between the clinical activities to make full use of that system.
The architecture of an electronic patient record system is
an issue in itself. There the goal is to provide enough structure and flexibility to allow intelligent storage and retrieval
of patient data. In Europe CEN TC 251 has produced a
prestandard for the architecture (26) and in the United States
the Medical Record Institute is working for the same goal
(27).
Resource and Management Information Systems
Care processes also have to be managed in order to use the
available resources effectively. This requires assessment of
what resources are needed and used in a certain clinical service chain and attaching costs to these. The goal is to gather
enough data on how clinical service lines operate, in terms of
outcomes and resource utilization, in order to optimize their
use of resources and to optimize outcomes and costs. Activitybased costing, DRG, case-mix, and local variations thereof are
forms of resource management.
Continuity of Care
In current thinking the process model of care delivery extends
from the home of the patient/client to the care facilities and
back to the home covering the whole care cycle. Continuity of
care (seamless care) is necessary for high-quality care with

optimal resources. This means that several service providers
have agreed to collaborate in solving the problem of the
patient/client. This implies the need for an information network integrating the service providers, the individual care
plans, and patient data [community health information networks (CHIN) and regional information networks].
The concepts of and solutions for common services, clinical
guidelines, electronic patient records, and resource management apply equally in this environment. The only difference
is that, instead of one service provider, there are several who
have agreed to collaborate according to an agreement. As the
number of actors increases the development and implementation of an information-management strategy is more demanding.
Minimally the chain includes a primary care provider (general practitioner) and specialist unit (hospital). Increasingly
nations are using general practitioners as gate keepers and
as case managers for specialist services. The GP must then
have an information system that assists in keeping track on
how the care plan of the patient is being executed/modified,
even when the patient has been referred to another service
provider. Similarly, telemedicine applications support consultations between clinicians, thus reducing the need for patients to transfer from one site to another. Such applications
naturally need to comply with national legislation on data
privacy and secrecy.
Personal computers, the Internet, telephone networks, and
wireless communication offer additional extensions to continuity of care. Certain medical procedures can be done in the
home setting (home care). Patients themselves can perform
certain procedures (self-care). The process paradigm of ordering resources for problem-solving is being turned around in
order to recognize that the patient has a dual role as both a
subject and an object of care. In the 5th Framework Program
of the European Union, the health telematics activity revolves
around the concept citizen-centered care (28). This also
aligns with health-promotion goals of making citizens more
responsible for their well-being and health (wellness). With
mobile communication these services are available everywhere. Finally, for elderly people information systems mean
solutions for independent living and security, thus extending
their ability to remain in their normal environment when
their physical and cognitive abilities are deteriorating.
Telemedicine
Telemedicine uses telecommunications technologies to deliver
health care over distances. It provides diagnostic, therapeutic,
monitoring, and follow-up activities as well as management,
education, and training. While the explosion of interest in telemedicine over the past few years makes it appear new, telemedicine has existed for more than 30 years. Currently it
overlaps with what is considered to be covered by the term
medical informatics (medical information systems). Examples
of issues that overlap are regional systems, the integrated
electronic patient record, and applications using the Internet.
A partial explanation of this overlap is that telemedicine is
promoted by teleoperators who are seeking means to add
value to their basic services. The medical information systems
industry is more fragmented for reasons explained elsewhere
in this article. Teleoperators approach health care from the
515
bottom up [from the telecommunications infrastructure toward the application layer, Fig. 6)]. The MIS industry delivers applications and systems integration services. As health
care delivery becomes more integrated and extends to homes
and individuals the borders between telemedicine and MIS
are disappearing.
The original meaning of telemedicine was medicine at a
distance. Developments in telecommunications, telematics,
computers, and multimedia have amended this definition.
Now the emphasis is on the access to shared and remote expertise independent of where the patient or the expertise is
located, the multimedia nature of such contacts, and the
transfer of electronic medical data (e.g., high resolution images, sounds, live video, patient records) from one location to
another. Distances and geography are no longer obstacles to
delivering timely and quality health care.
Teleradiology is the most often cited telemedicine application. Other applications include dermatology, ophthalmology,
pathology, psychiatry, transmission of images and signals
generated by ultrasound and endoscopy and by physiological
transducers for diagnostic and monitoring purposes (29).
Taylor (30,31) separates telemedicine into systems and
services. The first deals with the technology needed to deliver
the second. Telemedicine is still mostly in the technology
phase. Numerous experiments and pilots have been conducted (and some are still running) that have established that
the technology works. However, because they have been
mostly closed environments with special funding, the pilots
have not survived in real life. Once the pilot is over it has
proved to be extremely difficult to build a convincing case to
continue with the service on a real cost basis. Teleradiology,
however, is an exception. There is evidence that it is costeffective at case loads that are realistic in typical clinical
practice. As teleradiology has been around the longest it is
reasonable to assume that as other telemedicine services mature in the coming years they will diffuse into clinical practice. The fact that health care services are becoming integrated on community and regional basis and that the
telecommunications infrastructure necessary to support this
change is growing also strengthens the case for telemedicine.
A telemedicine system consists of input/output stations
and a communications channel. The performance requirements depend on the application. In the case of teleradiology
these include:
Image capture, either directly from digital imaging modalities or indirectly from films scanned with digitizers.
Transmission of image and associated patient data
through a data channel. Depending on the speed requirements, the channel can be an ordinary phone line, an
ISDN line, or even ATM. Satellite communications is
used.
Because the size of a digitized X-ray image file is large
(an image of 1000 pixels and a 12-bit gray scale means
that the file size is 12 Mbit) and the bandwidth of the
data channel is limited, the files are usually compressed
at the sending side. Efficient compression algorithms are
lossythat is, all image detail cannot be re-created during decompression. Much effort has been invested into
researching what compression ratios are acceptable in
various radiology applications.
516
Workstations to display X-ray images and associated patient data. The features needed are different at the sending and receiving sides.
User interface to operate the system.
Teleradiology systems are either standalone or integrated with other ISs at both ends. In such cases, image
and patient data communications is usually based on DICOM and HL7 standards, respectively.
Diagnostic and therapeutic telemedicine services include teleconsulting, teleconferencing, telereporting, and telemonitoring (31). Although experiments and pilot projects for numerous telemedicine applications are being conducted, the
development of telemedicine services used in routine clinical
work has been slow. This is explained by a number of factors,
the most important of which is that a telemedicine service is
an add-on to existing services. Therefore it must either offer
benefits that cannot be disputed or replace a less cost-effective service. The argument that it provides a means to deliver
care over a distance is not enough. It must be supplemented
with facts about quality, acceptance, and cost in comparison
with the services it is replacing or augmenting. So far there
is not much data available on the utility of telemedicine with
the exception of teleradiology (32). Reimbursement polices are
another barrier for telemedicine. However, with communityand regionwide integration of service providers, this barrier
will probably disappear.
A further problem is reliability and liability. Can users
trust what they access? Who is responsible if something goes
wrong? When the expert consulted is a human being, the
usual rules of practicing medicine apply. For servers available
through the Internet, however, the situation is quite different. Consequently as this concern has been vioced mechanisms have been created to provide guidelines and certification of these servers. Health on the Net (HON) is one such
service (33).
Data Confidentiality and Data Security
Confidentiality and security of patient data are issues that
cannot be compromised. National legislation defines how the
privacy of a person (even a patient) must be protected. Other
legislation provides the framework in which health care is
practiced. All MISs, MIS environments, and information
management strategies must minimally provide what is required by the relevant laws (34). Some countries require that
all software used in health care be certified that it meets the
national regulations (35).
Organizations should establish an information risk management plan for the implementation of these requirements
into operational processes and MIS. Elements to be included
in such a plan are:
1. How authorization to access patient data is obtained
from the patient (e.g., using individual health cards)
2. How access rights of health professionals are controlled,
maintained, and verified (e.g., audit trails and strong
authentication with electronic signature)
3. How patient data are grouped with different access
rights
4. How patient data are secured (e.g., by encryption)
5. How the training and education of health care professionals and patients in these issues is organized
Data confidentiality however, is not a black-and-white issue.
In real life and especially in health care, every situation that
will arise cannot be legislated nor can the normative requirements be applied in all situations. Common sense must prevail in such situations.
THE FUTURE
The utilization of information technology applications in
health care is influenced by progress in IT, medicine, clinical
practice, and health care delivery. These elements are highly
intertwined with one feeding the others. In IT the major
trends are the Internet, Web technology, and mobile communication. Web browsers are an easy way to provide uniform
user interfaces within an organization. Similarly, Extranets
are a way for the organization to be in contact with its clients
(citizens, patients) without compromising data confidentiality
and security (although there are still doubts about the security features of Web implementations). Mobile communication, fueled by the explosive growth in cellular phones and
value-added services, seems to offer a limitless range of applications.
However, these are just technologies. They need to be applied in a way that results in benefits for the clients/patients,
users, and organizations. User organizations should be careful not to be too enthusiastic about the possibilities offered by
new technologies. New technologies obey the life cycle of
early adaptation by technology enthusiasts and then early
adapters. These provide the testing ground to perfect the
technology and to make it available at an affordable price to
all. If the technology does not survive the tests of the early
adapters it dies (36).
The process approach and the need to manage care jointly
are pushing service providers toward collaboration in order to
meet the needs of their customers and solve the problems of
their patients effectively and efficiently. The scenario of Fig.
7 and Table 4 rests with the idea that IT can integrate data
Body of
medical
knowledge
Care plans
Expected outcomes
Data to be collected
Planned resource needs
Clinical guidelines
Quality
improvement
(TQM, CQI)
Clinical
research
Patient
Care processes
Outcome
Data collected
Resources used
Resources
Organization
Figure 7. A care scenario combining care processes, plans, and clinical guidelines, and with quality improvement both at the organizational and medical research levels.
Table 4. Characteristics of a Health Care Environment of

Today and the Future
Today
Disease and illness
management
Hospital-based care
Authoritarian and profession centered
Patient record centered
Future
Health promotion and wellness, independence and security
Virtual care (front lines and centers of excellence)
Client-centered care
Seamless service chains, logistics, community health information networks
(CHIN)
and make it and medical knowledge available in the right format anywhere and at any time. From the IT viewpoint, health
care will become virtual and transparent.
The development of MIS applications that are transportable and integratable naturally starts with identifying
user needs. User involvement in the development, testing,
and evaluation phases is equally important. The concept of a
user, however, needs to be viewed as widely as possible. This
means that one should include all categories of users from
daily end users to management. It also means that efforts
should be made to involve more than one health care organization. It also means that when the resulting product is taken
into use, its costs are offset by benefits and/or savings in
other areas, thus justifying the investment in that specific
product. According to Gremy and Sessler the key elements in
this are the respect of professional identity and a mutual effort for mutual understanding (37). The medical professions
should be empowered by the MIS applications instead of being forced into one working pattern.
BIBLIOGRAPHY
1. ISO/IEC DIS 13235-1 Information technologyOpen Distributed
ProcessingOpen trading functionPart 1: Specification, ISO
Standard.
2. ANSA, Advanced Networked Systems Architecture consortium
[Online]. Available http://www.ansa.co.uk/Research/
3. OMG, Object Management Group [Online]. Available http://
www.omg.org
4. W. Robson, Strategic Management and Information Systems: An
Integrated Approach, London: Pitman, 1994.
5. ICD-10, International Statistical Classification of Diseases and
Related Health Problems. 10th rev., Geneva: WHO, 1992. Also
[Online]. Available http://www.who.ch/hst/icd-10/icd-10.htm
6. SNOMED, The Systematized Nomenclature of Human and Veterinary Medicine [Online]. Available http://snomed.org
7. Cochrane, The Cochrane Collaboration [Online]. Available http://
hiru.mcmaster.ca/cochrane/default.htm
8. DRG, Health Care Financing Administration (HCFA) DRG version 3 [Online]. Available http://www.hcfa.gov/stats/pufiles.htm
9. HL7,
Health
Level
7
[Online].
Available
http://
www.mcis.duke.edu/standards/HL7/hl7.htm
10. DICOM, DICOMDigital Imaging and Communications in Medicine, ACR-NEMA Digital Imaging and Communications in Medicine (DICOM) Standard version 3.0 [Online]. Available http://
www.xray.hmc.psu.edu/dicom/dicom home.html
11. ASTM, The American Society for Testing and Materials [Online].
Available http://www.astm.org
517
12. Edifact, United Nations directories for Electronic Data Interchange for Administration, Commerce and Transport [Online].
Available http://unece.org/trade/untdid and Electronic Data Interchange [Online]. Available http://www.premenos.com/standards
13. CEN TC 251, European Committee for Standardization, Technical Committee for Health Informatics [Online]. Available http://
www.centc251.org
14. CORBAmed, Object Management Group Activity Focusing on
Healthcare Services [Online]. Available http://www.omg.org/
corbamed/corbamed.htm
15. Microsoft HUG, Microsoft Healthcare Users Group [Online].
Available http://www.mshug.org
16. STAR, Seamless Telematics Across Regions, a European Union
Supported Project in the Telematics Applications Program, Sector Health [Online]. Available http://www.mira.demon.ac.uk/star
17. HANSA, Healthcare Advanced Networked System Architecture,
a European Union Supported Project in the Telematics Applications Program, Sector Health [Online]. Available http://
www.effedue.com/hansa/
18. Andover Group, Andover Working Group for Open Healthcare Interoperability [Online]. Available http://www.dmo.hp.com/
mpginf/andover.html
19. HISA, Medical InformaticsHealthcare Information Systems
ArchitecturePart 1: Healthcare Middleware Layer,European
preStandard ENV 12967-1 [Online]. Available http://
www.centc251.org/ENV/12967-1/12967-1.htm
20. HIF, Medical informatics, Helathcare Information Framework,
European preStandard prENV 12443 [Online]. Available http://
www.centc251.org/ENV/12443/12443.htm
21. Prestige, Patient Record Supporting Telematics and Guidelines,
A European Union Supported Project in the Telematics Applications Program, Sector Health [Online]. Available http://
www.rbh.thames.nhs.uk/rbh/itdept/r&d/projects/prestige.htm
22. MYCIN, E. H. Shortliffe, Computer-based Medical Consultations:
MYCIN, New York: Elsevier, 1976, and B. G. Buchanan and
E. H. Shortliffe, Rule-Based Expert Systems. The MYCIN Experiments of the Stanford Heuristic Programming Project, Reading,
MA: Addison-Wesley, 1984.
23. M. J. ONeil, C. Payne, and J. D. Read, Read Codes Version 3: A
user led terminology, Meth. Inform. Med., 34: 187192, 1995; also
[Online].
Available
http://www.mcis.duke.edu/standards/
termcode/read.htm
24. UMLS, Unified Medical Language System [Online]. Available
http://www.nlm.nih.gov/research/umls/UMLSDOC.HTML
25. Galen, Generalised Architecture for Languages, Encyclopaedias
and Nomenclatures in medicine, a European Union Supported
Project in the Telematics Applications Program, Sector Health
[Online]. Available http://www.cs.man.ac.uk/mig/giu
26. EHCRA, Medical Informatics, Electronic Healthcare Record Architecture, European preStandard, prENV 12265 [Online]. Available http://www.centc251.org/ENV/12265/12265.htm
27. MRI, Medical Records Institute [Online]. Available http://
www.medrecinst.com
28. Telematics Applications Report of the strategic requirements
board, Program, Sector Health Care, 1998 [Online]. Available
http://www.ehto.be/ht projects/report board/index.html
29. Telemedicine Information Exchange, TIE [Online]. Available at
http://208.129.211.51/
30. P. Taylor, A survey of research in telemedicine. 1: Telemedicine
systems. J. Telemedicine and Telecare, 4: 117, 1998.
31. P. Taylor, A survey of research in telemedicine. 1: Telemedicine
services. J. Telemedicine and Telecare, 4: 6371, 1998.
518
MEDICAL SIGNAL PROCESSING
32. D. Lobley, The economics of telemedicine. J. Telemedicine and

Telecare, 3: 117125, 1997.
33. Health On the Net Foundation, HON [Online]. Available http://
www.hon.ch/
34. R. C. Barrows and P. D. Clayton, Privacy, confidentiality and
electronic medical records, JAMIA, 3: 139148, 1996.
35. CNIL, Commission Nationale pour lInformatique et les Libertes
(CNIL) [Online]. Available http://www.cnil.fr
36. G. A. Moore, Inside the Tornado: Marketing Strategies from Silicon
Valleys Cutting Edge, New York: HarperCollins, 1995.
37. F. Gremy and J.-M. Sessler, Medical Informatics in Health Organizations and Assessment Issues, Joint Working Conf. of IMIA
Working Groups 13 and 15, Helsinki, 1998.
NIILO SARANUMMI
VTT Information Technology
518

Medical signal and image processing entails the acquisition
and processing of information-bearing signals recorded from
the living body. The extraction, enhancement, and interpretation of the clinically important information buried in these
physiological signals and images have significant diagnostic
value for clinicians and researchers. This is due to the fact
that they help probe the state of the underlying physiological
structures and dynamics. Because the signals are in general
acquired noninvasively (i.e., from the surface of the skin, using sensors such as electrodes, microphones, ultrasound
transducers, optical imagers, mechanical transducers, and
chemical sensors etc.), medical signal and image processing
offers attractive clinical possibilities.
Particularly in noninvasive (and therefore indirect) measurements, the information is not readily accessible in the
raw recorded signal. To yield useful results, the signal must
be processed to compensate for distortions and eliminate interference. The medical environment is commonly noisy, and
the recorded signals are noise corrupted. Signals and systems
engineering knowledge, particularly signal processing expertise, is therefore critical in all phases of signal collection and
analysis.
Engineers are called upon to conceive and implement processing schemes suitable for medical signals. They also play a
key role in the design and development of medical monitoring
devices and systems that match advances in signal processing
and instrumentation technologies with medical needs and requirements.
This article provides an overview of contemporary methods
in medical signal processing that have significantly enhanced
our ability to extract information from vital signals.
MEDICAL SIGNALS
The need for noninvasive measurements of signals generated
by the body presents unique challenges that demand a clear
understanding of medical signal characteristics. Therefore,
the success of signal processing applications strongly depends
on a solid understanding of the origin and the nature of the
signal.
Medical signals are generally classified according to their
origin. Bioelectrical, bioimpedance, biomagnetic, bioacousti-
cal, biooptical, biomechanical, and biochemical signals refer

to the physiological sources that they originate from or to the
induction process imposed by the measurement. The dynamic
range and the frequency content of these signals have been
tabulated (1).
The source of bioelectric signals detected from single excitable nerve and muscle cells, for example, is the membrane
potential. When bioelectric signals are sensed noninvasively
by skin electrodes as in the case of surface electrocardiogram
(ECG), which emanates from the beating heart (see also
ELECTROCARDIOGRAPHY), a multitude of excitable cardiac cells
collectively create an electric field that propagates through
the biological medium before giving rise to the measured signal. Other clinically used spontaneous bioelectric signals recorded from surface electrodes include the electroencephalogram (EEG), which records the electrical activity of the brain
(see also ELECTROENCEPHALOGRAPHY), surface electromyogram
(EMG) detected from the muscles (see also ELECTROMYOGRAPHY), electrooculogram (EOG) from the eye, and electrogastrogram (EGG) from the gastrointestinal track.
Evoked potentials (EP) which measure brain response to
specific visual, auditory, and somatosensory stimulus are also
bioelectric signals. Bioelectric signals measured invasively
(i.e., transcutaneously) with needle electrodes include the
electroneurogram (ENG) that represents the electrical activity of a nerve bundle, electroretinogram (ERG) of the retina,
electrocorticogram of the cortex, and single fiber EMG.
In bioimpedance measurements, by contrast, high-frequency (50 kHz), low-amplitude (10 A to 10 mA range) currents are applied to the tissue, organ, or whole body, and the
impedance is measured. The impedance characteristics provide information including tissue composition, blood volume,
and blood distribution.
The brain, the heart, and the lungs also emit extremely
weak magnetic fields that often complement and augment information obtained from bioelectric signals. The main impediment to wide clinical use of biomagnetic signals is the very
low signal-to-noise ratio, which presents engineering challenges in detection.
Many physiological phenomena create sound signals. The
bioacoustic signals originate from blood flow in vessels, air
flow in the lungs, the joints, and the gastrointestinal tract
and are noninvasively recorded by microphones. The development of fiber optics and photonics has given rise to a multitude of clinical applications based on biooptical signals. Blood
oxygenation, for instance, can be monitored by detecting the
backscattered light from tissue at a variety of wavelengths.
The mechanical functions of physiological systems give rise
to motion, displacement, flow, pressure, and tension signals.
Biomechanical signals are often detected by local transducers
and hence require invasive measurements because mechanical signals are rapidly attenuated. Monitoring ion concentration or partial pressure of gases in living tissue or in
extracted samples yield biochemical signals. Recent developments in biochemical signal detection underlie the design of
new-generation biosensors intended for medical as well as environmental and industrial applications.
Processing of Medical Signals
The rapid pace of development in discrete-time signal processing (2) (see also DISCRETE TIME FILTERS) coupled with the
advent of digital computing and, in particular, the rise of vir-
tual instrumentation have given rise to efficient, flexible, and

cost-effective methods to acquire and analyze biomedical data
in digital form (3). Nonelectrical medical signals are generally
transduced to electrical form to readily digitize and process
them by digital means. Continuous-time signals are preprocessed by an analog antialiasing filter prior to analogdigital
conversion (see also ANALOG-TO-DIGITAL CONVERSION). This presampling amplifier stage is introduced in order to satisfy the
sampling theorem, which demands that the sampled signal
be band-limited and that the sampling frequency exceed twice
the highest frequency contained in the signal. Standards for
a minimum sampling rate in clinical recordings have been
introduced (4). In medical signal acquisition, it is critically
important to preserve the time morphology of the waveforms;
thus, phase distortions must be avoided by selecting a presampling amplifier with linear phase within the passband (5).
Although generic signal acquisition, time-domain analysis
(see also TIME-DOMAIN NETWORK ANALYSIS), digital filtering [see
also DIGITAL FILTERS; FIR FILTERS, DESIGN; IIR FILTERS], and
frequency-domain and spectral analysis (see also SPECTRAL
ANALYSIS) methods are often used, the low amplitudes and
strong interference from other bioelectric signals and noise
contamination typical in medical signals require special treatment (614). Fourier decomposition implemented using the
fast Fourier transform (FFT) is widely used to study the frequency or harmonic content of medical signals. When the frequency content of the interfering signal (e.g., high-frequency
muscle contraction or EMG noise) does not overlap with that
of the signal under study (e.g., ECG), digital filters can be
used to separate the two signals (e.g., ECG can be low-pass
filtered to reject the EMG). Similarly, slow baseline drift resulting from low-frequency motion artifacts and effects of
breathing in bioelectric recordings can be eliminated by highpass filtering. The 50 Hz or 60 Hz power-line noise can be
drastically reduced by notch filtering. As mentioned earlier,
the need to prevent phase distortion in medical signal processing suggests the use of linear-phase finite impulse response (FIR) filters with center symmetric impulse response
or, equivalently, filter parameters. Several powerful design
methods have been developed for linear-phase FIR filters and
are available in commercially available digital signal processing libraries.
Digital filters can not be used, however, when the frequency content of the signal and the interference are not distinguishable. A case in point is the extraction of evoked potentials from the background EEG. An understanding of the
deterministic or stochastic nature of the underlying physiological process is necessary for the classification of a medical
signal as such, hence the selection of the appropriate processing methods and algorithms. Signals described by exact
mathematical formulation, graphical form, rule, or look-up table are recognized as deterministic signals. Nondeterministic
or stochastic signals defy mathematical formulation and
evolve in an unpredictable random manner (see also STOCHASTIC SYSTEMS). In the detection of EP from the EEG, it is natural to assume that EP is a deterministic signal embedded in
EEG, which is assumed to be a zero-mean stochastic signal.
Signal averaging with proper time alignment of the EPs is a
commonly used technique whenever the signal (EP) and the
noise (background EEG) are uncorrelated and the noise is
characterized by a white (i.e., constant) spectral distribution.
The nonparametric spectral analysis of medical signals,
which can be assumed to be stationary stochastic processes,
519
can be accomplished using algorithms based on direct Fourier

transform of the time series or the inverse Fourier transform
of the autocorrelation sequence. To taper the edges of data
records to reduce leakage, time-windows are applied to time
sequences and lag-windows to autocorrelation sequences.
Parametric spectral analysis casts the estimation problem
into a framework in which the medical signal is modeled as
the response of a rational system function to a white
Gaussian noise signal. Powerful methods to identify the
model parameters have been introduced, and test criteria to
select the appropriate order and type of the model have been
developed. Depending on the problem at hand, these models
can have all-pole or auto-regressive (AR), all-zero or moving
average (MA), or pole-zero or ARMA form (15).
Conventional linear signals and systems theories have
been founded on assumptions of stationarity (or, equivalently,
time-invariance in deterministic systems) and Gaussianity
(or, equivalently, minimum-phaseness). Also, the assumption
of a characteristic scale in time constitutes an implicit assumption. Investigators engaged in the study of medical signals have long known that these constraints do not hold under the test of practice. There is evidence that new methods
are better suited to capture the characteristics of medical
signals.
The investigation of higher-order statistics and polyspectra
in signal analysis and system identification fields has provided processing techniques able to deal with signal nonlinearities and non-Gaussian statistics as well as allowing for
phase reconstruction. In cases where the power spectrum is
of little help, higher-order spectra (HOS) calculated from
higher-order cumulants of the data can be used. The aim behind using HOS rather than any other system identification
method is threefold:
1. Suppress Gaussian noise of unknown mean and variance in detection and parameter estimation problems;
2. Reconstruct phase as well as the magnitude response of
signals or systems;
3. Detect and characterize the nonlinearities in the signal
(16,17).
The power of neural-network based methods in medical
signal processing is widely acknowledged. In essence, neural
networks mimic many of the functions of the nervous system.
Simple networks can filter, recall, switch, amplify, and recognize patterns and hence serve well many signal processing
purposes. Neural networks have the ability to be trained to
identify nonlinear patterns between input and output values
of a system. The advantage of neural networks is the ability
to solve algorithmically unsolvable problems. Neural networks are trained by examples instead of rules. The most
widely used architecture of neural networks is the multilayer
perceptron trained by an algorithm called backpropogation
(18,19). (See also NEURAL NET ARCHITECTURE.)
Time-frequency representation (TFR) of a one-dimensional
signal maps the time-varying (nonstationary) signal into a
two-dimensional set where the axes are time and frequency.
There are certain properties that a TFR should satisfy: covariance, signal analysis, localization, and inner products. Many
innovative TFRs have been proposed to overcome the problems inherent in others, including the spectogram, Wigner
520
distribution, Cohens class of TFRs, affine class of TFRs, and

scalograms (20,21).
Many physiological dynamics are broadband and hence
evolve on a multiplicity of scales. This explains the recent success and popularity of medical signal processing applications
based on time-scale analysis and wavelet decomposition
(22,23). The wavelet transform responds to the need for a tool
well suited for the multiscale analysis of nonstationary signals by using short windows for high frequencies and long
windows for low frequencies (24). Continuous wavelet transform (CWT) decomposes a random signal onto a series of wide
bandpass filters that are time-scaled versions of a parent
wavelet. The parent wavelet is in fact a bandpass filter with
a given center frequency and parametrized by a scale factor.
As this factor is changed, the time domain representation of
the parent wavelet is either compressed or expanded. As the
scale factor is increased in value, the bandwidth of the signal
in the frequency domain is compressed and vice versa. This
duality between the time-frequency domains is the advantageous side of the wavelet transform that results in an increased time-frequency localization. However, the signal is
expressed in the wavelet-domain as a function of time and
scale rather than frequency components as is the case for the
Fourier transform. This property is of utmost advantage
when the signal in interest has multiscale characteristics.
The discrete wavelet transform (DWT) uses orthonormal
parent wavelets for its basis in handling multiscale characteristics of signals. When a dyadic orthonormal basis is chosen,
a statistically self-similar, wide sense stationary random signal can be decomposed by DWT to yield wavelet coefficients
parametrized by the scaling (dilation) factor and the translation factor. It has been demonstrated that orthonormal bases
are well suited for the analysis of 1/f a processes. 1/f a or
power-law processes are characterized by a power spectrum
that obeys the power-law attenuation (25).
A large class of naturally occurring phenomena exhibit
power-law or 1/f a type spectral attenuation over many decades of frequency and, hence, scale or reveal new details
upon time magnification. Medical examples include biological
time series such as voltages across nerve and synthetic membranes and physiological signals such as the EEG and the
heart rate variability (HRV) (26,27). HRV is analyzed to diagnose various disorders of the cardiovascular and the autonomous systems. HRV is derived from the heart beat signal by
detecting the peaks of the QRS waveforms in the ECG. If the
duration between consecutive beats is presented as a function
of the beat-number, the resulting HRV is referred to as a
tachogram.
The application of scaling concepts and tools such as wavelets to medical signals has uncovered a remarkable scaling
order or scale-invariance that persists over a significant number of temporal scales. Methods to capture scaling information in the form of simple rules that relate features on different scales and to relate them to physiological parameters are
actively investigated. As in the case of spatial fractals that
lack a characteristic length scale and consist of a hierarchy of
spatial structures in cascade, 1/f a processes cannot be described adequately within the confines of a characteristic time
scale. This property is beginning to be recognized as the dynamical signature of the underlying complex physiological
phenomena (28).
Compression of Biomedical Signals

Medical signal processing applications generate vast amounts
of data, which strain transmission and storage resources. The
need for long-term monitoring as well the desire to create and
use multipatient reference signal bases places severe demands on storage resources. Future ambulatory recording
systems and remote diagnosis will largely depend on efficient
and preferably lossless biomedical data compression. Data
compression methods tailored to biomedical signals aim to
eliminate redundancies while retaining clinically significant
information (29).
Signal compression methods can be classified into lossless
and lossy coding. In lossless data compression, the signal
samples are considered to be realizations of a random variable, and they are stored by assigning shorter or longer codewords to sample values with higher or lower probabilities, respectively. In this way, savings in the memory volume are
obtained. Entropy of the source signal determines the lowest
compression ratio that can be achieved. In lossless coding, the
source signal can be perfectly reconstructed by reversing the
codebook assignment. For typical biomedical signals, lossless
(reversible) compression methods can achieve compression ratios on the order of 2 : 1, whereas lossy (irreversible) techniques may produce ratios of 10 : 1 without introducing any
clinically significant degradation. [See also DATA COMPRESSION, LOSSY.]
The promise of medical signal and image processing will
be fulfilled when its techniques and tools are seamlessly
merged with the rising technologies in multimedia computing, telecommunications, and eventually with virtual reality.
Remote monitoring, diagnosis, and intervention will no doubt
usher the new era of telemedicine. The impact of this development on the delivery of health care and the quality of life will
be profound.
MEDICAL IMAGING
Medical images are of many forms: X ray, ultrasound, and
microscopic, to name a few. Increasingly, the formation, storage, and display of these images is based on extensive and
intensive utilization of signal and image processing. Some imaging modalities such as X-ray imaging onto film are mature
and stable, whereas other modalities, such as magnetic resonance imaging, are recent and undergoing rapid evolution.
There is a vast range of procedures: projection X-ray images
of the body, dynamic studies of the beating heart with contrast medium, and radioisotope images that depict glucose
utilization, a biochemical activity.
This section consists of the following subsections:
1.
2.
3.
4.
5.
6.
X-ray imaging,
Computer tomography,
Radionuclide imaging,
Magnetic resonance imaging,
Ultrasound, and
The ecology of medical imaging.
The first five sections deal with the medical imaging modalities. For each modality, we will describe the physical and
imaging principles, discuss the role of image processing, and
X-Ray Imaging
X-ray imaging is the most widely used medical imaging modality. A radiograph is a projection image, a shadow formed by
radiation to which the body is partially transparent. In Fig.
1 we show radiographs of a hand. The quantitative relation
between the body and the radiograph is quite complex. Some
radiographs have extremely high resolution and contrast
range. Both static and dynamic studies are performed, and
contrast agents are used to enhance the visibility of internal
structures. Recent developments in instrumentation are computed radiology and solid-state image enhancement systems.
The most commonly used image processing technique is subtraction. Even though extensive research has been performed
on image enhancement and recognition of X-ray images, there
has been very little practical acceptance of these techniques.
[See the article on X-RAY APPARATUS.]
Physical Principles. X rays are electromagnetic waves
whose frequencies are higher than those of ultraviolet light.
The frequency of X rays is normally not denoted in hertz but
rather in photon energy in (kilo) electron volts, which is re-
10
Fat
Muscle
Bone
t
I0
I1
(a)
(cm1)
describe recent developments and future trends. The last section will present the clinical and technical setting of medical
imaging and image processing and indicate the relative importance of the various image modalities.
An article of this length cannot fully describe all medical
image processing areas. We have selected the more important
methods that are currently used in modern hospitals. We do
not review microscope imaging, endoscopy, or picture archival
and communications systems (PACS). Microscopy is the essential tool of pathology, and much image pattern recognition
work has been done in this area. Endoscopic imaging is optical visualization through a body orifice (e.g. brachioscopy,
viewing lung passages; sigmoscopy, viewing the intestine) or
during minimally invasive surgery. PACS, comprising electronic storage, retrieval, transmission and display of medical
images, is a very active field, which includes teleradiology
(the remote interpretation of radiological images) is described
in a separate article on TELEMEDICINE.
521
0.1
10
100
(b)
Figure 2. (a) Interaction between X-rays and a homogeneous object.

An incident X-ray photon can be absorbed or scattered. The X-rays
that are neither absorbed nor scattered constitute the attenuated
beam, useful for imaging. Scattered radiation impairs image quality.
(b) Linear attenuation coefficient for soft tissue (muscle, fat, blood)
and for bone as a function of photon energy. Linear attenuation coefficients of different soft tissues are proportional to their densities,
which range from about 0.95 g/cm3 for fat to about 1.04 g/cm3 for
muscle.
lated to frequency by Plancks formula E h, where E is the

photon energy, h is Plancks constant, and is the frequency.
Photon energies for normal diagnostic use range from 10 keV
to 100 keV. X rays are a form of ionizing radiation; low exposures to X rays may cause cancer whereas high exposures can
be fatal. X-ray imaging systems can be analyzed by using ray
optics because diffraction effects are negligible.
Diagnostic X radiation is generated with X-ray tubes,
which produce polyenergetic radiation (radiation over a broad
range of frequencies). The primary energy (center frequency)
is determined by the voltage applied to the tube and the
amount of X-ray power by the tube current. The amount of
X-ray radiation per unit area is called exposure, measured in
roentgens, a unit proportional to the energy flux per unit
area, where the proportionality factor depends on photon energy. The amount of X-ray energy per unit mass absorbed by
a body is called the dose, measured in Grays.
If a monoenergetic (single-frequency) X-ray beam falls on
a homogeneous object as shown in Fig. 2(a), the relation between the incident and the transmitted flux is
I1 = I0 et
(1)
where I0 is the incident flux, I1 is the transmitted flux, t is the

thickness of the object, and is the linear attenuation coefficient, a property of the object that depends on its atomic composition, density, and the photon energy. Linear attenuation
coefficients for body tissues decrease over the diagnostic energy range: plots of attenuation coefficients for soft tissue and
bone are given in Fig. 2(b). X rays at different energies combine in an additive way. If a homogeneous body of constant
thickness is illuminated with an X-ray beam with spectral
intensity I0(V) in units of energy per unit area and photon
energy, the transmitted energy per unit area is given by

E = I0 (V )e(V )t dV
(2)
Figure 1. Two radiographic views of a hand. A fracture in the middle
finger can be seen more easily in the lateral view on the left. Courtesy
of Dr. Harold Kundel, University of Pennsylvania.
where V is the photon energy, (V) is the attenuation coefficient as a function of photon energy, and t is the thickness of
the object.
522
;;
;
Field-defining
diaphragm
X-ray
source
Recorder
Subject
Filter
Figure 3. A conventional X-ray imaging system.
X-ray interaction with an object in the diagnostic energy

range can occur in two ways: absorption and scattering. In
absorption the photon is stopped and gives up all its energy
in the vicinity of the interaction point, whereas in scattering
the photon gives up part of its energy and continues in a different direction. Equation (1) includes both absorbed and
scattered photons. However, scattered photons can leave the
body and impinge on the detector. Scattered radiation is undesirable: it contributes to noise in the image and reduces
contrast.
X-Ray Imaging. A typical X-ray imaging system is shown in
Fig. 3. An X-ray beam from a generator is restricted to the
appropriate region with a field-defining diaphragm, passes
through the body being imaged, and is detected with a recording system. In radiography one or several still images are
recorded. In cineradiography, a moving image sequence is recorded on film or videotape. In fluoroscopy, the image is
viewed by a radiologist or a surgeon while the patient is in
the X-ray apparatus. Images may be formed from the intrinsic contrast between tissues or by injecting or ingesting a contrast agent, a substance that enhances the radiographic difference between tissues (such as between blood and surrounding
soft tissues). Contrast studies are invasive; they typically
cause greater discomfort or danger than noninvasive procedures. We will discuss the factors that affect the image contrast and resolution and describe some X-ray image recording systems.
Image Contrast and Resolution. Image contrast is best at low
photon energies, and the energy available for image formation
increases with photon energy. These considerations lead to
the use of lower voltages for imaging thin body sections, such
as limbs and the female breast, and high voltages for thick
body parts, such as the abdomen. To see how radiological
physics affects this choice, consider the configuration shown
in Fig. 4, which shows a schematic view of a uniform soft
tissue region of thickness t containing a blood vessel of diame-
r
I0
I1
I0
I2
d
Figure 4. A schematic view of a blood vessel surrounded by soft tissue. The contrast depends on the size of the blood vessel and on the
difference between attenuation coefficients (see text).
ter d. The attenuation coefficients of tissue and blood are 1

and 2, respectively. Applying the law given in Eq. (1), the
intensity of the beam passing through the tissue is I1 I0
exp(1t), whereas the rays passing through the center of the
blood vessel have intensity I2 I0 exp[1t (2 1)d]. For
the blood vessel to be visible the contrast C
C=
|I1 I2 |
= |1 exp[(2 1 )d] |(2 1 )d|
I1
(3)
should be high. To obtain high contrast there should be a

large difference between the attenuation coefficients of the
blood vessel and surrounding tissue. Also, contrast increases
with d, the size of the object being imaged. From Fig. 2(b), we
see that it is desirable to use a low photon energy because
attenuation coefficient values decrease as energy increases.
On the other hand, I0 is limited by the need to limit the dose
to the patient, while the ability to form the image requires a
sufficiently large value of transmitted energy, which is proportional to exp(1t). Because the attenuation coefficient decreases with photon energy, we see that for large values of t
(thick body sections) we need to use higher-voltage X rays.
Contrast can be improved by introducing a material that
changes the attenuation coefficient. Contrast angiography is
the visualization of blood vessels by using externally injected
agents. Elements with high atomic number, such as iodine,
have a much higher attenuation coefficient than normal body
tissues. By injecting such a material into the circulatory system, we can increase 2 in Eq. (2), making blood vessels visible in thick body sections that require high-voltage X rays.
Contrast agents, however, disturb the body so that these procedures may be used only when the risk of not treating the
disease outweighs the danger of the diagnostic procedure.
One of the factors affecting resolution (the ability to see
small objects) is the focal spot, the size of the area on the
generator that emits X rays. The issues in this phenomenon
can be seen in Fig. 5(a), which shows a shadow cast by a small
object when irradiated with a finite source. The shadow exhibits a penumbra, a region around the edges of the projected
image of the objects in which the contrast is reduced. If the
object is too small the penumbra may be greater than the
image. The contrast of the object is reduced and may be too
low to see the object. The extent of degradation depends on
the size of the X-ray generator source (focal spot), the size of
the structure being imaged, the distance from the source to
the object, and the distance from the object to the recording
system. For best resolution the focal spot should be small, and
the object-recorder distance should be smaller than the
source-object distance: if the object is thin and in contact with
the recording surface, no blurring is caused by focal spot size.
Focal spot size is limited by energy dissipation in the X-ray
generator, and X-ray intensity on the detector varies inversely as the square of the source-detector distance, so that
an optimal resolution is obtained by an appropriate balance
of these factors.
A second limitation on resolution is imposed by the X-ray
recording system: currently available technologies require a
trade-off between resolution and sensitivity. As an example,
consider a screen-film system for radiography, which consists
of an intensifying screen placed in contact with a photographic
film [see Fig. 5(b)]. The intensifying screen is made from a
material with a high attenuation coefficient that fluoresces in
523
Penumbra
Subject
Penumbra
Incident
gamma-ray
photons
Source
Recorder
(a)
response to the absorption of X-ray photons: one X-ray photon

can produce as many as 1000 light photons. Resolution is limited by the distance between the fluorescing spot and the film:
the size of the area illuminated by one X-ray photon is proportional to the distance between absorption point and the film
emulsion. An ideal intensifying screen should absorb most incident X-ray photons, should have adequate conversion gain
(light photons per X ray), and should be thin enough to produce negligible blurring. Currently available materials and
manufacturing techniques require compromises between
these factors.
Image Recording Systems. The primary categories we consider are radiographic systems (still images) and fluoroscopic
systems (moving images). In radiography, one may use direct
film recording, screen-film systems, digital radiography systems, or screens coupled to recording cameras. Direct film recording is used for dental radiography and for imaging thin
structures, such as the hand. Even though the sensitivity of
direct film recording is not as high as for screen-film systems,
these procedures produce a relatively low dose because the
structure being imaged is relatively thin. The screen-film recorder was described in the previous section.
A computed radiography system captures (records) X rays
on a photostimulable phosphor screen. This forms a latent
image that is read by scanning the plate with a laser beam
that induces fluorescence proportional to X-ray exposure. The
scanner produces a digital image, which may be viewed on a
monitor. The plate is cleared by exposing it to visible light
and reused. This system has a wider latitude (dynamic range)
than film, although the resolution is not as high as the best
screen-film combinations. Even though the plates and scanner involve considerable capital investment, computed radiography can lead to savings in operating expense, especially
when images are stored and distributed electronically. Some
radiology departments are planning for a future when no
more film will be used.
In screen-film and computed radiography systems a latent
image is formed that requires further processing. In digital
radiography the image could be recorded with a sufficiently
large electronic camera. Another option is to optically reduce
the image produced by an intensifying screen, and to capture
the reduced image with a conventional charged-coupled-device (CCD) camera. However, with present technology, this
leads to a reduction in image quality.
Film
(b)
Figure 5. Source and recorder effect in

X-ray resolution. (a) The blurring of an
image depends on focal spot size and distances between the object, X-ray source,
and the recording medium. (b) A screenfilm system for radiography. X-rays are
captured by screen materials that emits
bursts of light photons which, in turn, expose photographic film. The cassette
shown uses two screens to provide higher
X-ray detection efficiency.
In fluoroscopy, where a moving image is viewed in real

time, best performance is achieved by using image intensifiers. Fluoroscopic exams require prolonged exposure and viewing so that dose reduction is critical. Much of this dose reduction is achieved by using systems that trade resolution for
sensitivity. Traditional radiological image intensifiers are
vacuum tubes that contain, on the input side, intensifying
phosphors coated with photoemissive targets. X rays produce
an optical image that falls on the photoemissive material,
which in turn produces an electron image emitted into the
vacuum on the interior of the intensifying tube, where they
are imaged with electron optics onto a small phosphor target
(similar to that in a cathode-ray tube display). This small optical image is easily coupled to a television camera with no
subsequent loss in image quality. The television signal can be
displayed on a monitor, recorded onto videotape, or digitized
and stored on computer media. Another option in fluoroscopy
is to use digital scan converters: rather than use continuous
exposure of the patient, images can be captured occasionally,
as required, or at a slow frame rate. These images are digitized and stored in random-access memory (frame buffer)
which subsequently generates a conventional television signal
that drives a television monitor.
X-ray Image Processing. The most common image processing procedure in X-ray imaging is subtraction angiography, where an image without contrast material is subtracted
from a radiogram produced with an injected contrast agent,
thereby enhancing the visibility of the structures containing
the contrast agent. In Fig. 6 we see a radiogram of the head
before and after contrast injection, and the difference image.
The arteries are clearly seen. This concept was introduced in
the 1930s, when photographic processing was used to produce
the difference image, but contemporary digital image processing techniques vastly improve the speed and convenience
of these procedures.
Image processing for radiological image enhancement was
first proposed in the 1960s. It seems plausible that such processing can improve contrast and correct for image blurring
caused by finite focal spot size. Although appropriate contrast
manipulation for image enhancement is effective in fluoroscopy systems, where an electronic image is viewed on a monitor, resolution enhancement has met no success, and there
are as yet no accepted techniques for enhancing film radiograms. This is probably due to the fact that film radiography
524
indicates suspicious areas to a radiologist, who makes the final decision. This type of interactive diagnosis system has the
promise of combining a radiologists extensive knowledge and
ability to interpret complex images with a machines capability to perform a thorough, systematic examination of a large
amount of data.
Computer Tomography
Figure 6. Subtraction angiography. Radiograph of a head before (a)

and after (b) contrast agent injection. The difference image (c) shows
only the arteries. Courtesy Dr. Barry Goldberg, Thomas Jefferson
University.
systems are optimized with the proper choice of X-ray focal

spot size and recording system sensitivity and resolution, so
that detail enhancement increases noise as well as resolution
producing no noticeable improvement in image quality.
One of the hurdles in successful implementation of X-ray
image enhancement is the high resolution and contrast range
in film images. For example, mammography (imaging of the
female breast for cancer screening) produces images 25 cm
20 cm, which may have significant detail as small as 50 m
in extent. This may require digitization to 6000 6000 pixels
with 16-bit density resolution. Merely scanning such images
in a way that preserves all the visual detail is a substantial
task with present technology. Obtaining images that can be
enhanced to show detail not apparent on the original film image may require even higher-quality digital image acquisition.
Extensive research has been done on systems for automatic interpretation of radiograms. Some success has been
achieved in the measurement of the cardiothoracic ratio (the
ratio of the width of the heart to the width of the chest), measurement of arterial stenosis (narrowing) in contrast arteriograms, and detection of degenerative change in lungs.
These methods have not yet achieved practical acceptance.
Some reasons for the lack of progress is the extreme complexity of X-ray images, which are projections (shadows) of threedimensional bodies, and many structures may overlap in one
area. Radiologists, who undergo extensive training, visualize
the three-dimensional structures in the body, compare a specific image with others that they have seen before, and base
their diagnosis on an extensive knowledge of anatomy and
pathology. It is, at present, not feasible to construct an automatic system that can operate at this level. One recent line
of research is in promising computer-assisted diagnosis (CAD)
which is being explored for screening mammography interpretation. A processing algorithm examines a scanned image and
A tomogram is an image of a section through the body. Tomographic X-ray images were first formed in the 1930s by moving the source and film in such a way that one plane through
the body remained in focus while others were blurred, but
they were used for only very specialized investigations. X-ray
computer tomography, which is abbreviated as Cat or X-Cat,
was introduced in 1973 and gained almost instantaneous acceptance.
In Cat, transmitted X-ray intensity measurements are
made for a large number of rays passing through a single
plane in the body. These measurements are subsequently processed to compute (reconstruct) the values of attenuation coefficient at each point in the plane. This, in effect, produces a
map of the density through the tomographic section plane.
The geometric arrangement for performing these measurements is shown in Fig. 7. A ring of X-ray detectors surrounds
the body being scanned, and an X-ray generator rotates in an
orbit between the body and the detector ring. For each position of the source, intensities of the radiation passing through
the body and impinging on the detectors are digitized and
captured by a computer. A sectional image is computed from
the data collected while the source rotates through a full
circle.
Medical CT images typically have a resolution of 500
500 pixels, which is about the same as that of television, and
much lower than the resolution of radiograms. Radiograms
are projection images and superpose many anatomical structures onto one plane, but all structures in a tomogram are
shown in their correct geometrical relationship. Figure 8
shows a CT scan through the upper abdomen. The spleen
(lower right) contains a hematoma (arrow), a pool of blood
caused by an injury. The hematoma density is only about 5%
higher than that of surrounding tissues, and it would not be
visible in a conventional radiograph.
X-ray detector ring
Subject
X-ray
source
Figure 7. Schematic diagram of an X-ray tomography (CT) scanner.

Signals are collected from 500 to 1000 detectors located on a ring.
The X-ray source rotates on an orbit between the detectors and the
subject.
525
Scintillating crystal
Subject
Collimator Photomultiplier
tubes
Figure 8. CT scan through the upper abdomen. The arrow points to

a hematoma (a pool of blood) in the spleen. The density difference
between the blood and surrounding tissues is about 5%. Courtesy of
Dr. Richard Wechsler, Thomas Jefferson University.
Figure 9. Schematic diagram of a gamma camera. Only gamma rays

normal to the camera are admitted by the collimator. An electronic
system collects signals from the photomultiplier tubes and computes
the coordinates of gamma-ray photons detected by the scintillating
crystal.
CT reconstruction algorithms are perhaps the most successful use of image processing in medicine. Unlike X-ray projection images, which are inherently analog, CT images are
inherently digital and are impossible to obtain without sophisticated signal and image processing technology.
a plate of scintillating material, such as sodium iodide. Each

gamma ray photon produces a pulse of light, which is detected
by a set of photomultiplier tubes. The current pulses from the
tubes differ; the output of the detector closest to the point of
scintillation is largest, and those farther from the source are
smaller. The pattern of these current pulse intensities is used
to compute the location of scintillation at a resolution much
finer than that of the tube spacing. The energy of the gamma
ray is estimated by computing the total amount of light falling on the protomultiplier tubes. Since scattered gamma rays,
which degrade the image, have lower energies and are excluded by the processing system. These gamma rays form a
severely defocused image of the object, thus reducing the contrast and increasing the noise level. The image of the radiation distribution is developed by recording individual detected
gamma rays and their positions.
Tomographic images of radionuclide distribution can be
formed by collecting images while a gamma camera moves
along an orbit around the body. A typical system is shown,
schematically, in Fig. 10, where three gamma cameras collect
projection data while they rotate around a patient. Because
each camera collects data for many planes through the body,
the data from such an examination facilitates reconstruction
of a three-dimensional or volumetric image. The algorithms
used for emission computer tomogaphy (ECT) are similar to
those for CT, but the reconstruction problem is more complex:
emissions from deeper structures are attenuated more than
Ga
c m
a
me ma
ra
Radionuclides are atoms that are unstable and decay spontaneously. Certain decay reactions produce high-energy photons, which are called gamma rays. There is no difference between gamma and X rays; the difference in terminology
reflects the process that produces the radiation, and not its
physical nature. In nuclear medicine a radiopharmaceutical
agent (a compound containing a gamma-ray-producing radionuclide) is injected into a body. Images formed from the emitted radiation can be used to visualize the location and distribution of the radioactivity. Radiopharmaceutical agents are
designed to migrate to specific structures and to reflect the
functioning of various organs or physiological processes. Consequently, radionuclide imaging produces functional images,
in contrast to the anatomical images formed by conventional
X rays. Some nuclear medicine procedures require only bulk
measurements of radioactivity; for example, the time-course
of radioactively labeled hippuric acid in the kidney is an indication of kidney function. In this article, we will concentrate
on the procedures that require image formation. We will describe the operation of a scintillation camera, the most commonly used device for radionuclide imaging. We will also discuss single-photon emission computer tomography (ELT), an
imaging method that produces three-dimensional data. We
will not discuss positron emission tomography (PET) scanners, which have unique technical and biological advantages
over ECT but are more expensive and are used primarily for
research.
Projection images of radionuclide distributions are most
commonly formed with a scintillation camera, which consists
of two major components: a collimator and a position-encoding radiation detector, as shown schematically in Fig. 9. The
collimator is made form a highly absorbing substance, such
as lead, and contains a set of parallel holes that admit only
radiation normal to the collimator. These gamma rays fall on
a
mm a
Ga mer
ca
Radionuclide Imaging
Subject
Gamma
camera
Figure 10. Schematic diagram of an Emission Computed Tomography (ECT) system. Three gamma cameras collect projection data from
the subject while rotating about a common center.
526
Figure 11. ECAT images of a heart. Shown are a set of slices

through the left ventricle obtained after exercise (top) and at rest
(bottom). The darker region in the upper right of the stress images
indicates an ischemic (reduced blood supply) region of the heart muscle. Courtesy of Dr. David Friedman, Thomas Jefferson University.
those from the surface, and reconstruction algorithms must

correct for this effect. There is no elegant solution to the general attenuation correction problem for ECT, although empirically developed methods produce satisfactory results.
Figure 11 shows several sections through the left ventricle
of a heart obtained with an ECT scanner. The radiopharmaceutical is absorbed by metabolically active muscle. Two scans
are obtained: one after exercise and one after a period of rest.
The image has lower intensities in regions that are not receiving blood circulation, such as parts of the heart that may have
been injured by a heart attacka blockage of circulation due
to an occlusion in an artery that supplies oxygen and nutrients to the heart muscle. The image is formed over a period
of about 10 minutes. This is necessary because the amount of
radioactivity is very low, so that photons must be collected for
a long period to produce sufficiently high signal-to-noise ratio
in the imaging. Blurring caused by heart motion is reduced
by cardiac gating; the electrocardiogram identifies times at
which the heart is in the same position, and data from these
time intervals are pooled. The resolution in Fig. 11 is very
low, but the technique provides unique information about
muscle functioning that is not available from anatomical
images.
Signal formation and image processing are an integral part
of radionuclide imaging. The formation of the image from a
gamma camera requires extensive real-time signal processing. A gamma camera produces a sequence of xy coordinates of detected gamma rays that are processed to produce
a pixel array of gray values. The image so obtained typically
contains subtle geometric distortions and nonuniformities,
which are corrected with appropriate algorithms. Further
processing is required if cardiac gating is used or if tomographic images are produced. Certain studies measure parameters based on dynamic (time-course) data and use sophisticated statistical parameter estimation techniques. The
viewing of three-dimensional data from ECT requires rendering methods that produce sectional or projection images.
Magnetic Resonance Imaging
X-ray and radionuclide imaging use electromagnetic radiation
of a frequency so high that the human body is partially transparent. Magnetic resonance imaging (MRI) uses electromagnetic frequencies in the range of 40 MHz to 100 MHz, which
also penetrate living tissues. At these frequencies, the wave-
length is too long to allow focusing as an imaging principle.

MRI uses the nuclear magnetic resonance effect to induce signals from the body. Signals from different locations in the
body are frequency-encoded by applying spatially varying
magnetic fields, and the emitted signals are reconstructed to
produce images. The principles used for MRI are unlike those
used for any other imaging technique. We will first describe
the physical principles of nuclear magnetic resonance, which
govern the production of the magnetic resonance signal. We
will subsequently describe the methods used to generate the
signals that form the images and also describe some of the
processes and properties of the body that are viewed in medical MRI.
Nuclear Magnetic Resonance. Nuclear magnetic resonance
is a process associated with the magnetic moments possessed
by the nuclei of many atoms. There magnetic moments are
present, typically, in nuclei whose atomic number (sum of the
number of protons and neutrons) is odd. Any of these nuclei
can be imaged through MRI. Medical MRI is almost exclusively based on the imaging of hydrogen nuclei or protons, so
that in the subsequent discussion we will talk about the magnetic resonance and imaging of hydrogen nuclei. Each nucleus
also has an angular momentum, which is aligned with the
magnetic moment. Magnetic resonance is based on the joint
action of these magnetic and rotational moments.
Thermal motion causes nuclear magnets to take random
orientation, but an external magnetic field (the dc field) pulls
at least some of these nuclei into alignment with the field
direction. This alignment process is not instantaneous but follows an approximately exponential time course whose time
constant, the so-called the longitutional relaxation time (T1),
depends on the physical and chemical state of the material,
and ranges from about 0.25 s to 1 s for protons in biological
tissues. Should such a nuclear magnet be deflected from the
direction of external field, a torque that tends to move it toward alignment is generated. The nuclear angular momentum resists this so that the nucleus behaves like a spinning
top: its axis maintains a constant angle in a relation to the dc
field and spins (precesses) around it. This phenomenon is
called nuclear magnetic resonance. The rate of this precession
(the Larmour frequency) is proportional to the magnitude of
the magnetic field and the proportionality constant, for protons, is 42.6 MHz/T. The magnitude of this oscillating magnetization decays exponentially at the so-called transverse relaxation time (T2), which has a range from 20 ms to 100 ms for
protons in biological tissues, depending on tissue type. The
resonance of this oscillation is very sharp: the quality factor
Q, which is a product of the resonance frequency and T2,
ranges from about 5 106 to 25 106 for protons in biological tissues.
The precessing magnetic moments set up an external oscillating field that can be detected by placing appropriate conductor loops called pickup coils near the body. These detect
an induced voltage (due to the Faraday effect), which is proportional to the product of the magnetic moment and the frequency. Both the magnetic moment and the frequency are
proportional to the external magnetic field; consequently, the
voltage is proportional to the square of the field. It therefore
is desirable to have as high a dc field as possible, and magnets
with 1.5 T are now fairly common. To deflect (excite) nuclear
magnetic moments from their equilibrium direction, an oscillating magnetic field (the RF field) is applied at right angles
;;;;
;;;
Dc magnet
;;;;
;;;
Gradient and
RF coils
Pickup coil
Figure 12. Schematic diagram of a MRI scanner.
to the dc field. If this field is at the Larmour frequency, nuclear magnets tip away from the direction of the dc field. The
deflection angle from equilibrium ( flip angle) is proportional
to the product of the RF field strength and duration.
Magnetic Resonance Imaging Principles. A schematic diagram of a magnetic resonance imager is shown in Fig. 12. The
bulkiest and most expensive component is the dc magnet. The
RF coils excite the MR signal, and the pickup coils detect it.
To form an image gradient coils impose linearly varying magnetic fields that are used to select the region to be viewed
and to modulate the signal during readout. In a typical data
collection step, the RF field is applied in the presence of a
gradient field (slice selection gradient), so that only one slice
through the body is at the Larmour frequency. After the magnetic moments are excited, another gradient field is applied
(readout gradient) so that the moments in different portion of
the slice radiate at different frequencies. The signal from the
pickup coil is amplified and digitized. The Fourier transform
of this signal gives a map of the amount of magnetic materials
at regions of different gradient field value. A number of such
data collection steps, each with a different readout gradient,
are required to collect enough data to produce a sectional image. This sequence of data collection is programmed by
applying pulses of current to slice selection coils, RF coil, and
readout gradient coils. A magnetic resonance imager is a flexible instrument: image characteristics, such as slice thickness
and resolution are determined by this pulse sequence. The
slice orientation, thickness, and resolution can be varied.
Image intensity is basically proportional to proton concentration, or the amount of water in different tissues. Figure 13
527
shows a magnetic resonance image of the human head. This

sectional plane, located along the plane of symmetry of the
head, cannot be obtained with a single CT scan. Because bone
contains less water than the soft tissues, the skull shows a
lower image brightness. Further contrast is provided by differences in water content and chemical characteristics of various brain tissues, such as gray matter, white matter, and cerebrospinal fluid. Appropriate pulse sequences can provide
images weighted by T1 and T2, which show characteristic tissue-specific variations. Other pulse sequences are designed to
enhance signals due to differences in chemical composition
(chemical shift imaging) which produce small changes in resonance frequency. Pulse sequences can also differentiate between moving and stationary tissues. MRI can produce angiograms (images produced by flowing blood) without using
contrast agents. A relatively recent use of MRI is to form
functional images, which show areas of brain activity that are
related to sensory, motor, or cognitive brain activity. Functional imaging can be used to identify areas of the brain affected by stroke before the tissue has undergone a physical
change caused by loss of blood circulation. This can guide
treatment to restore circulation and avoid brain injury.
From this description, we see that an MRI system functions through electronic and electromagnetic devices such as
coils, amplifiers, and digital/analog converters that produce
pulse sequences and detect signals. The signals from an MRI
scanner are digitally processed to form images and to extract
information such as proton density, relaxation times, motion,
and chemical composition. Further processing is required to
correct for magnetic field nonuniformity and motion-induced
artifacts. Postprocessing of magnetic resonance images and
image sets is an important area of signal processing and pattern recognition applications. Sets of images obtained with
different pulse sequences are combined to enhance signals
caused by tumors, neurological plaques, to provide three-dimensional reconstructions of complex structures and to plan
surgical procedures.
In contrast to X-ray and radioisotope imaging, MRI uses
low-frequency electromagnetic radiation, which is nonionizing. Therefore there is no concern about producing tumors or
cancer; the only known source of harm from MRI is heating
produced by the RF fields, which can easily be monitored and
does not reach harmful levels in clinical equipment. On the
other hand, high magnetic and RF fields in MRI may interfere
with implanted electronic devices such as cardiac pacemakers, and MRI may be unusable or dangerous in the presence
of implanted metal prostheses. MRI exams are long, uncomfortable, and expensive. Because MRI scanners are large and
require extensive shielding they are available only in hospitals and other special facilities. Magnetic resonance image
resolution is lower than that obtainable from X-rays. Because
of these factors, in spite of its versatility and safety, MRI is
not likely to replace other modalities in the near future.
ULTRASOUND
Figure 13. Sectional MRI image of a head.
Acoustical imaging uses ultrasound (mechanical vibrations at

frequencies above the range of human hearing) to form echo
images of the interior of the human body. The operation of an
ultrasound scanner is illustrated in Fig. 14. A transducer
emits acoustic pulses that propagate along narrow beams
Transducer
528
Figure 14. Schematic diagram of an

ultrasound scanner. In (a), a transducer
emits a sequence of acoustic pulses that
travel along paths shown in gray. Echoes
are produced when these pulses impinge
on the front or back surface of the object.
In (b) an image is formed by displaying
echo signals along monitor scan lines.
Object
(a)
through the body. Reflections (echoes) from tissues radiate toward the transducer, which converts them to electrical energy. An image is formed by collecting a set of echoes from
beams sent out along parallel lines. We will describe the main
physical properties that govern propagation of sound in tissues and relate them to factors that affect the quality of images obtained with ultrasound. [See also the article on ULTRASONIC MEDICAL IMAGING.]
Propagation of ultrasound through tissue is governed by
the acoustical wave equation, and the wavelength of acoustical waves places critical limits on the resolution of ultrasound
scanners. Wavelength, frequency, and acoustic velocity are related by f c; in soft tissues c 1500 m/s, so that, for
example, the wavelength at 5 MHz is 0.3 mm. The resolution
of the scanner in the depth direction is limited by the duration of the acoustic pulse and in the lateral direction by the
width of the sound beam. In Fig. 15 we show the schematic
pattern of acoustical energy produced by a concave (focused)
transducer excited with a constant frequency. The transducer
has diameter D and the focus (center of curvature of the surface) is at a distance F from the transducer. The field pattern
is quite complex: approximately, the acoustic beam converges
toward the focus where its diameter w 0.6F/D. The width
of the beam is relatively constant over a depth of field d
2wF/D. To both sides of the region of focus the beam width
increases linearly with distance. To obtain high resolution
(small w), one should use a short wavelength (high frequency)
and or a large transducer of diameter D, which leads to a
small depth of field. This physical limitation can be overcome
by the principle of dynamic focusing with an array transducer. A flat transducer is decomposed into elements. When
an echo is received, signals from various transducer elements
are delayed by appropriate amounts, producing the same sort
of focusing obtained with a curved transducer surface. These
delays are varied with time, allowing focus to be maintained
for a large range of depth.
In practice, only soft tissues (muscle, fat, and fluid-filled
cavities) can be imaged with acoustical imaging. Bone and
d
Figure 15. Depth dependent beam width in a focused ultrasound
transducer.
(b)
air-filled spaces such as lungs do not propagate ultrasound

well enough and are, in effect, opaque. Because resolution is
limited by the wavelength, it is desirable to use as high a
frequency as possible. Usable frequency is limited by attenuation. Sound power in a plane wave propagating through tissue
is given by the formula P(z) P0 exp(2z), where P0 is the
power at the surface of the body, P(z) is the power at depth z
in the body, and is the attentuation coefficient, which depends on frequency. The attenuation of ultrasound in soft tissue is approximately proportional to frequency, so that higher
frequencies are attenuated more strongly and deeper structures can be viewed only with lower frequencies. The frequency used for abdominal ultrasound is about 5 MHz, and a
resolution of about 0.3 mm in the depth direction and about
0.4 mm laterally can be obtained to a depth of about 20 cm.
Smaller organs, such as the female breast, are scanned with
higher frequencies (7.5 MHz), and still higher frequencies are
used to examine the eye.
The time to receive echoes from structures at a depth of 20
cm is about 270 microseconds, so that it is possible to emit a
sequences of acoustic scanning beams and construct images
in real time, at approximately video rates. One of the important uses of ultrasound is to view the heart in real time.
Pathologies such as abnormal heart valve motion can be diagnosed from the temporal appearance of the moving image.
The relation between the acoustic echo and tissue characteristics is complex and not well understood. There are two
main sources of acoustic echoes. Reflections are produced at
tissue boundaries, where there is a change in tissue density
or acoustic velocity. These echoes appear as lines at organ
boundaries, or at boundaries between tissue and tumor. Such
reflections are not seen if the surface is parallel to the beam
direction. So-called soft-tissue echoes arise from microscopic
inhomogeneities such as blood cells and muscle fibers. An
acoustical image typically contains both types of signals. Figure 16 shows an ultrasound image of a pregnant uterus. The
head and chest of the fetus are easily seen. The body is outlined with surface echoes, while soft-tissue echoes are produced by the lungs. Much of the image is composed of speckle,
which is largely due to soft-tissue echoes, but may be caused
by focusing defects or multiple reflections. The interpretation
of echosonograms requires extensive training and experience,
perhaps more than for other image modalities.
In addition to conventional scanners, there are many other
techniques for producing acoustic images. In Doppler imaging, frequency differences between incident and reflected
pulses are used to measure blood velocity and produce images
that allow diagnosis of circulatory problems. Harmonic imaging produces signals from nonlinearities of acoustic propa-
gation. Acoustic contrast agents allow increased visibility of

tissue differences. Very small imaging transducers can be introduced into body cavities or blood vessels. They operate at
very high frequencies and produce high resolution images of
anatomy of blood flow.
The formation of acoustical images is impossible without
signal processing. Originally, acoustical scanners were constructed with analog signal processing circuitry, but modern
instruments make extensive use of digital signal processing.
Dynamic focusing of the transducer, Doppler signal detection,
and harmonic signal analysis must be performed at high rates
in real time. There is growth in the use of image processing to
enhance images produced by acoustical scanners: for example,
promising techniques have been proposed for reducing acoustical speckle.
Medical ultrasound images have a relatively low resolution
and noisy appearance. Nonetheless, echosonography offers a
number of important advantages. Ultrasound radiation is
nonionizing. Even though high levels of ultrasonic energy can
produce heating or mechanical injury, there are no known
harmful effects from acoustic radiation at the levels used in
diagnostic scanners. Consequently, ultrasonography is the
preferredalmost exclusiveimaging modality for examining the pregnant uterus, where there is great concern about
possible birth defects or tumors in the fetus that may be
caused by X radiation. Ultrasound images are formed in real
time so that the body can be explored to locate the region to
be imaged and moving structures such as the beating heart
can be viewed.
ECOLOGY OF MEDICAL IMAGING
What is an ideal medical imaging instrument? Cost, convenience, and safety are important considerations. However, the
most important issue is whether the imager can provide definitive information about the medical condition of the patient. To have a better appreciation of these issues, it is necessary to review the conditions under which medical imaging
is performed.
Medical imaging may be used as part of therapeutic procedures, to detect disease where no symptoms are present
(screening), or to monitor the process of healing. The most
common use of medical imaging is for differential diagnosis,
where a patient exhibits symptoms that can be caused by one
529
Table 1. Modalities, Subdivisions, and Annual Case Load in a

Large Teaching Hospital
Modalities
X ray
Computer tomography
Nuclear medicine
Magnetic resonance
imaging
Ultrasound
Subspecialities
Chest, bone, gastrointestinal,
genitourinary, neurological,
neurosurgical, angiography,
mammography
Body, neurological (head)
Studies
157,958
14,343
10,772
12,380
15,844
or more diseases, and there is a need to ascertain the specific

condition so that proper treatment can be applied. The physician orders a specific imaging procedure, which may require
patient preparation (for example, ingestion of a contrast material), the use of specific imaging instruments and settings
(X rays of a certain voltage, intensifying screens, and films),
and proper positioning of the patient. Most procedures are
performed by properly trained technicians, but some very
dangerous or invasive procedures, such as injecting contrast
agents directly into the heart, are performed with direct participation of a medical doctor. The majority of imaging procedures are administered by a radiology department of a hospital, and images are interpreted (read) by radiologists, medical
doctors who undergo post-MD training (residency) in this specialty. The choice of procedure for a given medical problem is
based on medical knowledge and on guidelines promulgated
by medical and/or health insurance organizations. We see
that most medical imaging tasks are highly specialized and
are performed in a very structured setting.
Table 1 shows the divisions of a radiology department of a
large teaching hospital. The divisions have been grouped by
modality. It also shows the number of studies performed in
one year in each modality. It is clear that X-ray imaging is
still the predominant technique. Other modalities are growing more rapidly, but none are likely to displace X rays in the
near future.
Medical imaging procedures and modalities exist in an extremely competitive environment. For example, heart disease
may be diagnosed either with angiography or nuclear medicine. Each procedure, to survive, must fit an ecological niche:
it must satisfy a need where it has advantages over other
methods. The same issue arises when new image processing
techniques are introduced: images produced with these procedures must have an advantage over other imaging procedures. The advantage will probably be application dependent.
An image processing technique that is effective for mammography may not be of much value for cardiology, and vice
versa.
IMAGE PROCESSING AND INTERPRETATION1
Signal and image processing play an essential and growing
role in medical imaging. To date, most of the applications of
Figure 16. Ultrasound image of a pregnant uterus. The head and

chest of a festus are shown. Courtesy of Dr. Barry Goldberg, Thomas
Jefferson University.
The work reported in this section is supported by the National Cancer Institute and the National Institutes of Health under grant numbers CA52823 and P41 RR01638.
530
signal processing have been in design and construction of imaging devices: indeed, novel modalities such as CT, radioisotope imaging, MRI, and acoustic scanning are impossible
without signal processing. There is a modest but growing application of post-processing, the use of image processing techniques to produce novel images from the output of conventional imaging devices. The most common medical images,
radiograms, are largely formed with analog methods. However, there is great promise for digital storage and transmission of these images. This is likely to be a large area for development and application of image compression techniques.
There is also the promise, to date largely unrealized, of using
pattern recognition techniques to improve on human interpretation of medical images.
As in other biomedical areas, medical image processing is
driven by the needs and constraints of the health care system.
Innovations must compete with existing methods and require
extensive clinical testing before they are accepted.
In conclusion, the fusion of several indices and local decisions leads to a more reliable global decision mechanism to
improve on diagnostic imaging.
In this part of the article, we gave a few examples of medical applications that make direct use of various aspects of machine vision technology. We also showed how to formalize
many medical applications within the paradigm of image understanding involving low-level as well as high-level vision.
The low-level vision paradigm was illustrated by considering
the problems of extracting a soft tissue organs structure from
an ultrasound image of the organ, and that of registering a
set of histological 2-D images of a rat brain sectional material
with a 3-D brain. For the high-level vision paradigm, we considered the problem of combining local decisions based on different aspects or features extracted from an ultrasound image
of a liver to arrive at a more accurate global decision.
BIBLIOGRAPHY
1. A. Cohen, Biomedical signals: Origin and dynamic characteristics; Frequencydomain analysis, in J. Bronzino (ed.), Biomedical
2. A. V. Oppenheim and R. W. Schafer, Discrete-Time Signal Processing, Englewood Cliffs, NJ: Prentice-Hall, 1989.
3. W. J. Tompkins (ed.), Biomedical Digital Signal Processing, Englewood Cliffs, NJ: Prentice-Hall, 1993.
4. M. D. Menz, Minimum sampling rate in electrocardiography, J.
Clin. Eng., 19 (5): 386394, 1994.
5. L. T. Mainardi, A. M. Bianchi, and S. Cerutti, Digital biomedical
signal acquisition and processing, in J. Bronzino (ed.), Biomedical
6. R. E. Challis and R. I. Kitney, The design of digital filters for
biomedical signal processing. I. Basic concepts, J. Biomed. Eng.,
4 (4): 267278, 1982.
biomedical signal processing. II. Design techniques using the zplane, J. Biomed. Eng., 5 (1): 1930, 1983.
biomedical signal processing. III. The design of Butterworth and
Chebychev filters, J. Biomed. Eng., 5 (2): 91102, 1983.
9. N. V. Thakor and D. Moreau, Design and analysis of quantised
coefficient digital filters: Application to biomedical signal processing with microprocessors, Med. Biol. Eng. Comput., 25 (1):
1825, 1987.
10. R. E. Challis and R. I. Kitney, Biomedical signal procesing. I.

Time-domain methods, Med. Biol. Eng. Comput., 28 (6): 509
524, 1990.
11. R. E. Challis and R. I. Kitney, Biomedical signal processing. II.
The frequency transforms and their inter-relationships, Med.
Biol. Eng. Comput., 29 (1): 117, 1991.
12. R. I. Kitney, Biomedical signal processing. III. The power spectrum and coherence function. Med. Biol. Eng. Comput., 29 (3):
225, 1991.
13. C. L. Levkov, Fast integer coefficient FIR filters to remove the
AC interference and the high-frequency noise components in
biological signals, Med. Biol. Eng. Comput., 27 (3): 330332,
1989.
14. M. Akay, Biomedical Signal Processing, New York: Academic
Press, 1994.
15. M. Akay, Detection and Estimation of Biomedical Signals, New
York: Academic Press, 1996.
16. C. L. Nikias and A. P. Petropulu, Higher-order Spectra Analysis:
A Nonlinear Signal Processing Framework, Englewood Cliffs, NJ:
Prentice-Hall, 1993.
17. A. P. Petropulu, Higher-order spectra in biomedical signal processing, in J. Bronzino (ed.), Biomedical Engineering Handbook,
18. A. S. Miller, B. H. Blott, and T. K. Hames, Review of neural network applications in medical imaging and signal processing, Med.
Biol. Eng. Comput., 30 (5): 499, 1992.
19. E. M. Tzanakou, Neural networks in biomedical signal processing, in J. Bronzino (ed.), Biomedical Engineering Handbook,
20. X. Wang, H. H. Sun, and J. M. Van De Water, Time-frequency
distribution technique in biological signal processing, Biomed. Instrum. Technol., 29 (3): 203, 1995.
21. G. F. Boudreaux-Bartels and R. Murray, Time-frequency signal
representations for biomedical signals, in J. Bronzino (ed.), Biomedical Engineering Handbook, Boca Raton, FL: CRC Press,
1995.
22. M. Akay, Time-Frequency and Wavelets in Biomedical, Signal Processing, Piscataway, NJ: IEEE Press, 1997.
23. N. V. Thakor and D. Sherman, Wavelet (time-scale) analysis in
biomedical signal processing, in J. Bronzino (ed.), Biomedical Engineering Handbook, Boca Raton, FL: CRC Press, 1995.
24. M. Vetterli and J. Kovacevic, Wavelets and Sibband Coding, Upper Saddle River, NJ: Prentice-Hall, 1995.
25. G. Wornell, Signal Processing with Fractals, Upper Saddle River,
NJ: Prentice-Hall, 1996.
26. R. W. DeBoer et al., Comparing spectra of a series of point events
particularly for heart rate variability data, IEEE Trans. Biomed.
Eng., 31: 384387, 1984.
27. H. G. Steenis et al., Heart rate variability spectra based on nonequidistant sampling: Spectrum of counts and the instantaneous
heart rate spectrum, Med. Eng. Phys., 16: 355362, 1994.
28. B. Onaral and J. P. Cammarota, Complexity, scaling, and fractals
in biomedical signals, in J. Bronzino (ed.), Biomedical Engineering Handbook, Boca Raton, FL: CRC Press, 1995.
29. E. Getin and H. Koymen, Compression of Digital Biomedical Signals, in J. Bronzino (ed), Biomedical Engineering Handbook, Boca
BANU ONARAL
OLEH TRETIAK
FERNAND COHEN
Drexel University
MEMORY ARCHITECTURE
MEDICAL TELEMETRY. See BIOMEDICAL TELEMETRY.

MEDICAL ULTRASOUND. See BIOLOGICAL EFFECTS OF
ULTRASOUND.
MEISSNER EFFECT AND VORTICES IN SUPERCONDUCTORS. See SUPERCONDUCTORS, TYPE I AND II.
MEMORIES, SEMICONDUCTOR. See SRAM CHIPS.
531
MICROELECTRODES
BIOELECTRODES
MICROPIPETTES
Microelectrodes have traditionally been developed and
used to measure voltage inside and outside biological
cells, and much of our understanding of the nervous system has come from these recordings. Microelectrodes have
been adapted with some clever arrangements to measure
membrane voltage and ion concentrations simultaneously.
More recently, microelectrodes capable of measuring partial pressures of various gases and concentrations of physiologically relevant chemical substances, such as neurotransmitters, have also been designed. The advantages of
these microelectrodes is that they allow direct measurements within biological tissues to give information about
the local microscopic milieu. Therefore, these electrodes
must be small to minimize interference with physiological function and the damage generated by their insertion. Although these small structures are more fragile than
macrosensors, they usually have better time constants.
MEMBRANE POTENTIALS
All cells in the body have a nucleus and a cytoplasm surrounded by a lipid membrane. The cytoplasm is a good conductor with a resistivity varying between 50 cm and
300 cm. The cytoplasm is separated from the outside of
the cell by a thin (7.5 nm to 10 nm), resistive and capacitive membrane composed almost entirely of proteins and
lipids. Electrical potentials exist across this membrane in
practically all cells of the body. The resting potential difference is generally around 70 mV, and the inside of the
membrane is negative with respect to the outside (See Bioelectric potentials). This resting potential is generated
by the equilibrium of two forces, the diffusion of ions across
the membrane through protein channels and the electrical
force generated by accumulation of ions at the membrane.
The resting potential is sustained by pumps that maintain
diffusion gradients across the membrane. Some cells, such
as muscle bers and neurons, are excitable and generate
large voltage signals (about 100 mV) either spontaneously
or when stimulated. Figure 1 shows the resting potential and action potential in a neuron. The action potential
is generated by nonlinear voltage-sensitive ion channels.
Sodium (Na) channels are normally at rest and open when
the membrane voltage reaches a threshold value. A large
inux of Na current depolarizes the membrane to positive
potentials. The Na channels turn themselves off (inactivation), and the potassium channels then open bringing the
membrane voltage down to it resting value (1). This action
potential is an all-or-none phenomenon and carries information from sensory inputs to the brain, from the brain to
the muscles, and within various parts of the brain. Therefore, the measurement of the membrane potential is crucial to our understanding of the activity of excitable and
nonexcitable cells. To measure the electrical activity of a

cell, the transmembranous voltage or transmembranous
current must be evaluated. An electrode must be located
directly inside or make electrical contact with the inside of
the cell. Because the cells recorded from can be as small as
1 m, electrodes with submicron dimensions must be used
to collect information without damaging the cell. This can
be achieved with micropipettes. A micropipette as a microelectrode to record membranous voltage in muscle cells
was rst used by Graham and Gerard in 1946 (2).
GLASS MICROELECTRODES (MICROPIPETTES)

Micropipettes are made of very thin glass tubes lled with a
conducting electrolytic solution. A 1 mm diameter borosilicate or aluminosilicate glass tube is placed in a pipette
puller. The two ends of the glass tube are clamped to a
pulling device, and a heating lament is placed in the middle of the glass [Fig. 2(a)]. As the heating lament melts the
glass, the ends of the tube are pulled apart until separation
takes place, forming two electrodes with very small diameter (<0.2 m) [see Fig. 2(b)]. The length of the electrodes
shank is controlled by cooling the glass with an air puff following heating (3). The tip diameter and the taper of the
electrode are also controlled to make various types of microelectrodes. Intracellular potential is recorded with submicron diameter electrodes lled with an electrolyte similar to that found within the cell. KCl is often used because
the intracellular potassium concentration is high with respect to other ions. To facilitate lling the electrode, a small
glass capillary is inserted in the tube by the manufacturer
(Fig. 2). The other end of the electrode is inserted into an
electrode holder which contains a reference electrode made
of sintered Ag/AgCl. Then the electrode is connected to the
input of a high input impedance amplier. The potential
inside the cell is measured with respect to the voltage outside the cell using a reference electrode located in the extracellular space, also made of Ag/AgCl (see Fig. 1). Microelectrodes are also used to measure the voltage outside the
cell (extracellular recording). The diameter of extracellular
electrodes is larger than intracellular electrodes (around
1 m or 2 m), and the electrodes are lled with NaCl
because the sodium chloride concentration is signicantly
higher outside than inside the cell. A third type of microelectrode, the patch clamp electrode, is made to measure
the current through the membrane and through ion channels (4). These electrodes are also pulled to a diameter of 1
m to 2 m but have a steep taper generated by pulling the
electrode in a two-step procedure. Then the pipette is lled
and placed on the surface of the cell. The glass forms a high
impedance seal (G) with the membrane. Then the patch
of membrane inside the electrode is removed by applying
a small amount of suction, and direct access to the cell is
provided for voltage or current measurement (whole-cell
patch clamp). By pulling the electrode and patch from the
cell, an insideoutside patch is formed (cytoplasmic surface
toward the bath). By pulling the electrode slowly away from
the cell, an outside-out patch can sometimes form (extracellular surface toward the bath). The membranous current
from these various combinations is obtained by clamping
Micropipettes
Ag/AgCl
Pellet
KC
Ve
ic
ro
mV
Extracellular
medium
p
pi
te
Ve
Reference electrode
(Ag/AgCl)
et
Figure 1. Recording neural activity with a microelectrode. (a)

Neuron and intracellular electrode. (b) Intracellular voltage obtained with electrode showing the resting potential and the action potential.
70
Nichrome wire
F
Glass capillary
tube
I
(a)
Shaft
Shank
ms
Tip
(b)
Metal
Glass
(c)
Figure 2. Microelectrodes. (a) Micropipette puller. A glass capillary tubing is heated and pulled. (b) Two small diameter electrodes
are produced (c) Supported metal microelectrode with glass insulation.
the voltage at various values and measuring the current.

The impedance of these micropipettes signicantly inuences the measurement of both voltage and current. Intracellular electrodes, in particular, have high impedance
and capacitance and require special circuits for signal processing (see later).
METAL MICROELECTRODES
Although micropipettes can be made very small, they have
large impedance and are also very fragile. Microelectrodes
for recording the activity of small neurons can also be
made with sharpened and insulated metal wires. These

electrodes have very small tips and can be placed within
a few micrometers of the cell to be studied. Moreover, they
can be stiff enough to penetrate tough tissue, such as the
protective membrane around the brain or nerves. These
electrodes can record neural activity but also provide very
localized stimulation by passing current large enough to
excite the neurons in the vicinity of the electrode (5).
The most common type of electrode uses tungsten etched
to a sharp tip and insulated to within a few micrometers of
the tip. These electrodes are shaped with a shallow taper
to ease penetration. A more recent design uses iridium because it is stiffer than tungsten, is extremely resistant to
corrosion, and can be electrochemically activated (6). In
the recording mode, these metal electrodes operate at small
voltages, and the impedance between the metal and the tissue is dominated by the double-layer capacitance. Therefore, the electrical model of the electrode consists of a small
capacitance in series with the access resistance and the tissue resistance (see next section). The impedance is determined mainly by the size of the tip and can be in the M
range. The noise generated by the electrode is typical thermal noise but does not come entirely from the resistance
of the electrode. The electrochemical process, which generates the double-layer capacitance, involves movement of
charge carriers and contributes to the noise (7). Therefore,
low-noise recording ampliers must be used to maximize
the signal-to-noise ratio.
Metal microelectrodes can also be used to stimulate neuronal activity. Electrons are charge carriers in the metal,
but in the bodys aqueous solution the current is carried by
ions. A conversion must be made at the interface, which
necessarily involves electrochemical reactions. Some of
these reactions are reversible, but others are not. The irreversible reactions cause pH changes, gas formation, metal
dissolution, and corrosion (8). All of these by-products are
potentially damaging to the tissue but can be minimized
by using biphasic current pulses which keep the voltage
Micropipettes
across the interface below threshold values for these reactions and at least partially reverse some of the reactions.
For microelectrodes designed to stimulate neural tissue,
the charge-carrying capacity is the most important characteristic. This capacity is proportional to the surface area
of the electrode. To increase the amount of charge which
can be delivered by a small electrode, a layer of metal oxide (iridium oxide for an iridium electrode) is deposited
on the surface. The oxide layer formation or activation is
achieved by cycling the electrode between the anodic and
cathodic voltages which generate electrolysis. Iridium oxide is a conductive layer which exchanges an electron for
an hydroxyl ion across the interface. The charge-carrying
capacity of the electrode is effectively increased by adding
an iridium oxide layer. The impedance of the electrodes
is also reduced by this activation process by an order of
magnitude (6). The impedance decreases as a function of
frequency and is best modeled by a capacitor in series with
a resistor which has a 300 Hz cutoff.
Metal electrodes are insulated with various types of
polymeric materials, such Teon or Parylene-C. Insulation
at the tip of the electrode is removed with electric elds or
lasers to burn the material away. By combining the properties of glass and metals, supported metal electrodes are
fabricated [Fig. 2(c)]. A micropipette is lled with a metal
which has a melting point below that of glass and is pulled
to a ne tip. Thin metal lms can also be deposited onto a
solid glass rod pulled to a sharp point. Then the electrode is
insulated with a polymer, except at the tip, to form a sharp
microelectrode (9).
ELECTRICAL PROPERTIES OF MICROELECTRODES
Micropipettes and metals microelectrodes consist of a conductive cylindrical core surrounded by an insulating layer
made of polymer or glass. The electrode is inserted within
a tissue which is a relatively good conductor (about 100
cm). The electrode is pulled to a very small diameter, and
therefore the narrow shank region of the electrode generates large resistance. This region is usually tapered, and
assuming a small amount of taper, the resistance is given
by
where is the specic resistivity of the electrodes conducting material, L is the length of the electrode shank, and d
is the internal diameter of the electrode. A distributed capacitance is also formed along the immersed part of the
shank between the interior and the extracellular (or intracellular) space. This capacitance, again assuming a small
amount of taper, is given by
where D is the outside diameter of the electrode, l is the

length of the shank inserted into the tissue, and is the
dielectric constant of the insulating material. There are
several impedances which affect the signal, and detailed
Re
Cm
Rm
Ce
Vm +
Vhc
+
(a)
Rs
Cs
Ri
Vi
Ce
Vhc
+
(b)
Figure 3. Microelectrode impedance. (a) Equivalent circuit for

micropipette. (b) Equivalent circuit for metal microelectrode.
equivalent circuits of these microelectrodes have been presented (10).

For micropipettes, the impedance characteristic is dominated by the resistance RE (about 100 M for an intracellular microelectrode) and the small capacitance CE (a
few pF) of the shank. The cell is modeled by an equivalent
membranous resistance Rm and capacitance Cm in series
with a voltage source Vm . A simplied equivalent circuit
is shown in Fig. 3(a). The resistance of the cells Rm is often as large as the source resistance of the electrode. This
can cause saturation of the amplier when current is injected within the cell. Special circuits have been designed
to remove this effect (see bridge circuit in next section).
For metal microelectrodes, the impedance is dominated
by the metalliquid capacitance Cs and the series resistance Rs . For extracellular recordings with a metal microelectrode, the source is modeled by a voltage generator Vi
with a source resistance Ri . A simplied equivalent circuit
is shown in Fig. 3(b). In both cases, there is a dc battery Vhc
added to the equivalent circuit which takes into account all
of the half-cell potentials throughout the circuit.
ELECTRONIC CIRCUITS FOR MICROPIPETTES

The high resistance and capacitance of micropipettes can
severely affect the signals recorded and require special circuits for processing (11). In particular, most researchers
inject current inside through the cell to measure its input
resistance or to depolarize the membrane. To inject current and record voltages with the same electrode, a circuit
is used similar to that shown in Fig. 4(a). The membranous voltage is measured with the high impedance buffer
amplier (A). The injected current Ii is given by the following equation assuming that the input impedance of A is
Micropipettes
innite,
Vs
Ii
+
Rs
where B is a nite gain summing amplier. If AB is set

equal to 1, the circuit becomes a current source because
the injected current equals BVs /Rs and depends only on
Vs and Rs . A is implemented as a high-impedance buffer
and B as a summing amplier with a gain of 1. The input
resistance Rin of the circuit is obtained by considering that
the resistance Rs is bootstrapped between the input and
the output. Then the Miller theorem gives the resistance
Rm as
Vm
Is
Vm
A
Re
Electrode
Rm
Cm
Vr
(a)
The product AB can be designed to be less than but very

close to 1 thereby generating a very high input resistance.
Then this circuit allows voltage recording with high input
impedance and current injection with a grounded voltage
source.
The high resistance of the electrode generates a large
voltage in response to an applied current which is added
to the cells response. To remove the effect of the electrode
resistance from the response of the cell, a bridge circuit
shown in Fig. 4(b) is implemented. The circuit takes advantage of the fact that the electrodes time constant is
signicantly faster (in most cases) than the response of
the cell to a step increase in current. A current generator
injects the same current into a variable resistor that is injected in the cell. The voltage recorded by the electrode is
subtracted from the voltage across the variable resistance.
By varying the resistance, the effect of the electrode resistance can be removed from the recorded voltage. When the
recorded signal is properly compensated for, then the resistance of the electrode is equal to the value of the variable
resistance. Therefore, this simple circuit can compensate
for the electrode resistance and allows direct measurement
of the electrode resistance.
The bridge circuit depends on the relative speed of the
rise time of the amplier and the membranous voltage of
the cell. The capacitance of the micropipette can be large
enough to slow down the response of the amplier until it is
no longer possible to distinguish between the cell response
and the electrode response. Therefore, a circuit compensating for the capacitance of the electrode has been developed.
The circuit [Fig. 4(c)] introduces a negative capacitance at
the amplifer input. The input impedance of the circuit, assuming an amplier with an innite input resistance is
given by
where CF is the feedback capacitor. Therefore, the input

impedance of the circuit is the impedance of capacitor Ceq
given by
If the product AAc is adjusted to a value greater than 1, the

input capacitance becomes negative and is placed in paral-
+
Is
Re
Electrode
Is
Rb
Cm
Rm
Vr
Cell
(b)
CF
Vo
Vi
+
A
+
Ac
Ceq
R2
R1
(c)
Figure 4. Circuits for microelectrodes. (a) Current injection and
voltage measurement. (b) Bridge circuit for electrode resistance
measurement and compensation. (c) Negative capacitance compensation.
lel with the capacitance of the electrode. Then the net capacitance is decreased, improving the frequency response
and the rise time of the electrode-amplier recording system (12). To implement the circuit, the amplier (A) is a
buffer and is the same as in Fig. 4(a). The gain Ac amplier
is made variable and greater than 1.
Microelectrodes have also been used for voltage and
patch clamping of cells. These techniques allow researchers
Micropipettes
to x the voltage and measure the membraneous current of

a whole cell, a patch of membrane, or individual channels.
The methods required specialized circuits which measure
a current while maintaining the voltage of the membrane
(4).
ION-SELECTIVE GLASS MICROELECTRODES

By adding a membrane material to a glass micropipette,
the activity of ions in sample solutions can be measured.
The selectivity depends mainly on the type of membrane
used. The pH of solutions can be measured with a glass
membrane, and recent developments in polymer membranes have made it possible to measure the activities
of Na+ , K+ , Ca2+ , Cl , and other ions. These ion-selective
microelectrodes are inexpensive and are small enough to
measure the activity of ions in both extracellular and intracellular spaces. Abnormal cellular states often result
from an imbalance of ionic concentration. Therefore, the
fact the activity of ions (not just the concentration) can be
measured with very small electrodes compatible with glass
micropipette technology has provided researchers with a
powerful method to analyze both the normal and diseased
physiological states.
A membrane is chosen to allow the diffusion of certain
ions selectively and then is inserted into the glass micropipette. The potential difference generated between an
external reference electrode (Ag/AgCl, for example) and the
ion-selective microelectrode immersed in the sample solution is given by the Nicolsky equation (13):
where Vr is a constant and takes into account the voltages

within the electrochemical cell, ai is the activity of the ion
in the sample, aj is the activity of the interferention j, kij
is the selectivity constant of the electrode for ion j relative to ion i, x the charge of ion i, R is the gas constant,
T is the absolute temperature, z is the charge of the ion,
and F is the Faraday constant. Typically, electrochemical
cells have very large impedances, and the voltage difference must be measured with high input resistance ampliers (1015 ), such as electrometers. The selective membranes can be grouped into three classes: glass, liquid, and
solid state. Glass membranes are very sensitive to the concentration of H+ ions. Liquid membranes are commonly
used in biological preparations for other ions. For example, potassium ion activity can be measured using a liquid membrane containing the antibiotic Valinomycin combined with a resin. When applied to the extracellular space
of brain tissue undergoing epileptiform activity, the potassium concentration observed rises from its resting value
of 7 mM to 11 mM. The rise in potassium activity is accompanied by increased neuronal activity and is similar to
neural activity observed during epilepsy. The last class of
membranes is made of solid-state material, such as crystals or insoluble salts. For example, a pellet of silver sulde
can be used to detect Ag+ with very high sensitivity.
MICROELECTRODES FOR CHEMICAL MEASUREMENTS

The small size of microelectrodes makes them ideal for
measuring local concentrations of various chemical species,
such as neurotransmitters, or the partial pressure of various gases (see Biosensors). The carbon ber microelectrode is very popular. These bers have diameters as
small as a few microns and can be inserted into glass micropipettes for insulation by exposing a short length of
the electrode. Contact between the ber and the electronic
circuit is made by lling the electrode with mercury or
silver paint and sealing it with epoxy. Then the assembly is inserted within physiological tissue and selective
measurement of various compounds is carried out using a
combination of selective membranes, enzyme coatings, surface modications, and various electrochemical techniques.
For example chronoamperometry has been used to detect the release of easily oxidized neurotransmitters, such
as dopamine, serotonin, and epinephrine. In chronoampersmetry, a constant voltage is applied between the carbon ber and a ground electrode. The current is measured
and is directly related to the level of oxidation. Carbon
ber microdisks are typically more useful than microcylinders since they provide a better geometry and excellent
resolution. They are also more difcult to fabricate and
have lower current amplitudes. A three-electrode system
is used to generate a constant voltage between the working electrode and the medium around it. A reference electrode is located close to the working electrode to estimate
the mediums voltage. Using feedback, a current is applied
between the working electrode and a ground electrode to
maintain a constant voltage. For the very low currents of
microelectrodes and microdisks, a two electrode system is
sufcient since the ohmic drop and polarization of the reference electrode are negligeable. The amplitude of the current is modulated by oxidation of the compound to be measured and is proportional to its activity. To improve selectivity, a differential method is used whereby the current
amplitudes obtained at two voltages are subtracted and
the contribution of the interfering species, such as ascorbic acid, can be reduced or eliminated. The electrode can
also be coated with selective membranes. Naon is a commonly used membrane because it prevents various charged
molecules from interfering with measurement. The surface
of the carbon ber can also be modied chemically or with
a laser to improve the selectivity and sensitivity of the electrode (14). By depositing appropriate enzymes on the surface of the electrode, very selective electrodes can be made
to measure glucose, nitric oxide, acetylcholine, etc.
Field effect transistors (FET) have also been adapted to
allow measuring various ions or chemicals. Ion-sensitive
eld-effect transistors (ISFET), for example, are enhanced
MOSFET transistors but use an ion-sensitive membrane
instead of gate metallization (14). Then the transistor is
immersed in a solution containing the ions to be measured.
An electrochemical potential is established at the interface
between the solution and the gate dielectric. This potential
is established with respect to a reference electrode located
in the solution and can modulate the conductance of the
channel under the gate. The electrodes can be made very
small but have drift and selectivity problems.
Micropipettes
MICROELECTRODE ARRAYS
The silicon technology used to make integrated circuits
can be adapted to manufacture arrays of microelectrodes.
The activity of single cells can be recorded with the micropipette technology discussed previously. Neuroscientists are now increasingly interested in recording simultaneously from a large number of cells. Moreover, by stimulating a large number of cells in the spinal cord or in
the brain selectively, it should be possible, in principle,
to restore motor function in paralyzed patients or vision
in blind patients, for example. Therefore, multiple arrays
of electrodes capable of recording or stimulating the nervous system are clearly important to understanding nervous system function and to designing neural prostheses.
Three silicon-based types of microelectrode arrays have
been developed: (1) A 1-D beam electrode where a thinlm platinum-iridium is deposited on a thin layer of silicon
substrate (15). This thin substrate provides a surprising
amount of exibility and can be utilized for the leads and
the electrode pads. (2) A 2-D array for recording the activity
of neurons grown in cultures and axons in nerves. A thin
lm microelectrode array is made of gold electrodes covered
with platinum black on a silicon substrate. The assembly
is built into the bottom of a neuron culture dish. Neurons
grow over these electrodes and make direct contact with
them (16). In another design, micromachining of a silicon
wafer generates a matrix of 64 square holes with a side
dimension of 90 m. Gold pads and leads are deposited
near each hole. Then the thin wafer is inserted between
the two sides of a severed nerve. As the axons grow inside
the holes, it is possible to record from a selectively small
groups of axons (17). (3) A 3-D array for cortical recording
and stimulation. The 1-D beam electrode discussed previously can be assembled to form three-dimensional arrays.
The longitudinal probes are inserted perpendicularly into a
silicon platform. The leads from each probe are transferred
to the silicon probe and are routed to a digital processing
unit (18). Current work also involves including low noise
amplication directly on the platform. Another implementation involving micromachining and etching techniques
was used to fabricate a 10 10 electrode array. One hundred conductive needlelike electrodes (80 m at the base
and 1.5 mm long) are micromachined on a 4.2 mm 4.2 mm
substrate (19). Aluminum pads are deposited on the other
side of the substrate and make contact with each needle
electrode. The tips of the electrodes are coated with gold or
platimum. A high-speed pneumatic device is used to place
the array into cortical tissue because the high density of
the electrodes makes insertion difcult. Then the microelectrode arrays are available for recording from a large
number of cortical sites.
NANO ELECTRODES AND NANO ARRAYS

Progress in nanotechnology has reduced the size of microelectrodes into the 1-100 nm range (20). These electrodes
have improved spatial and temporal resolution for imaging
and analytical measurements (20). Nano-wires arrays with
integrated electronic circuits have been able to record neu-
ral activity along axons with a resolution of 400 nm (21).

Electrodes as small as 1 nm have been fabricated (22). Concentrations as small as 1021 (23) have been measured with
these nano-electrodes thereby getting closer to achieving
single molecule detection.
ACKNOWLEDGMENTS
This chapter was prepared with the nancial support of
NSF grants IBN 93-19591 and INT 94-17206. I would also
like to thank Zhenxing Jin for his help with drawing the
circuit diagrams.
BIBLIOGRAPHY
1. E. R. Kandel, J. H. Schwartz, Principles of Neural Science, 3rd
ed., Norwalk, CT: Appleton and Lance, 1991.
2. J. Graham, R. W. Gerard, Membrane potentials and excitation
of impaled muscle cells, J. Cell. Comp. Physiol., 28: 99117,
1946.
3. K. T. Brown, D. G. Fleming, New microelectrode techniques
for intracellular work in small cells, Neuroscience, 2: 813827,
1977.
4. T. G. Smith, Jr., et al. (eds.), Voltage and Patch Clamping with
Microelectrodes, Bethesda, MD: Americal Physiological Society, 1985.
5. I. A. Silver, Microelectrodes in medicine, Philos. Trans. R. Soc.
Lond. Series B. Biol. Sci. 316: 161167, 1987.
6. G. E. Loeb, R. A. Peck, J. Martyniuk, Toward the ultimate
metal microelectrode, J. Neurosci. Meth., 63: 175183, 1995.
7. J. Millar, G. V. Williams, Ultra-low noise silver plated carbon
microelectrodes, J. Neurosci. Meth., 25: 5962, 1988.
8. D. M. Durand, Electrical stimulation of excitable tissue, in J.
D. Bronzino (ed.), Handbook of Biomedical Engineering, Boca
Raton: CRC Press, 1995, pp. 229251.
9. J. G. Webster (ed.), Medical Instrumentation, Application and
Design, 3rd. ed., New York: Wiley, 1998.
10. L. A. Geddes (ed.), Electrodes and the Measurement of Bioelectric Events, New York: Wiley-Interscience, 1972.
11. A. D. McClellan, Theoretical and practical considerations concerning an active current source for intracellular recording
and stimulation, Med. Biol. Eng. Comput., 19: 659661, 1981.
12. R. S. C. Cobbold, Transducers for Biomedical Measurements,
Principles and Applications, New York: Wiley, 1974.
13. M. E. Meyerhoff, W. N. Opdycke, Ion-selective electrodes, Adv.
Anal. Chem., 25: 147, 1986.
14. D. G. Buerk, Biosensors, Theory and Applications, Lancaster,
PA: Technomic Publishing, 1993.
15. K. D. Wise, J. B. Angell, A. Starr, An integrated circuit approach to extracellular microelectrodes, IEEE Trans. Biomed.
Eng., 17: 238247, 1970.
16. W. Nish et al., Biosensors and Bioelectronics, 9: 734741, 1994.
17. G. T. A. Kovacs et al., Silicon-substrate microelectrode arrays
for parallel recording of neural activity in peripheral and cranial nerves, IEEE Trans. Biomed. Eng., 41: 567577, 1994.
18. A. C. Hoogerwerf, K. D. Wise, A three-dimensional microelectrode array for chronic neural recording, IEEE Trans. Biomed.
Eng., 41: 11361146, 1994.
Micropipettes
19. K. E. Jones, P. K. Campbell, R. A. Normann, A glass/silicon
composite intracortical electrode array, Ann. Biomed. Eng., 20:
423437, 1992.
20. D. W. Arrigan. Nanoelectrodes, nanoelectrode arrays and their
applications. Analyst. 2004 Dec; 129(12): 115765. Epub 2004
Nov 9. Review.
21. F. Patolsky, B. P. Timko, G. Yu, Y. Fang, A. B. Greytak, G. Zheng,
C. M. Lieber. Detection, stimulation, and inhibition of neuronal signals with high-density nanowire transistor arrays.
Science. 2006 Aug 25; 313(5790): 11004.
22. R. M. Penner, M. J. Heben, T. L. Longin, and N. S. Lewis. Fabrication and Use of Nanometer-Sized Electrodes in Electrochemistry, Science. 1990 Nov 23; 250: 11181121
23. J. J. Watkins, J. Chen, H. S. White, H. D. Abruna, E. Maisonhaute, C. Amatore. Zeptomole voltammetric detection and
electron-transfer rate measurements using platinum electrodes of nanometer dimensions. Anal Chem. 2003 Aug 15;
75(16): 396271.
DOMINIQUE M. DURAND
Case Western Reserve
University, Cleveland, OH
PACEMAKERS
Today, implantable cardiac pacemakers are used with exceptional success as a long-term, safe, and reliable form of therapy for different kinds of cardiac rhythm disturbances. Numerous pioneering, technological developments, such as
highly integrated circuits or lithium batteries, are significant
milestones in this field. During the short history of the pacemaker, beginning with the first implantation in 1958, a turbulent technical evolution has occurred. The early devices
were heavy, simple impulse generators with two transistors
and short operating lifetimes. These could pace only the right
ventricle asynchronously with impulses of a constant rate and
amplitude. Modern, rate-adaptive dual-chamber pacemakers
possess highly complex integrated circuits with several hundred thousand transistors. They can pace in the right atrium
and the ventricle, monitor the intrinsic cardiac activity, adapt
automatically to changing needs of the heart, be programmed
through inductive telemetry in a variety of ways, and guarantee an operating lifetime of 8 years or longer. These devices
now weigh approximately 25 g and are getting increasingly
smaller, given continued advances in electronics and battery
technology. Multidisciplinary cooperation between the diverse
fields of physiology, physics, electrochemistry, electronics, and
materials science have made these impressive developments
possible. Consequently, the goal of pacemaker therapy has exceeded that of merely a life-maintaining function. It has instead become more adept at the reestablishment of a high
quality of life for the patient.
Through different pacing functions, modern, rate-adaptive,
dual-chamber pacemakers can adjust to an optimal mode of
pacing and pacing rate at any time. Thus, interference between the pacemaker stimulus and the cardiac intrinsic rate
is prevented through continuous intracardiac electrogram
(IEGM) monitoring. If the intrinsic activity is absent for a
certain anticipated interval, the pacemaker commences pacing to prevent a sinking heart rate or cardiac arrest. Today,
all single- and dual-chamber pacemakers are designed to pace
on demand. The ever increasingly used, rate-adaptive pacemakers can control the stimulation rate with a physiological
sensor if a sufficient sinus rhythm is no longer present. A rate
increase is additionally accompanied by a dynamic shortening
of the atrioventricular (AV) delay to guarantee atrial and ventricular synchrony. This allows for the best possible pumping
performance at any time. In addition, modern pacemakers record large quantities of diagnostic information in an internal
memory over long periods of time. At follow-up, the physician
can use the programmer to analyze data that greatly contribute to therapy optimization. Rhythm disturbances requiring
treatment and the basic, technical solutions in a modern cardiac pacemaker will be described to augment understanding
of the versatile functions of the pacemaker.
heart via the parasympathetic and sympathetic branches.

Both branches have an opposite effect on the pumping efficiency. The sympathetic branch increases the cardiac output,
and the parasympathetic branch inhibits it. Under resting
conditions, the volume pumped per heartbeat, or the stroke
volume, is about 70 mL on average. At a resting heart rate of
70 beats per minute (bpm), the cardiac output is about 5 L/
min. Under heavy exercise, this value can rise up to 25 L/min
in trained athletes. To increase the cardiac output, both factors are increased simultaneously, but to different degrees.
Even though the stroke volume can increase only moderately,
by 50% at the most, the heart rate rises under heavy load
conditions to 180 bpm or higher. A simple way to estimate the
maximum heart rate (in beats per minute) for an individual is
done by subtracting the age in years from 220. For example,
the maximum heart rate for a 40 year old woman would be
180 bpm. The following four, qualitatively different effects influence heart rhythm:
THE HEART RHYTHM

The autonomic nervous system (ANS) mainly controls the adaptation processes and innervates different regions of the
1. Chronotropypertains to the heart rate. Heightened

sympathetic activity increases the heart rate; this is
termed a positive chronotropic effect. A heightened
parasympathetic tone can reduce the heart rate, even
lowering it to a cardiac arrest. This is referred to as a
negative chronotropic effect.
2. Inotropypertains to contractility. The sympathetic
transmitter substances that ligate to specific receptors
in the myocardial cells effect different changes in the
cellular ion balance, especially in increasing the amount
of released calcium. Called positive inotropy, this intensifies the electromechanical coupling, which is the conversion of ionic balance equilibria to muscle contraction.
With the release of more calcium, the contraction velocity increases, leading to a higher pressure rise velocity
in both ventricles and to the maximum contractile force
attainable. Although the parasympathetic innervation
of the ventricular myocardium is physiologically less developed, a negative inotropic effect can nevertheless be
detected. This acts both through an increase in parasympathetic and a decrease in sympathetic activity (1).
3. Dromotropypertains to the AV conduction time.
Through sympathetic influence, the excitation conduction in the AV node is increased (i.e., the time between
the atrial and ventricular contractions is shortened),
this being a positive dromotropic effect. Parasympathetic activation slows conduction, a negative dromotropic effect. In extreme cases, this can lead to a temporary total AV block. The AV conduction time is
shortened with an increase in heart rate. Normally, the
interval at 70 bpm is 190 ms and at 130 bpm is 120 ms.
Because the rate increase is largely attained by shortening the diastolic interval, shortening the AV delay is
key to maintaining synchrony between the atria and the
ventricles. The delay is important for an optimum ventricular filling, also at higher rates.
4. Bathmotropypertains to the excitability. Through
sympathetic activation, the pacing threshold of the
483
484
PACEMAKERS
Impaired
conduction
Impaired
pacemaking
Normotopic
Figure 1. Schematic classification of

heart rhythm disturbances, which can result either from pacemaking or conduction disorders. For each of the depicted
rhythm disturbance subclasses, specialized forms of electrotherapy have been developed.
Sinus tachycardia
Sinus bradycardia
Sinus arrhythmia
myocardial cells is lowered, so the cells become more

excitable. This is a positive bathmotropic effect. Parasympathetic influence has an opposite effect by raising
the threshold potential for the cells and reducing excitability. This is called a negative bathmotropic effect.
However, bathmotropy plays only a minor role in cardiac pumping.
Rhythm Disturbances
The prerequisite for maintaining homeostasis, which applies
to keeping all bodily functions constant, is the coordinated
interaction of all local autonomic and neural regulation mechanisms of the heart and circulation. Through illness or congenital anomalies, the natural cardiac pacing and conduction
mechanisms can be disturbed in different ways or be completely absent (Fig. 1). Without treatment, these disturbances
can lead to either death or to a permanent or temporary blood
flow insufficiency.
A pacemaking disturbance is present, for example, when
excitation still exits the sinus node (normotopic pacemaker),
but the sinus rate is too low (bradycardia), too high (tachycardia), or irregular, as with an arrhythmia. Ectopic rhythms
also belong in this category. This applies to a condition with
sinus node failure or a conduction disturbance (e.g., AV
block), where the remaining parts of the conduction system
are capable of spontaneously pacing but have lower intrinsic
rates the farther away they are from the sinus node. This
phenomena is referred to as ectopic pacing, because excitation
originates from an abnormal site. Especially dangerous members of this group are the actively occurring rhythms, such as
those independent of the sinus node function. These include
ventricular flutter and ventricular fibrillation, which ensues
at an even higher rate. In the latter case, coordinated filling
and emptying of the heart are no longer possible. Ventricular
fibrillation is the most common cause of death in electrical
accidents and is responsible for some fraction of cardiac infarctions. These states can be interrupted only by electrical
defibrillation.
A conduction disturbance can occur between the sinus
node and the sections following it (SA Block), in the region of
the AV node (AV block), or even within the ventricle. To fur-
Ectopic
Sinoatrial
Atrioventricular
Passive
Active
AV escape rhythm
Ventricular
escape rhythm
Extrasystoles
Ventricular
fibrillation
Ventricular flutter
Ventricular
tachycardia
Intraventricular
ther subdivide these, a medical differentiation is made between conduction disturbances of the first degree (delayed
conduction), the second degree (occasional interruption), and
the third degree (total interruption). A first-degree AV block
is diagnosed if the AV conduction time, the PQ interval on
the ECG, is greater than 210 ms. For second-degree AV block,
two types exist. In type I (Wenckebach or Mobitz I block), the
PQ interval is extended with each heartbeat, until a ventricular contraction is missing. In type II (Mobitz II block), the PQ
interval is constant as a rule, whereby individual atrial activity is either irregularly or regularly not continued on to the
ventricle (e.g., in a 2 : 1, 3 : 1, 4 : 1 rhythm). A total block of
conduction is called third-degree AV block. Here, a ventricular substitute rhythm is normally formed within a few seconds below the block that has a much lower rate, but it secures patient survival. The situation referred to as an
AdamsStokes attack occurs when there is a temporarily diminished, cerebral circulation resulting from an acutely occurring rhythm disturbance, as a corollary to some of the conditions described previously. Depending on the type and
length of the rhythm disturbance, symptoms of failure include
intermittent dizziness, loss of consciousness, cramps, and arrested breathing. If no substitute rhythm has formed after a
3 to 4 min arrest, even death can occur. Because of the high
level of technology attained in modern pacemakers, an artificial cardiac pacemaker can treat practically every kind of
bradycardic rhythm disturbance today. Also, in another category of electronic therapeutic devices, the implantable defibrillators are increasingly able to treat tachycardic rhythm
disturbances. Thus, rectifying a slow and a fast heart rate is
now achievable with state-of-the-art electronic technology.
THE DEVELOPMENT OF PACEMAKER THERAPY
The roots of contemporary pacemaker therapy can be traced
back to the eighteenth century. In 1791, Luigi Galvani documented, for the first time, experiments in which the heart and
muscle tissue were electrically stimulated. In the nineteenth
century, additional experiments, conducted by Bichat in 1800,
electrically stimulated animal hearts or decapitated humans.
The syndrome of bradycardia and syncope, which would be
PACEMAKERS
later named after its discoverers, was described by Adams in

1827 and by Stokes in 1846.
Hyman, who coined the term pacemaker, constructed the
first transportable, artificial pacemaker in 1932. The device
consisted of an electromechanical construction that had to be
carried on the patients back. It contained a mechanical clockwork as its energy source. In 1952, Zoll treated Adams
Stokes attacks with external stimulation through plate electrodes. The treatment could be performed only in emergency
situations in the clinic because of numerous difficulties
(burns, pain, skeletal muscle stimulation). Furman and Robinson were able to solve these problems later with an endocardiac electrode that still had to be attached to an external
pacemaker. Only the availability of semiconductor components made the construction of the first implantable pacemaker possible. Developed by Elmqvist and implanted by
Senning in 1958, this grandfather of all contemporary, implantable pacemakers possessed an externally rechargeable
battery and consisted of a simple impulse generator with two
transistors, delivering impulses at a constant rate of 70 bpm.
The pulse rate was determined solely by the time constant in
the RC network. Detecting intrinsic cardiac activity was not
yet possible. The asynchronous ventricular pacing provided
by these devices, however, could not rule out the interference
between the pacemaker and the intrinsic rhythm.
The next logical step was taken in the 1960s through the
introduction of R-wave synchronizing or inhibiting. Introduced by Castellano in 1964, the demand pacemaker had an
additional input amplifier and could then detect cardiac intrinsic events with the ECG measured by the electrode. If intrinsic activity was detected, it was possible to inhibit impulses. Delivering pacemaker impulses during the vulnerable
phase was avoided, and pacemaker-induced arrhythmias
were largely eliminated. This generation was constructed
with discrete transistors, thus attaining only limited functionality. Programmable pacemakers were only possible in the
early 1970s with the advent of integrated circuits, which had
significantly expanded functions. These systems allowed different stimulation parameters to be adjusted postoperatively,
greatly improving the individual patient treatment throughout the operation time of the device. At about the same time,
it was possible to extend the pacemaker operating lifetime
with lithium batteries to such an extent that it lasted as long
as the life expectancy of most patients.
Physiological pacemakers were developed in parallel with
the continued development of single-chamber pacemakers.
Ventricular pacing through the depolarization signal of the
atrium was a decisive step in improving electrotherapy. With
485
this, in addition to eradicating bradycardic symptoms, atrial

and ventricular contractions could be synchronized. For patients with AV block and intact sinus node function, adapting
the cardiac output under physical stress was possible. This
was a decisive step in increasing the quality of life. The dualchamber pacemaker with atrial and ventricular demand pacing and P-wave synchronous ventricular pacing characterized
the next breakthrough in re-creating the coordinated course
of contractions between the atrium and ventricle.
Rate-adaptive pacemakers are increasingly gaining significance in bradycardic rhythm disturbance therapy, in addition to re-creating the natural course of contraction. With
measurement of appropriate physiological values, these adaptive devices allow load-dependent adaptation of the pacing
rate in cases of sinoatrial excitation formation disturbances.
This enables cardiac output to be adaptable over a greater
range. In 1974, Camilli described the first experimental, rateadaptive pacemaker based on pH-levels in the blood. However, the first rate-adaptive pacemakers were not commercially available until 1983. The most commonly used principle
in the past, the motion sensor, is integrated in the pacemaker
(e.g., as a piezoelectric crystal) and is known for its simple
implementation and easy-to-comprehend functioning principles. The pacing rate determined in this fashion correlates
however rather imprecisely with actual metabolic demands.
For this reason, researchers are currently focusing their efforts on perfecting pacemaker technology in rate-adaptive
systems. These perfected models will determine metabolic
needs with higher specificity and ideally re-create the natural
feedback control loop for heart rate adaptation.
OPERATING MODES AND PACEMAKER CODES

To clearly mark different operating modes, an international,
uniform system of nomenclature was established (2). Table 1
shows this generic pacemaker code. It consists of letters symbolizing: the paced chamber (the first letter), the sensed
chamber (the second letter), the type of reaction to the sensed
signal (triggering or inhibiting, the third letter), programmable and rate-adaptive functions (the fourth letter), and antitachycardic functions (the fifth letter). The code is an extension of the three-letter code from 1974. Consequently, it is
still common to refer only to the first three letters of the basic
pacing principle. The earlier described single-chamber pacemaker that paces and senses in the ventricle only, is called
VVI. And if it includes rate-adaptive impulse delivery, it is
termed VVIR. The dual-chamber pacemaker that senses atri-
Table 1. The NASPE/BPEG Generic (NBG) Pacemaker Code (2)

Position
I
Chamber(s) Paced
II
Chamber(s) Sensed
III
Response to Sensing
IV
Programmability,
Rate Modulation
V
Antitachyarrhythmia
Function(s)
0 None
A Atrium
V Ventricle
D Dual (A V)
0 None
A Atrium
V Ventricle
D Dual (A V)
S Single (A or V)
0 None
T Triggered
I Inhibited
D Dual (T I)
S Single (A or V)
0 None
P Simple programmable
M Multiprogrammable
C Communicating
R Rate modulation
0 None
P Pacing (antitachyarrhythmia)
S Shock
D Dual (P S)
486
PACEMAKERS
Symptomatic bradycardia
No
Is AV
No conduction Yes
presently
appropriate?
DDDR
No
Is SA mode
Yes
function
presently
appropriate?
Are there atrial

arrhythmias?
Yes
Paroxysmal
SVT
Yes
Chronic atrial
fibrillation
VVIR
VVI
VDD
DDD
DDDR
DDDR
+ dual
demand
AAIR
DDDR
No
Is SA node
Yes
function
presently
appropriate?
No
Is AV
conduction
intact?
Yes
DDIR
AAI
DDD
AAIR
DDDR
VDD
Figure 2. Flow chart of pacing mode selection and its dependence on the rhythm disturbance.
Ventricular single-chamber pacing (VVI) should be applied only for chronic atrial fibrillation.
Otherwise, DDD(R) orif AV-conduction is still presentAAI(R) modes are indicated.
ally and uses the natural intrinsic activity to pace the ventricle, functions in the VAT mode. However, it is often necessary
to pace and sense in both chambers (DDD, DDDR). A flow
chart in Fig. 2 illustrates the selection of the most important
pacing modes given the profile of the disease. With the exception of chronic atrial fibrillation, which today still requires
implantation of single-chamber systems (VVI, VVIR), rateadaptive dual-chamber pacemakers are becoming more and
more the standard for providing therapy.
ELECTRODES
Equally important progress has been made in electrode technology in parallel with semiconductor technology. Through
new electrode technologies and coating methods, decisive
progress was achievable for better electrode-tissue interfacing. Improved ingrowth characteristics and a drastic reduction of the pacing threshold, and thus the energy consumption, are examples of these innovations. The pacemaker
impulse is conducted through a special electrode (or in the
case of a dual-chamber pacemaker, through two separate electrodes) to the atrial and/or ventricular myocardium to trigger
depolarization. To this end, principally only the chambers on
the right side are considered (right atrium or right ventricle).
Introducing the pacemaker electrode(s) is feasible only with
a low-pressure system approach. One such case would be a
transvenous entry, through the right half of the heart. Excitation of the tissue is done through the electrically conductive
tip of the electrode, which is formed or surface-treated for optimal charge transmission (3). The geometric (macroscopic)
electrode surface is designed to attain the highest current

density possible with limited impulse energy and reduced contact resistance of the fibrotic tissue that builds up immediately after the in-growth process around the electrode. Even
though the macroscopic surface is kept relatively small (typically 4 to 20 mm2), an attempt is made to increase the electrically active (microscopic) surface by depositing a porous structure so that losses through dynamic resistance and electrode
polarization are reduced. The different methods that have
been applied are sintering, deposition of gridlike structures,
sputter coating, and physical vapor deposition (PVD) coating.
Under controlled process conditions, it is possible to create
surfaces possessing a fractal structure through PVD coating
with TiN or IrN. These surfaces are larger than the geometric
surface by a factor of more than a 1000 (4).
For passive fixation, either a conical formed collar or
small tines made of silicone rubber, polyurethane, or another
lead-insulating material are located directly behind the electrode tip (Fig. 3). Because of the requirements for flexibility
and stability, the lead is principally constructed with a coil
design. At the other end of the electrode is a standardized
axial pin connector IS-1, which connects and mechanically
fixates the electrode to the pacemaker by a screw-in contact.
In addition to unipolar designs, bipolar electrodes exist with
an additional ring contact at a distance 15 to 30 mm away
from the electrode tip, which has a separate point of contact
through a separate lead. Unipolar pacing forms the negative
pole through the electrode tip (the stimulating or different
electrode) and the positive pole through the pacemaker housing (the indifferent electrode). However, bipolar electrodes
PACEMAKERS
Tipconnection
to inner conductor
Ringconnection
to outer conductor
Insulation
Helical coil
conductors
Tip electrode
(cathode)
Ring electrode
(anode)
Tines
Figure 3. Typical shape of a bipolar pacing electrode. It is connected

to the pacemaker by a standardized IS-1 pin connector (upper end);
two helical coil conductors (only one for unipolar leads) guarantee
sufficient mechanical flexibility. The electrode tip and tines show just
one possible shape; several other designs (e.g., screw-in leads for active fixation) are also available.
provide bipolar pacing, in which the ring electrode functions

as the indifferent electrode. Both configurations are also used
for sensing. A frequently applied configuration is the combination of bipolar sensing and unipolar pacing, which offers
the following advantages:
1. Avoids inhibition resulting from muscle potentials,
2. Results in less cross-talk by ventricular activity during
atrial sensing,
3. Improves the signal-to-noise ratio of sensing, and
4. Creates clearly visible pacing pulses on the surface
ECG.
BATTERY
In estimating the energy consumption of an implantable cardiac pacemaker, the following basic information is to be considered. In most cases, stimulation impulses are 5 to 19 mA
with 0.1 to 9.6 V and 0.1 to 2 ms impulse widths at a rate of
30 to 150 bpm. With a typical setting of 10 mA with 5 V and
0.5 ms at 70 bpm, the power consumption is 30 W (5). The
first implanted pacemaker from 1958 was equipped with a
rechargeable (secondary) NiCd battery with a 190 mAh capacity with 1.25 V. Charging was done inductively and took an
hour per week (6). This battery type was used only five times
in implantable pacemakers worldwide because the service
lifetime was significantly lower than that of primary (nonrechargeable) batteries. In addition, it was too risky to place
the responsibility of recharging the battery in the hands of
the patient. Instead, the zincmercury battery became the
standard energy source in implantable pacemakers from 1958
into the early 1970s (7). Normally, three to six cells were connected in series to achieve an operating voltage of 4 to 8 V.
The problem of hydrogen evolution from operating these devices was solved by using epoxy encapsulation of the pacemaker, which was gas permeable. Despite continual improvement until 1970, these cells had only a 2 year lifetime.
Although the capacity would theoretically suffice for 5 years,
487
high internal losses (10 to 20%) limited the actual service lifetime (5). Various other solution concepts such as biological,
piezoelectric, biogalvanic, and even nuclear batteries were researched as well. With the exception of the nuclear battery,
none of these methods went beyond the experimental stage.
Since 1972, different types of lithium batteries have been
used and in the meantime have come to be the standard solution. Although improved rechargeable generators have been
available since the early 1970s, this has not yet had an impact on present pacemaker application. This is because lithium batteries have enabled pacemaker operation for more
than 10 years because of their significantly higher energy
density and low internal energy dissipation. An additional advantage to the lithium battery is that of no gas evolution. This
allows pacemakers to be encapsulated largely by titanium for
a hermetic seal. Electronic circuit reliability has therefore
been significantly improved because circuit failures caused by
released hydrogen have been eliminated. Of the numerous
types developed (5), mainly the LiI battery (Fig. 4) is used for
pacemakers, which delivers a voltage of 2.8 V with capacities
of 0.8 to 3 Ah. The voltage falls gradually through discharging, resulting in a very flat curve the first year. However, increased discharging results in a much steeper curve. Concurrently, internal resistance increases greatly, from
approximately 100 with a new battery to approximately 40
k at the end of the service lifetime. This point in time is
usually expressed in two stages. If the voltage has attained a
value between 2.0 and 2.2 V [internal resistance then being
20 to 30 k (8)], the beginning of the elective replacement
interval (ERI) is indicated. ERI is defined as the interval between this stage of battery depletion and the end-of-service
(EOS) stage, where the voltage has further dropped down to
1.8 V and the internal resistance has increased to 40 k. This
additionally discharged state, where the device no longer
functions regularly, is reached about six months after start of
ERI, depending on the device and pacing mode. The pacemaker should be replaced during ERI and must be immediately replaced when the EOS criteria are met.
Glass feedthrough
Insulating tube
Collector
Lithium anode
Cathode
Stainless steel housing

Figure 4. Internal construction of a central lithium anode/casegrounded pacemaker battery. The precoated lithium anode, which is
surrounded by the cathode material, is connected to the negative pin
by a central collector and a glass feedthrough. The housing itself
serves as the cathode current collector.
488
PACEMAKERS
Depending on the pacemaker model, ERI is shown by a

defined decrease in the pacing frequency, (e.g., by 11% as opposed to the programmed rate with the device PHYSIOS TC,
Biotronik, Inc). Alternatively, ERI may be indicated by a variation in the magnet placement frequency, the rate at which
the pacemaker paces for a certain number of cycles, (e.g., 10),
when a magnet that is usually in the programming head is
placed over the pacemaker. Aside from this, all important
electrical parameters, including battery voltage and the battery internal resistance, can be interrogated in modern pacemakers through telemetry. The respective voltage threshold
values are not uniform, and they are determined by each
manufacturer.
BASIC FUNCTIONS OF THE DUAL-CHAMBER PACEMAKER

The multiprogrammability of the function and pacing parameters enables an optimized, individual form of electrotherapy
through postoperative, noninvasive pacemaker programming
for each indication. This function is designated by the M in
the pacemaker nomenclature code. Besides the basic rate and
pacing mode, other programmable parameters are the stimulation energy (i.e., the impulse amplitude and duration), sensing sensitivity, refractory period, and hysteresis setting. In
addition, diagnostic and electrophysiologic examinations can
be conducted postoperatively with these implants. The same
applies for temporary antitachycardia pacing, which is performed noninvasively with an external programming device
using a chosen algorithm.
With the programmability of the stimulus impulse parameter, energy can be effectively conserved with the possibility
of extending the operating lifetime. Thus ineffective pacing,
resulting for example from changing pacing thresholds, can
be controlled. The pacing threshold is the minimum pacing
energy needed to trigger the myocardial depolarization. It is
measured after pacemaker implantation and during follow-up
examinations by both the pacemaker and the programmer.
Pacing energy is adjusted via two parametersimpulse amplitude and duration. The adjustment is done to a value above
the pacing threshold by a defined safety margin. By adapting
sensitivity and choosing between unipolar and bipolar electrode configurations, extensive protection is afforded against
external and intrinsic disturbing signals. Rate and refractory
period programming optimizes the significant cardiologic parameters of the artificial pacing system according to therapy
requirements. It also suppresses depolarization phenomena,
crosstalk, and extrasystolic arrhythmias.
Above all, rate programming allows intrinsic heartbeats to
ensue in patients who need only occasional pacing. The hysteresis setting avoids pacemaker-mediated syncope resulting
from intermittent conduction at nearly identical rates of intrinsic and artificial pacing (9). The pacemaker engages only
when the natural sinus rhythm falls below a certain beat
rate, but then stimulates at a higher (adjustable) rate. Hysteresis programming also provides the advantages of compensating for missing atrial contribution by pacing at a higher
rate (with single-chamber pacing) and limiting pacemaker activity to only when it is absolutely necessary. With an intact
sinus node and the presence of conduction disturbances, Pwave-controlled ventricular pacing (VAT) can function like an
electronic bridge for the AV interruption. However, abnormal
excitation, such as premature supraventricular or retrogradely conducted ventricular extrasystoles, and ectopic ventricular rhythms, limits the wide application of atrial control.
Only through additional control functions for the ventricle
can pacemaker-mediated parasystoles be suppressed. The
dual-chamber pacemaker with an atrial and ventricular control circuit for pacing (DDD) solely guarantees physiological
atrial and ventricular pacing. This is accomplished by recognizing atrial, as well as ventricular, arrhythmias under
nearly all conditions of pacemaking and conduction disturbances (10). The case of sick sinus syndrome is an exception,
in that no physiologic rate-adaptation is possible under conditions of stress, unless a rate-adaptive pacemaker is implemented.
The technical challenge for re-creating the natural course
of excitation is that of synchronizing the ventricles with the
atria. The AV delay and the atrial and ventricular refractory
periods regulate the course of pacing. Adapting AV delays automatically to the heart rate optimizes the atrial hemodynamic contribution for filling the ventricle, which contributes
up to 30% to the cardiac output. By reducing the artificial
conduction time with sensed P waves, the latency period between the atrial stimulus and atrial excitation is additionally
compensated. Arrhythmias can be eliminated by selecting appropriate control intervals. These rhythmic disturbances may
result from ventricular extrasystoles or the abnormal spread
of excitation. For the majority of clinically significant rhythm
disturbances, this option allows for an automatic adaptation
of pacing to the therapy requirements (11). In the event of
extreme arrhythmias, programmability offers specific control
programs that interrupt supraventricular and ventricular reentry tachycardia by prematurely exciting the myocardium.
TECHNICAL SOLUTION
The block diagram of a multiprogrammable dual-chamber
pacemaker in Fig. 5 affords an overview of the basic components. Atrial and ventricular channels each possess separate
input and output amplifiers connected to the myocardium
through the atrial or the ventricular electrode. The program
memory digitally controls the adjustment of the impulse energy through the amplitude and the impulse duration of the
stimulus and the sensitivity of the input amplifier. The central, quartz-controlled clock and the counter keep the timing
of all control processes, such as pacing rate, refractory period,
hysteresis, and program transmission. Programming is performed inductively through a program amplifier with a decoder, a control circuit, and a register that stores the temporary and permanent programs.
Figure 6 shows the construction of an implant, in which
the two significant components are the lithium battery, occupying the right two-thirds of the housing, and the hybrid circuit. The pacemaker housing consists of a dual-sectioned titanium capsule, which is reliably sealed by laser welding and
has a long-term resistance to corrosion. The electrode connection block consists of a cast epoxy resin head with hermetically sealed feedthrough. The hybrid substrate contains all
electronic components, including a coil that enables bidirectional inductive telemetry for postoperative programming
through a programmer. Some applications for this programmer are an inductive transmission of pacing and function pa-
PACEMAKERS
Sensing
amplifier
A
Noise
detection
Refractory
period
PR
Battery
+
EOL
EOLindicator
PR
Sensing
amplifier
V
Noise
detection
Refractory
period
Main
control
logic
489
Output
amplifier
A
A
PR
Timer
marker
CLK
Event
counter
PR
External
trigger
Quartz
crystals
Oscillators
AV delay
RUN-AWAY
protection
Output
amplifier
V
Temporary
program
register
A/D
converter
Memory
(RAM)
8 KB 8
Reed
switch
Bus
control
Permanent
program
register
Decoder
Receiver
Coil
IEGM
Histogram
Analog data
Coder
Transmitter
Figure 5. Block diagram of the circuitry of a multiprogrammable dual-chamber pacemaker. It

shows the following major components: input and output stages for the atrial and ventricular
channels, main control logic, memory for internal data storage, and the communication unit for
bidirectional data transmission, which is activated by closing the reed switch with the magnet
in the programming head.
rameters or intracardiac signals and such operating parameters as battery current, voltage, and internal resistance;
electrode impedance; and patient information (12). The value
of programmability not only lies in the possibility of postoperative corrections, but allows also an effective reduction of the
variety of pacemaker types, lowering the overall costs involved.
Connector block
Setscrew
Fixation hole
Header
Feedthrough
O-ring
Lead ports
Lead
Housing
Sealing cap
Hybrid circuit
Welded seam
Battery
Figure 6. Major components of an assembled dual-chamber pacemaker. The hermetically sealed titanium housing contains the hybrid
(left), which holds the IC and all other electronic components, and
the battery (right), which occupies about two-thirds of the housing.
Leads are connected to the device via the cast epoxy resin head.
Technical Realization of Circuits

Because the implant often has a life-supporting function, high
demands are placed on the electronics in the pacemaker.
Some of these expectations are extreme reliability and operating safety, the smallest possible and lightest construction,
operation at low supply voltages of 2 V, and constancy of all
function parameters, even with falling battery voltage as a
result of discharging. These requirements are met by a host
of special technical solutions for circuits and highly developed
production processes.
Integrated Circuit. The biological demands for miniaturization and minimal energy consumption are best met through
the monolithic integration of as many components as possible
onto one chip. All analog and digital function blocks necessary
for standard pacemaking are integrated onto one chip using
low-voltage CMOS (complementary metallic oxide semiconductor) technology. The central control unit, the functional
units for pacing and sensing, telemetry, charge pump, program amplifier, oscillator, and reference voltage sources are
included in this group. The control signal is amplified with a
two-level bandpass with several ranges of sensitivity. The
pacing amplitude is created with the help of a charge pump
in combination with a programmable voltage multiplier and
is adjustable from 0.1 to 9.6 V. The programmability of closely
stepped settings and the large range for adjusting pacing amplitude and sensitivity with separate programmability for the
atrial and ventricular channel place great demands on the
semiconductor technology of the pacemaker circuit. A 2 m
silicon gate technology was used. Approximately 100,000
490
PACEMAKERS
Figure 7. Hybrid circuit of a rate-adaptive dual-chamber pacemaker (INOS, Biotronik, Inc.).

The back side (right) holds several discrete components such as quartz crystal, reed switch,
telemetry coil, capacitors, and resistors. The front side (center) contains two VLSI chips (pacemaker and monitoring chips) and carries a daughter board (left) with a third VLSI chip for the
rate-adaptation functions.
transistors were integrated on a surface with an area of 70

mm2. The operating voltage ranges from 1.8 to 3.0 V. Current
developments use technologies allowing structures between
0.3 and 0.8 m.
Hybrid Design. As in many other fields of hardware and
software development, modularity as a construction principle
plays a key role in pacemaker technology. The previously
noted complex functions of modern pacemaker systems are
realized by individual technical modules composed of circuits.
From here, a whole family of devices with different functions
emerges easily. These devices are each tailored to the individual therapeutic needs and provide optimal and cost-effective
therapy. The complexity of such circuits exceeds the present
technical possibilities of a monolithic solution. Thus, integrated and passive components are joined by a hybrid circuit.
Figure 7 shows the modular circuit construction of a completely assembled rate-adaptive dual-chamber pacemaker circuit using hybrid technology. The pacemaker circuit contains
three VLSI components. The pacemaker chip is made up of
the control logic, analog input and output amplifiers and telemetry; the monitoring chip is composed of the event counter
and the trend monitor (Fig. 7, center). The third VLSI-chip
mounted on a daughter board includes a microprocessor for

rate adaptation functions (Fig. 7, left). Passive components
including the telemetry coil, reed switch, and quartz crystal
can be seen in Fig. 7, right. The versatile pacemaker chip can
be applied with a dual-chamber system, as shown in the hybrid circuit, as well as in a single-chamber pacemaker. Likewise, the monitoring chip can be integrated into other systems. The chips are produced as application-specific
integrated circuits (ASIC) using low-voltage CMOS technology. The no-load current is approximately 7 A for the entire
dual-chamber pacemaker circuit. With atrial and ventricular
pacing at a rate of 60 bpm, approximately 30 A are taken
up from the lithium battery (2.0 Ah); that corresponds to an
operation time of 8 years.
As a form of interconnection technology, hybrid technology
plays a key role in producing electronics for implants. Its task
is to create the connection between the different integrated
components and the passive components. To meet the growing demands for reliability and packing density, production
technology for manufacturing modern multilayer hybrid circuits has consistently been developed further. Laminate technology with dielectric foils allows the necessary production
steps to be reduced. This is in contrast to thick film
PACEMAKERS
multilayer technology, in which conductor track levels, including the insulating dielectric layers, are produced by repeated printing and sintering. With fewer production steps
and by substituting screen printed for laminate dielectrics, a
higher level of reliability is reached. Additional advantages
are the possibilities to integrate passive components into the
layered structure and fabricate a three-dimensional substrate
by creating depressions. Both aspects lead to an increased
packing density and therefore additional miniaturization.
Reliability and Quality Assurance. Reliability in the discussed monolithic circuit and its hybrid construction has been
proven clinically in more than 600,000 implants, in which reliability can be attained with 107 /h at a 90% confidence
level for the hybrid circuit and 109 /h for ICs and passive
components. In pacemaker technology, this high reliability is
not attained by the normal methods of redundancy of critical
circuits and components. This is because operation time requirements can be fulfilled only via the current consumption
aspect, because of the volume-limited battery capacity.
Instead, it is necessary to research specifically failure
mechanisms and error sources and to eliminate them in the
development phase as a preventative measure. Above all, it
is important to provide consistent application of control measures during the different phases of production (13). This experience has led, for example, to special design rules during
dimensioning of the current mirror circuits. This is because
the design rules strongly influence the characteristics of the
ECG, as well as measuring amplifiers, VCOs, and reference
voltage sources. Especially critical production tolerance is
also identified in this manner. A parameter drift from an ionic
impurity in the semiconductor can be largely avoided by the
exact control of the process parameters during chip production (14).
Quality assurance measures for pacemaker production include a 100% final control of all components, semifinished
products, and the finished product. In general, the military
standards (MIL-STD-883 and MIL-M-38510) are used for implantable devices. In some cases, the specifications even exceed these standards. The requirements of the standards
(MIL-Q-9858 and MIL-STD-1772) must be met for the production of implantable semiconductors and hybrid circuits. All
aspects of quality control and assurance are included in these
measures. Standards set by the International Organization
for Standardization (ISO), the ISO 9000 standards, are gaining importance worldwide. Quality assurance measures for
the production process and measures for qualification and
type testing are contained in the ISO 9002 guidelines. The
ISO 9001 guidelines cover the realm of development, thereby
having an effect on the design process. In addition to strict
quality control during hybrid construction, it must be guaranteed that these high standards are being met in testing the
materials applied.
Input Amplifier. The input or sense amplifier must detect
the electrical intrinsic cardiac activity. That is, to amplify and
filter the IEGM as recorded by the electrode and to sense the
intrinsic activity (P waves in the atrium or R waves in the
ventricular signal). The central control unit (see Fig. 5) turns
off the amplifier during the pacing impulse (blanking). Protecting the amplifier and other components (e.g., the output
amplifier) during use of an external defibrillator is accom-
491
plished by discretely designed input protection diodes. The

amplifier circuit triggers a comparator, which senses via a
threshold value comparison. The actual electrogram contains
frequency components up to approximately 100 Hz. Higher
frequencies originate from interference sources (e.g., from the
muscle activity). To increase the selectivity, the input amplifier of the discussed dual-chamber pacemaker has a bandpass
consisting of a fourth-order high-pass (80 dB per decade) and
a second-order low-pass (40 dB per decade), and a center frequency of 70 Hz on the atrial and 40 Hz on the ventricular
channel. The input amplifier is constructed similarly to the
other analog circuits in SC (switched capacitor) technology
(5,15). This offers a number of advantages with regard to the
applicable semiconductor technology (e.g., CMOS) and improves the tolerances, especially for implants.
Control Logic. The digital control logic is housed on the
pacemaker chip with the analog components. It must coordinate all activities of the pacemaker (sensing and pacing in
one or two chambers) according to a predefined timing
scheme, so that the synchronization of the ventricles with the
atria is reestablished. The control logic is therefore realized
as binary automaton, meaning it can assume different states
depending on timing intervals and sensed signals. The course
of control is influenced by various timing intervals. Minimum
and maximum pacing rates are set by the lower rate interval
(LRI) and the upper rate interval (URI). The lower rate interval (also the basic interval or the automatic interval) is the
maximum time between two intrinsic impulses of the respective chamber during which no stimulus is triggered. The upper rate interval is the shortest interval between a sensed or
paced event and a (new) stimulus. The AV delay in modern
pacemakers varies according to two criteria. First, the AV interval is shorter after a sensed event than after a paced event
to compensate for the delay between the atrial stimulus and
the actual atrial depolarization (latency interval compensation). Second, the AV delay is shortened rate-dependently to
mimic natural dromotropy, guaranteeing an optimal synchronization during higher heart rates (dynamic AV delay). The
refractory periods for the atrial and ventricular channels
(ARP and VRP, atrial/ventricular refractory period) are programmable in their length and ignore sensed signals during
these periods. This prevents the QRS complex or the T wave
of the previous stimulus as well as afterpotentials appearing
after every stimulus from being misinterpreted as sensed
events and falsely inhibiting the following stimulus. Through
refractory period programming, certain physiological events
are blanked out, preventing these signals from being interpreted as intrinsic events. A relatively short blanking period prevents false detection of hardware activities (i.e., by the stimulus in the other chamber) (16).
Output Amplifier. The tissue is stimulated with an impulse
whose amplitude and pulse width are adjustable typically
from 0.1 V to 9.6 V and from 0.1 ms to 2 ms, respectively.
Both parameters are programmable for energy efficiency. The
stimulus must be sufficiently high to depolarize the tissue.
That is, to increase the membrane potential from its resting
value of 85 mV beyond the threshold potential to approximately 65 mV. This triggers an action potential, which
spreads throughout the entire myocardium. The necessary
impulse energy varies on a case-by-case basis, depending on
492
PACEMAKERS
2.0
Pulse amplitude (V)
the battery and electrode parameters, and a multitude of diagnostic data. This includes the number and histogram distribution of the paced and sensed events in each chamber that
were recorded during the previous follow-up interval. These
data provide the physician with valuable diagnostic information about the pacemaker function and changes in the patients symptomatology. The stored data can be roughly divided into four categoriespatient data, battery and
electrode parameters, pacing parameters, and diagnostic information.
Pacing threshold
1.5
1.0
2 VRh
0.5
Rheobase VRh
Patient Data
Chronaxie time
0
0.2
0.4
0.6
0.8
Pulse width (ms)
1.0
1.2
1.4
Figure 8. Chronaxie-rheobase curve showing the relationship between the pulse width and amplitude values necessary for tissue
stimulation. Only typical values are shown; in practice, individual
values have to be measured for each patient to determine adequate
pacing parameters.
the exact position of the electrode, the thickness of the fibrotic

tissue, the pacing threshold, and so on. By varying the impulse width and amplitude parameters, a chronaxierheobase
relationship results. The graphical representation of this
function (Fig. 8) connects pairs of values from both parameters that were determined experimentally for a certain pacing
threshold. When the pulse width is increased up to infinity,
the graph converges asymptotically toward the rheobase
value. Chronaxie is the impulse duration corresponding to
double the rheobase voltage. It can easily be observed that an
impulse duration equaling the chronaxie at an amplitude of
twice the rheobase is best from an energetic viewpoint. In
practice, there is a 100% factor of safety (doubled pacing energy). One of the two parameters is doubled, preferably the
impulse duration, because it can be adjusted by the circuit
more simply. Consequently, an impulse amplitude equaling
double the rheobase value and an impulse duration equaling
double the chronaxie are normally chosen. The impulse delivery is realized by an output stage (pace amplifier), in which
the capacitor that delivers the charge to the tissue (capacitive
coupling) is charged between two stimuli. This allows charge
neutrality to be reestablished by the subsequent repolarization process and minimizes irreversible chemical reactions.
During an impulse, a charge of 0.1 to 20 C is delivered. To
increase the integration density, often one pace amplifier is
used for both channels. Switching between atrial and ventricular electrodes is done by a multiplexer (5). Because the impulse amplitudes can be programmed up to a multiple of the
battery voltage, the devices possess a voltage multiplier or
charge pumps. These functions are also integrated onto the
pacemaker chip.
TELEMETRY
Analog telemetry allows the bidirectional data transfer between the pacemaker and an external programmer. This enables a later correction of the pacing parameters during follow-ups, as well as an interrogation of the data stored in the
internal memory of the pacemaker (e.g., patient-related data),
The patient data memory stores the most important information about the patient and the pacemaker. The date of implantation, the date of the last follow-up, information concerning the symptoms, etiology, and ECG indication, in the
form of code letters according to an international code list,
are some of these particulars. Others include the implanted
electrode configuration, the initials of the patient, and the serial number of the pacemaker. Thus, the most crucial information can be retrieved in the case of an emergency, an unexpected change of the physician, or the loss of the pacemaker
identification card.
Battery and Electrode Parameters
Precise, prevailing values of the battery voltage, the internal
resistance of the battery, and the power consumption allow a
correct estimate of the operation time to be expected. Data
about the (real) lead impedance, which normally has a value
around 500 , indicate possible malfunctions such as breaks
in the lead or insulation defects. Furthermore, the pulse amplitude, current, energy, and charge are displayed separately
for the atrial and the ventricular channel.
Pacing Parameters
All rhythmological parameters, such as pacing mode (e.g.,
DDD), rates, AV delays, refractory periods, and hysteresis
setting, are considered pacing parameters. During diagnostic
examinations, temporary programs with other parameter settings need to be activated on a short-term basis. To simplify
this process, some pacemakers possess two complete program
memories. This makes it possible to switch quickly between
two programs (e.g., within one cardiac cycle).
Diagnostic Information
This category includes the internal event counter with trend
monitor. The event counter normally registers events that occur over very long periods, such as atrial sensing and pacing,
ventricular sensing and pacing, or also ventricular extrasystoles (ventricular sensing outside the AV delay). The trend
monitor graphically depicts the heart rate (paced or sensed)
over time, where the temporal resolution can be selected
within a wide range (several minutes to several days or even
months). Frequently, it can be differentiated between a rolling or a fixed mode. Thus, either the oldest values are constantly overwritten by new values, or the trend recording
stops after one complete run. The stored rates are not momentary values, but they equal the average value of the respective
scanning interval.
PACEMAKERS
Programming head
Programmer
493
value. This access code is checked for exact agreement to exclude an erroneous programming by the wrong programmer,
faulty transmission, or other interference signals (18). Also, a
parity bit follows at the end of each data package, which is
another of several safety measures during data transmission
alone. The pulse sequence is subsequently amplitude-modulated to a carrier frequency, which corresponds to the resonance frequency of the telemetry coils. This frequency is not
standardized, and it thus depends on the manufacturer. It is
usually within a range of 104 and 105 Hz.
Pacemaker
RATE-ADAPTIVE PACEMAKERS
Figure 9. Bidirectional data transfer between pacemaker and programmer via the programming head by means of inductive telemetry,
which is used to interrogate programmed parameters and stored
Holter data from the pacemaker at the beginning of each follow-up
and to reprogram it when necessary.
Aside from event counter and trend monitor, another application of analog telemetry exists for diagnostic purposes.
The PHYSIOS TC (Biotronik, Inc.), for example, offers the option to transmit the filtered or unfiltered atrial and ventricular IEGM with markers to the programmer in real time. The
sampling rate is 256 Hz for dual-channel and 512 Hz for single-channel operation. The transmitted IEGM and marker
signals can be read on the monitor or by the programmer
printer. They can also be read on a connected ECG recorder,
which allows for the simultaneous display of these signals
with the surface ECG. In this manner, the especially noisefree IEGM secures the diagnosis for atrial arrhythmias, simplifies the analysis of complicated ECGs, and recognizes muscle potentials.
Technical Realization
To interrogate and program the pacemaker, the programming
head, which is connected to the programmer, is positioned
over the pacemaker (Fig. 9). The magnet within the programming head closes a reed switch in the pacemaker, activating
the transmission and the receiving mode. The components
specific for telemetry are illustrated in the lower part of the
block diagram in Fig. 5. The data are transmitted inductively
between the coil within the programming head and the telemetry coil of the pacemaker, in most cases mounted on the
hybrid circuit (Fig. 7, right). During data transmission, a temporary program paces with a constant rate (usually asynchronous) to avoid malfunctions resulting from incomplete data
transmission. Because the relative positions of both coils are
important for inductive coupling, modern systems facilitate
exact positioning by indicating the optimum position using
LEDs in the programming head.
Manufacturers do not use a unified coding procedure. The
pulse-position coding is frequently used (17). Here, the information is transmitted by a sequence of pulses. One way is
that the value of a parameter to be transmitted is coded either in digital or analog form by the distance of pulse flanks
or of pulses of constant lengths. With digital coding, two different pulse distances are used (e.g., to code the binary values
0 and 1, which are transmitted in sequence, following a
defined start coding). Some manufacturers send a certain access code in each data package, in addition to the parameter
The heart rate is the most important correcting variable used

by the organism for increasing the cardiac output. This occurs
under physical stress or various other influences resulting in
an increased metabolic demand. As discussed earlier in this
article, the large range of variation for the cardiac output illustrates how important an adequate heart rate adjustment
is for maintaining a sufficient organ perfusion. The sinus-controlled and thus physiological heart rate adaptation (e.g., in
the VAT mode with AV block) is maintained only for pacemaker patients showing neither a compromised sinus node
function nor SA block nor atrial flutter or fibrillation. In these
cases, a fixed-rate dual-chamber pacemaker is usually sufficient. But if adequate sinus rate is not present, as is the case
with sick sinus syndrome, or if the rate increases only insufficiently under load conditions, as with chronotropic incompetence, then it is necessary to implant a rate-adaptive pacemaker. Even if there is not yet any chronotropic incompetence
at the time of implantation, the physician should determine
whether an affected sinus node is to be expected as a result
of a progressing disease within the operation time of the implant.
The challenge for biomedical engineering consists of determining the metabolic demand with suitable sensors possessing high sensitivity and specificity. Consequently, the
pacing rate can be adjusted to a sufficient degree and within
physiologic response times. Figure 10 shows the principle behind rate-adaptive pacemakers using either corporeal or cardiac input. A number of sensors have been studied to date
that are based on completely different physical measuring
principles. These various sensors observe many different
physiologic indicators. However, only some of them have attained any clinical significance.
Open- and Closed-Loop Systems
The sensor principles that have so far been studied can be
classified according to several criteria as follows:
1. The control strategy in open- or closed-loop systems
(19,20)
2. The origin of the signal in corporeal or cardiac control
parameters (5)
3. The physiological relationship between the sensor signal and the activity of the autonomic nervous system in
primary, secondary, or tertiary sensors (21)
4. The physical measuring principle (e.g., temperature,
impedance, or potential measurement)
494
PACEMAKERS
Hemodynamic
Thermal
Metabolic
Physician
control input
Corporeal control input
Cardiac control input

Programmer
Temperature
Motion
Position
pCO2, pO2
Vagus/sympath.
Venous pressure
Respiration
Sinus rate
ANS
QT interval
VIP
PEP
Stroke volume
Contractility
Pacemaker
Body process
setpoints:
Controller
Hemodynamic
Thermal
Metabolic
Circulation
center
Emotional
Controlled
variables
Body
Chronotropy
Heart rate
Neural
Adjusting
element
Cardiac
output
Humoral
Circulation:
hemodynamic
thermal
metabolic
Inotropy
Stroke volume
Controlled
system
Disturbance factors:
Hemodynamic Thermal Metabolic Emotional
Figure 10. Principle of rate-adaptive cardiac pacing using corporeal or cardiac control parameters. The lower part of the picture shows the major components of the cardiovascular control
system, while the upper part summarizes different strategies to reestablish chronotropic adaptation by the pacemaker.
Ultimately, the pivotal factor is the clinical quality of the

sensor signal, specifically, up to which degree the physiological circulation regulation is restored for patient well-being.
Based on this sole consideration, the superiority of the closedloop systems over the open-loop systems is evident. The latter
are characterized by the fact that only one variable disturbing
the circulation is measured. One example is the physical activity detected by the mechanical motion sensor to determine
the pacing rate from the sensor signal. With this approach,
body movement is the input signal for rate control in obtaining a physiologic rate response.
The principal shortcoming of all open-loop systems is that
no feedback and thus no successful physiological-based control exists, notwithstanding the accuracy of the executed rate
response. Additionally, only the influence of one disturbance
variable can be measured. Other disturbing influences remain
unconsidered. In contrast, the pacing rate always reacts back
upon the sensor signal in closed-loop systems, thus forming a
closed control loop. These systems can be further divided into
those with a purely metabolic feedback (e.g., the partial oxygen pressure sensor) and those with a feedback mediated by
the autonomic nerves. The latter group is composed of systems that deliver a signal via cardiac measuring parameters.
The signal depends on the myocardial contractility and thus

on the nervous tone. With this signal, the natural control loop
for heart rate adaptation can be closed again. Because the
system-correcting variables that control the heart rate in a
healthy organism are accessed, this group currently constitutes an ideal method for physiological rate adaptation. Such
nervous system-controlled adaptation has an optimal amplitude and response characteristic that represents adequate reaction to influences from higher centers, (e.g., emotional
stress). The following section outlines all relevant sensor systems for artificial rate adaptation that are currently undergoing a moderately large-scale testing or are being commercially distributed.
Corporeal Parameters
Activity. Rate-adaptive pacemakers equipped with an electromechanical motion or activity sensor within the pacemaker
housing have experienced the widest distribution so far. Such
a configuration determines corporeal activity or kinetic energy. To date, various constructions such as piezo-vibrators,
capacitive sensors, or a movable sphere inside a coil have
been used for the detection of mechanical motion. Because the
PACEMAKERS
movements of the large muscles comprise the main part of

the physical load, their kinetic energy is a suitable rate adaptation parameter (2224). Another advantage is the easy clinical handling of such sensors. Because the sensor is located in
the pacemaker housing, no additional probes or special electrodes are necessary. This system can be operated with standard electrodes (that are perhaps already implanted). Also,
this sensor has short response times. Despite continuous improvements, the principal disadvantage of this concept does
remain. Namely, only an external signal is used instead of an
intrinsic, corporeal, circulation-related parameter. The external signal correlates with the actual, hemodynamic demand
only in an indirect way, and it can be influenced by mechanical tremors, vibrations, and so on. Alternatively, some load
conditions cannot be recognized at all (e.g., work involved in
holding objects, standing, or psychological stress).
Blood Temperature. Another suitable parameter for rate
adaptation that has also been used in commercially available
pacemakers is the central venous blood temperature (2527).
The use of this parameter can be traced back to studies by
Weisswange et al. (28). With physical motion, the central venous blood temperature (CVT) increases depending on the
load because of the low efficiency of the musculature and the
connected heat production. The CVT rises from 37 to 38.5C
with maximum physical activity. The temperature is measured with a high-resolution (0.025C) thermistor integrated
into the pacing electrode. The most important advantage of
the temperature-controlled pacemakers is that the rate adaptation is based on a metabolically influenced parameter. This
parameter also reacts to metabolic changes that are not exercise-related, such as fever, circadian deviations (a slight temperature decrease during nighttime), orthostasis, or even psychological stress. However, this is countered by disadvantages
in response time and practicality. The temperature increase
starts only with a certain delay after stress begins, and it
occurs considerably more slowly than a physiological rise of
the heart rate. The main disadvantage for clinical practice
consists of the necessity of a special thermistor-electrode with
a tripolar connection. On the other hand, this method approximates the actual metabolic demand much more closely than
the motion sensors in spite of the fact that no central physiological control variable is measured, which means that a
closed control loop does not exist.
Mixed Venous Oxygen Saturation (SO2) and pH. Other metabolic parameters with special sensors are the mixed-venous
oxygen saturation (termed SO2) and the pH-value of the
blood, the latter being tested only under nonclinical conditions. The mixed-venous oxygen saturation is measured by
reflectance oximetry, where an integrated optoelectronic,
glass-sealed sensor element is attached to the electrode at the
height of the right atrium (29,30). The pH measurement is
accomplished with an iridium/iridium oxide ring electrode. A
silver/silver chloride electrode at the pacemaker housing
serves as a reference (31). These two types of measurements
have three disadvantages in common: a specific electrode is
required; the response times are too slow for a short-term control, as is required by a rate-adaptive pacemaker; and the
long-term stability is not guaranteed. One reason for this is
that deposits that distort the result of the measurement form
over time at the SO2 sensor. The silver/silver chloride elec-
495
trode for pH measurement also does not display long-term

stability, and it is additionally problematic in its biocompatibility. Because of these results, these sensors have not been
applied in commercially available implants to date.
Respiration. In contrast, rate-adaptive pacemakers based
on minute ventilation measurements are presently offered by
several manufacturers. As early as 1982, a rate-adaptive
pacemaker that used the respiratory rate for rate adaptation
was available (32). Respiratory rate and minute ventilation
are determined via impedance measurements. Even though
earlier devices required an additional electrode or special bipolar electrodes (with a very large distance between tip and
ring) for this measurement, modern devices need only bipolar
standard electrodes (33). To measure the impedance, subthreshold bipolar measuring current impulses are fed through
the electrode ring, while the voltage between electrode tip and
housing is simultaneously measured. Because of the low conductivity of air, the measured impedance rises during inhalation and falls during exhalation. With suitable algorithms,
not only the respiratory rate but also the minute ventilation
is determined from the gained impedance signal. The minute
ventilation shows a much better correlation to the natural sinus rate than just the respiratory rate. However, no closed
control loop exists in this system. Furthermore, motion artifacts impair the impedance measurement, and the rise in the
rate occurs only with a certain delay.
Cardiac Parameters
Pressure Gradient. Patients with sick sinus syndrome have
usually lost only chronotropy, but not the inotropic adaptation; that is, the increase in contractility conveyed by the
sympathetic nerve still occurs under physical stress. But
heightened sympathetic activity also occurs based on response by the ANS to other stimuli (e.g., emotional stress).
Because of the thus triggered changes in the mechanogram of
the muscle cell, the course of contraction and also the pressure curve for the right ventricle are modified. The maximum
value of the first derivative of the right ventricular pressure
(dp/dtmax) especially exhibits a good correlation with the natural sinus rate in some studies and is therefore attractive as a
control parameter for rate adaptation. The pressure is measured by a pressure sensor that is integrated into the electrode, 28 mm behind the electrode tip (34). At the moment,
several prototypes of the pressure rise-controlled pacing system are undergoing clinical trials (34,35). The correlation between sensor signal and load has been proven to be good; the
response times are also physiological. The technical difficulty
resides in the long-term stability of the sensor. This is a result
of fibrin deposition and burrowing by the electrode into trabeculae of the ventricular wall, which can falsify the measurements. Thus, more studies are needed before applying these
prototypes to the clinical setting. The research should focus
on sensor long-term stability as well as on the reaction during
ischemia, cardiac insufficiency, and simultaneous drug therapy. But the disadvantage of requiring a specific electrode remains.
Peak Endocardial Acceleration. A newly developed measuring system consists of a piezoelectric acceleration sensor integrated into the electrode tip, which directly detects the accel-
496
PACEMAKERS
eration impulses of the adjacent heart wall (36). This new

system exploits changes in contractility, in much the same
way as is done with a pressure sensor, while avoiding the
long-term problems associated with direct pressure measurement. A clinical trial with the first prototype provided a very
good correlation of the evaluated maximum acceleration
value, PEA (peak endocardial acceleration), with myocardial
contractility. Thus, this maintains synchrony with sympathetic activity and the existing sinus rhythm (37). Because
the measuring signal should not be distorted by the process
of ingrowth by the electrode, a good long-term stability of the
sensor is to be expected but must still be clinically proven.
However, this measuring system also requires a special electrode with a head that has a significantly larger diameter
than standard electrodes because of the integrated sensor.
QT Interval. In the search for physiologic measurement parameters that can be obtained with standard electrodes without an additional sensor, the IEGM is a consideration. This
is a convenient value to use because the pacemaker already
measures it through the pacing electrode for sensing intrinsic
actions. Since 1983, the QT interval has been used as a control parameter for rate adaptation. That is, the time interval
between the QRS complex of paced events, such as the pacemaker stimulus and the end of repolarization, namely the T
wave in the IEGM (38). The QT interval shortens with rising
metabolic demand, under stress, but also with a rising rate.
Thus, this sensor signal displays a positive feedback. This
normally disadvantageous behavior can be compensated by
adjusting the transmission factor for setting the rate change
per QT shortening (39). This transmission factor is not constant in the current generation of pacemakers but imitates
the nonlinear relationship between QT and HR. The shortening of the QT interval is thought to be caused by the influence of catecholamines on the repolarization (19). Included in
this influence is noradrenaline, the sympathetic cardiac nerve
transmitter. Thus this sensor reacts not only to physical but
also to psychological stress. This relationship also explains
the good proportionality observed between the rate adaptation of the pacemaker and the amount of stress. The disadvantages of this sensor are the delayed increase at the stress
origin and the brief continuation of a high rate after the
stress has ended. Various nonmetabolic influences such as antiarrhythmic medication, ischemia, and changes in the electrolyte composition can influence the QT interval and thus
the pacing rate. The relationship between rate and QT interval varies individually to a very large degree. It also undergoes temporal changes, requiring short follow-up cycles and
frequent reprogramming. This sensor principle is used in clinical application. The positive feedback on the heart rate parameter is, however, a significant shortcoming.
Ventricular Depolarization Gradient. Besides the QT interval, other parameters of the ventricular IEGM were also studied (e.g., the ventricular depolarization integral). If the integral of the IEGM is formed over time, then the ventricular
depolarization gradient is defined as the negative extreme
value of this integral. Therefore, it corresponds to the integral
of the IEGM in the depolarization phase (40). It is measured
by a bipolar standard electrode. The ventricular depolarization gradient diminishes during stress and under catecholamine influence, and it increases during a rate increase. The
parameter changes sufficiently fast at the initiation of stress
and somewhat slower after it terminates. In the intact heart

(i.e., with normal chronotropic adaptation), this parameter remains approximately constant. These favorable properties of
the parameter, provided by the negative feedback of the heart
rate, allow the construction of a closed system. There have
been problems, however, because of paradoxical reactions to
position changes (orthostasis). Also, certain drugs may give
rise to other problems. Consequently, the system is not commercially available at the moment, but it is the subject of ongoing research. With the advent of the fractal-coated electrode, it has become possible to suppress almost completely
the polarization artifact in the IEGM caused by the pacemaker stimulus. This enables a much more exact and undistorted measurement of the IEGM, opening up new perspectives in using the ventricular evoked response (VER) of the
heart for physiological rate adaptation (41).
Stroke Volume. The changes in the contraction dynamics
of the myocardium, resulting from a heightened sympathetic
activity (increased inotropy), are expressed in other physiological parameters as well. Some of these are the stroke volume or the right ventricular preejection phase (PEP). Measuring methods determine the right ventricular volume by
intracardiac impedance measurement. To date, catheters
with 8, 6, 4, 3 or 2 equidistant ring electrodes have been used
for intracardiac volume measurements (4244). Even though
good results have been achieved with such multipole catheters in the left ventricle, the irregular geometry of the right
ventricle complicates the determination of a sufficiently exact
volume. In addition, respiratory influences, movement or
bending of the electrode during contraction, and variable AVconduction times and intrinsic rhythm lead to compromised
measuring results. A key error occurs with the transition
from an upright to a supine position. The stroke volume increases as a result of a sudden increase in the venous return,
to which the pacemaker would respond with a rate increase.
In contrast to this, the heart rate decreases slightly with a
normal circulatory regulation. This sensor has also not yet
been introduced into commercially available systems.
Preejection Phase. The preejection phase consists of the
electromechanical delay (from the QRS complex/stimulus to
the beginning of the mechanical activation) and the isovolumetric contraction (presphygmic phase). This parameter
shortens with increased contractility, but also with increased
end-diastolic volume. It is thus suited as an indicator for
physical as well as psychological stress. A direct dependence
on the heart rate can be neglected for practical concerns. This
parameter can also be derived from the intracardiac impedance measurement. Chirife (45) used a tripolar electrode for
the volume measurement. The end of PEP is marked by the
beginning of the ejection phase (i.e., by a sudden decrease in
volume). Another approach was taken by Schaldach et al. using intracardiac impedance plethysmography (46). The measurement is performed with a unipolar standard electrode. In
a method of measurement analogous to that of extracorporeal
impedance plethysmography, a second parameter (PEP*) is
obtained. This parameter is defined as the time interval between the ventricular stimulus and the point of the steepest
negative rise of the unipolar intracardiac impedance. It
closely correlates to PEP (47). The first generation of rateadaptive pacemakers on the basis of PEP* confirmed the
physiological significance of this parameter (48). They also re-
PACEMAKERS
vealed technological difficulties in determining the exact time

interval, mainly caused by limited temporal resolution of the
scanning function in the digital portion of the circuitry. This
measurement principle was then further developed and led to
the concept of the ventricular inotropic parameter (VIP).
Here, the inotropic state of the myocardium is directly determined from the morphology of the unipolar intracardiac impedance signal without the detour of a time interval measurement.
Ventricular Inotropic Parameter. The increased inotropy of
the myocardium under sympathetic influence is controlled to
a large degree by the increased influx of calcium. This heightened release of calcium, coming from the extracellular space
and from the sarcoplasmic reticulum, intensifies electromechanical coupling. From a mechanical viewpoint, this process
is expressed as an accelerated contraction (i.e., a higher contraction velocity). It is also expressed as a higher maximum
value of the contraction power and, also, as an accelerated
relaxation. The higher contraction velocity is the most interesting for its measurement technique; not only can it be captured by mechanical parameters like dp/dtmax, but it is also
reflected in the unipolar intracardiac impedance signal. With
a suitable algorithm, that of the RQ algorithm, the effective
rise of the unipolar impedance is measured in a specially designed measuring window. It is then compared to reference
values for rest and stress and used to calculate the pacing
rate (5,49). This physiological principle of rate adaptation
(based on nervous information) is already commercially available in the second pacemaker generation. It has shown very
promising results in the clinical field in a large number of
patients. Ambulatory stress tests prove an adequate rate adaptation concerning amplitude and temporal response characteristics. In addition to this, psychological stress tests, the
rate course during everyday life activities (Fig. 11), and special pharmacological studies also confirm the close correlation
of the measuring signal with nervous control (50).
Currently, a manual initialization of the algorithm for rate
adaptation is necessary. The unipolar intracardiac impedance
signal sometimes displays a drift over several months. This
makes it necessary to check the initialization at certain time
100
Heart rate (bpm)
90
Walking Housework
MSR
Sleeping
Sports
Travelling
497
intervals and to renew it, if indicated. However, procedures

are already underway that will initialize automatically after
the implantation. These are also intended to update the pacemaker parameters continuously and automatically.
Multiple Sensors
This overview of the various measuring signals has shown
that every single sensor has some specific advantages and disadvantages for rate adaptation. For that reason, some manufacturers pursue the strategy to use two or more sensors simultaneously, offering the advantage of complementing each
other as much as possible in their properties. A particularly
attractive combination is that of the motion sensor and a metabolic sensor. The former quickly determines a load but only
imprecisely measures the level, and the latter responds with
a delay but detects the degree of stress more precisely and is
less sensitive to external disturbances (e.g., tremors). Currently pacemakers that have a motion sensor combined with
QT-interval measurement are available; others utilize the
motion signal and minute ventilation measurement. The clinical results are promising (19,33), but it should not be overlooked that the integration of several sensors is connected to
a higher consumption of power, an increased programming
effort for the physician, and, quite relevantly, higher costs.
Because of growing economic pressure and increasing time
limitations for the pacemaker follow-up, the measurement of
yet more measuring variables is questioned. Therefore, a different strategy is favored by other manufacturers.
If nervous activity, such as the sympathetic tone, can be
successfully determined with sufficient exactitude (e.g., with
the intracardiac acceleration sensor or the unipolar intracardiac impedance measurement), then access is established to
the ANS and thus to a widely ramified and highly complex
network of biological sensors (baroreceptors, chemoreceptors,
etc.). Information from the corporeal intrinsic sensor system
is thus used for rate adaptation that has already been processed in the circulatory center. This intrinsic sensor system
determines the various inner and outer disturbances much
more comprehensively and quickly, as well as more precisely
than an artificial multisensor system. Moreover, it also captures nonmetabolic influences such as emotional stress. Consequently, it is to be expected that those systems measuring
just one good indicator of nervous activity reliably will be superior to those measuring several other parameters with a
lesser correlation.
CONCLUSION
80
70
60
12:00
16:00
20:00
24:00
04:00
Time (h:min)
08:00
12:00
Figure 11. Twenty-four-hour trend of the adaptive pacing rate

stored in the internal memory of the pacemaker. Rate adaptation
based on unipolar intracardiac impedance measurement shows appropriate response to everyday activities.
In its comparatively short history, pacemaker therapy has

undergone a rapid development. Today, the innovations allow
the physician to treat complex heart rhythm disturbances
with a therapy that is reliable and specific to the individual.
Another advantage in contrast to drug therapy is that pacemaker therapy has less propensity to lead to side effects.
While in the early years, the life-supporting function was the
prominent focus, modern rate-adaptive dual-chamber pacemakers secure a high quality of life for the patient. This is
because of their ability to reestablish the synchrony of atrial
and ventricular contraction and their provision of physiological rate adaptation. With minimal dimensions and weight,
the present implants possess a service lifetime that corresponds to the life expectancy of most patients.
498
PACEMAKERS
Distant monitoring of pacemaker-dependent patients via

telemedicine will further increase the safety of the patients
in the future. Extended Holter functions of the internal memory will provide the physician with more precise diagnostic
information. Considering time and cost limitations, the focus
is no longer solely on quantity. Instead, the therapy-relevant
information must be selectively extracted by a suitable choice
of the parameters to be monitored, which is accompanied by
expert systems support. These enhanced diagnostic features
will allow pacemakers to monitor other functional aspects of
the heart, such as medication monitoring or ischemia. An application that has already proven successful is allograft rejection monitoring of patients who underwent heart transplantation by means of the ventricular evoked response signal (51).
To decrease the programming and follow-up efforts, automatic functions are increasingly used that guarantee an unvarying, optimal pacemaker function and thus enable the
physicians to focus more attention on the patient. Some of
these functions include the automatic monitoring and adjustment of the sensing threshold and pacing energy, which secures safe sensing and pacing with minimal power consumption. Further automatic functions will follow (e.g., automatic
compensation of a possible sensor drift in a rate-adaptive
system).
Paralleling this approach, several concepts are pursued to
also draw tachycardic rhythm disturbances into the indication spectrum. Promising methods are antitachycardia pacing
and multisite pacing. A direct stimulation of the afferent vagal nervous pathways in the myocardium can also contribute
to the reestablishment of the neurohumoral balance. With the
preventative measures mentioned, the access to neurohumoral parameters makes possible the recognition of lifethreatening tachycardias at the early stages, and their
suppression. This is done without having to trigger an electroshock by an implantable defibrillator. Access to the state of
tone of the autonomic nervous system by intracardiac measuring methods will also make a major contribution to the
therapy optimization for other applications, such as to ensure
cardiac activity after a cardiomyoplasty (52).
BIBLIOGRAPHY
1. H. Antoni, Funktionen des Herzens, In R. F. Schmidt and G.
Thews (eds.), Physiologie des Menschen, 24th ed., Berlin:
Springer, 1990.
2. A. D. Bernstein et al., The NASPE/BPEG generic pacemaker
code for antibradyarrhythmia and adaptive-rate pacing and antitachyarrhythmia devices, PACE, 10: 794799, 1987.
3. M. Schaldach, The myocardium-electrode interface at the cellular
level. In A. E. Aubert, H. Ector, and R. Stroobandt (eds.), Cardiac
Pacing and Electrophysiology, Dordrecht: Kluwer, 1994.
4. A. Bolz, R. Frohlich, and M. Schaldach, Pacing and sensing performance of fractally coated leads, Proc. Annu. Int. Conf. IEEE
Eng. Med. Biol. Soc., 16: 365366, 1994.
5. M. Schaldach, Electrotherapy of the Heart, Berlin: Springer, 1992.
6. A. Lindgren and S. Jansson, Heart Physiology and Stimulation,
Solna, Sweden: Siemens-Elema AB, 1992.
7. W. Greatbatch, Twenty-five years of pacemaking, PACE, 7: 143
147, 1984.
8. S. Furman, Basic concepts. In S. Furman, D. L. Hayes, and D. R.
Holmes (eds.), A Practice of Cardiac Pacing, 3rd ed., Mount Kisco,
NY: Futura Publishing, 1993.
9. H. D. Friedberg and S. S. Barold, On hysteresis in pacing, J.

Electrocardiol., 6: 12, 1973.
10. S. Furman, P. Hurzeler, and V. de Caprio, The ventricular endocardial electrogram and pacemaker sensing, J. Thorac. Cardiovasc. Surg., 73: 258266, 1977.
11. M. Schaldach, Entwicklungstendenzen in der physiologischen
Stimulation des Herzens. In A. Weikl (ed.), Physiologische Stimulation des Herzens, Erlangen, Germany: Perimed, 1984.
12. M. J. Castellanet et al., Telemetry of programmed and measured
data in pacing system evaluation and follow-up, J. Electrophysiol., 1: 360375, 1987.
13. D. Epstein, Hybrid microcircuit screening: Does it work?, Hybrid
Circuit Technol., 4: 3639, 1987.
14. M. Schaldach, Reliability considerations in the design of implantable products. Biomed. Technik, 32: 276284, 1987.
15. R. Castello, A. G. Grassi, and S. Donati, A 500-nA sixth-order
bandpass SC filter, IEEE J. Solid-State Circ., 25: 669676, 1990.
16. S. Furman, Comprehension of pacemaker timing cycles. In S.
Furman, D. L. Hayes, and D. R. Holmes (eds.), A Practice of Cardiac Pacing, 3rd ed., Mount Kisco, NY: Futura Publishing, 1993.
17. K. M. Hugo, External programming. In J. G. Webster (ed.), Design of Cardiac Pacemakers, New York: IEEE Press, 1995.
18. J. T. Hartlaub, R. S. McDonald, and L. W. Shearon, Temporary
and permanent programmable digital cardiac pacemaker, U.S. Patent No. 4,253,466, 1981.
19. C. P. Lau, The range of sensors and algorithms used in rate adaptive cardiac pacing, PACE, 15: 11771211, 1992.
20. M. Schaldach and H. Hutten, A physiological approach to different concepts of rate adaptive cardiac pacing, Med. Prog. Technol.,
16: 235246, 1990.
21. A. F. Rickards and R. M. Donaldson, Rate-responsive pacing,
Clin. Prog. Pacing Electrophysiol., 1: 1219, 1983.
22. T. C. Wong et al., Portable accelerometer device for measuring
human energy expenditure, IEEE Trans. Biomed. Eng., 28: 467
471, 1981.
23. D. P. Humen, W. J. Kostuk, and G. J. Klein, Activity sensing rate
responsive pacing: Improvement in myocardial performance with
exercise, PACE, 8: 5259, 1985.
24. M. Schaldach, ERGOSRate-adaptive pacing system based on
motion energy, Prog. Biomed. Res., 1: 1, 1989.
25. E. Alt et al., Rate control of physiological pacemakers by central
venous blood temperature, Circulation, 73: 12061212, 1986.
26. C. Naumann dAlnoncourt et al., Temperaturgesteuerter
Herzschrittmacher THERMOS 01 nach His-Bundel Ablation,
Herzschrittmacher, 7: 168174, 1987.
27. M. Schaldach, Herzschrittmacher mit temperaturgesteuerter
Frequenzanpassung, Herzschrittmacher, 9: 514, 1989.
28. A. Weisswange et al., Frequenzsteuerung von Schrittmachern
durch Bluttemperatur. Verh. Dtsch. Ges. Kreislaufforsch., 44:
152, 1978.
29. A. Wirtzfeld et al., Central venous oxygen saturation for the control of automatic rate-responsive pacing, Circulation, 64 (Suppl.
IV): 299, 1981.
30. K. Stangl and E. Alt, Mixed-venous oxygen saturation. In E. Alt,
S. S. Barold, and K. Stangl (eds.), Rate Adaptive Cardiac Pacing,
Berlin: Springer, 1993.
31. L. Camilli, Initial use of a pH triggered pacemaker, PACE, 12:
10001007, 1989.
32. P. Rossi et al., Respiratory rate as a determinant of optimal pacing rate, PACE, 6: 502507, 1983.
33. H. G. Mond, Respiration. In E. Alt, S. S. Barold, and K. Stangl
(eds.), Rate Adaptive Cardiac Pacing, Berlin: Springer, 1993.
34. I. Ovsyshcher et al., First derivative of right ventricular pressure,
dP/dt, as a sensor for rate adaptive VVI pacemaker: Initial experience, PACE, 15: 211218, 1992.
PACKAGING OF OPTICAL COMPONENTS AND SYSTEMS

35. T. Bennett et al., Development of a rate adaptive pacemaker
based on the maximum rate-of-rise of right ventricular pressure
(RV dP/dtmax), PACE, 15: 219234, 1992.
36. A. F. Rickards et al., An implantable intracardiac accelerometer
for monitoring myocardial contractility, PACE, 19: 20662071,
1996.
37. H. Langenfeld et al., Myokardiale Kontraktionsbeschleunigung
als Sensor fur die frequenzadaptive StimulationErste klinische
Erfahrungen mit dem Living-BEST-System, Herzschrittmacher,
17: 6975, 1997.
38. A. F. Rickards and J. Norman, Relation between QT interval and
heart rate: New design of physiologically adaptive cardiac pacemaker, Br. Heart J., 45: 5661, 1981.
39. Rickards, W. Boute, and M. J. S. Begemann, The QT interval. In
E. Alt, S. S. Barold, and K. Stangl (eds.), Rate Adaptive Cardiac
Pacing, Berlin: Springer, 1993.
40. I. Singer, B. M. Steinhaus, and J. Kupersmith, Closed-loop rateadaptive pacemaker based on the ventricular evoked response. In
E. Alt, S. S. Barold, and K. Stangl (eds.), Rate Adaptive Cardiac
Pacing, Berlin: Springer, 1993.
41. A. Bolz et al., Low polarization pacing lead for detecting the ventricular evoked response, Med. Prog. Technol., 19: 129137, 1993.
42. R. W. Salo et al., Continuous ventricular volume assessment for
diagnosis and pacemaker control. PACE, 7: 12671272, 1984.
43. J. Baan et al., Calibration and application of the conductance
catheter for ventricular volume measurement, Automedica, 11:
357365, 1989.
44. G. Boheim, Intrakardiale Impedanzmessung zur frequenzadaptiven Stimulation des Herzens, PhD Thesis, University of ErlangenNuremberg, 1988.
45. R. Chirife, Physiological principles of a new method for rate responsive pacing using the preejection interval, PACE, 11: 1545
1554, 1988.
46. M. Schaldach, W. Rentsch, and H. W. Rentsch, Advances in intracardiac impedance plethysmography, Proc. Annu. Int. Conf. IEEE
Eng. Med. Biol. Soc. 12: 711713, 1990.
47. W. Rentsch et al., Zur physiologischen Frequenzanpassung von
Herzschrittmachern mittels systolischer Zeitintervalle, Z. Ges.
inn. Med., 42: 386389, 1987.
48. M. Schaldach, PEP-gesteuerter Herzschrittmacher, Biomed.
Technik, 34: 177184, 1989.
49. M. Schaldach and H. Hutten, Intracardiac impedance to determine sympathetic activity in rate responsive pacing, PACE, 15:
17781786, 1992.
50. J. Witte et al., Frequenzadaptive Stimulation mittels eines KontraktilitatssensorsEine multizentrische Studie, Herzchrittmacher, 16: 816, 1996.
51. H. Hutten et al., CHARMComputerized heart acute rejection
monitoring, Biomed. Technik. 41 (2): 3540, 1996.
52. G. Bolotin et al., CardiomyoplastyThe beginning of a new era,
Isr. J. Med. Sci., 32: 321326, 1996.
MAX SCHALDACH
University of ErlangenNuremberg
499
PATIENT MONITORING
niques have already been developed and supplied commercially.

Safety is an important feature of any monitoring device
because monitoring is performed for a long period of time for
the critically ill patient. Electric safety is strictly required especially when the monitoring device has electric contacts to
the patient body. Sometimes, two or more monitors are applied to a patient. Leakage current should be avoided under
any failure of each device. Electromagnetic compatibility is
also important. Monitoring instruments should be immune to
any possible electromagnetic interference from telemetering
devices, mobile telephones or other noice sources such as electrosurgery.
Many patient monitors have an automatic alarm function.
When a monitoring item is expressed as a single value such
as heart rate, blood pressure, or body temperature, the alarm
condition is determined by setting a level or range, and the
monitor gives an alarm sign, such as warning and urgent,
according to the patients condition. When the monitoring
item is expressed in a waveform, such as the ECG, the alarm
system needs to be able to perform real-time waveform analyses. In any alarm system, two kinds of errorfalse positives
and false negativesmay occur. In critically ill patients, a
false negative may be fatal. While false positives may be tolerated to some extent, repeated false alarms may seriously
disturb the clinical staff. In general, any alarm system requires some logic, and some times highly intelligent signal
processing is required.
Modern medicine allows for the monitoring of high-risk patients so that medical treatment can be applied adequately as
their condition worsens. To detect changes in the physiological condition of each patient, appropriate monitoring is applied routinely according to the patients condition, at least in
well-equipped hospitals. Patient monitoring usually means
the physiological monitoring of high-risk patients using appropriate instruments.
In hospitals, there are many sites where patient monitoring is especially important. For example, in the operating
room, instruments such as a pulse oximeter are used for monitoring anesthesia; in the intensive care unit, vital signs are
monitored; in the coronary care unit, the patients electrocardiogram (ECG) is routinely monitored and analyzed automatically; and in the incubator, the vital signs of the infant as
well as the internal environment of the incubator are monitored. In addition, during examinations such as cardiac catheterization, and therapeutic procedures such as hyper- or hypothermia therapy, patient monitoring is required for
ensuring safety. Even in the general ward, monitoring is performed fairly often when some risks are suspected. By using
a telemetry system, the patient is not constrained to a bed.
Even out of the hospital, patient monitoring is still performed
in some situations. In the ambulance, postresuscitation management requires the use of a cardiac monitor. In the home
where medical care such as oxygen therapy and intravenous
infusion therapy is carried out, monitoring instruments are
helpful. A so-called Holter recorder is used in which 24-h ECG
is recorded for detecting spontaneous events such as cardiac
arrhythmia.
There are many parameters that are used for patient monitoring: Among them are heart rate, ECG, blood pressure, cardiac output, rate of respiration, tidal volume, expiratory gas
content, blood gas concentrations, body temperature, metabolism, electroencephalogram (EEG), intracranial pressure,
blood glucose levels, blood pH, electrolytes, and body motion.
Many types of monitoring techniques and instruments have
been developed to enable measurement of these parameters.
For high-risk patients, monitoring should be performed
continuously. The real-time display of the trend or waveform
of each parameter is helpful especially in a patient who is
experiencing cardiopulmonary function problems, because if
a sudden failure of respiration or circulation is not detected
immediately it may result in the physiological state of the
patient becoming critical. The reliability of monitoring is
quite important. In some situations, invasive procedures for
monitoring are allowed if they are considered essential. For
example, an indwelling arterial catheter is used when instantaneous blood pressure has to be monitored continuously.
However, invasive methods are undesirable if the patients
condition is less critical. In some situations, noninvasive
methods are preferred. Because noninvasive methods are always more difficult to perform or less accurate than invasive
methods, the development of reliable noninvasive monitoring
techniques is highly desirable; many smart noninvasive tech-
CONVENTIONAL PATIENT MONITORING TECHNIQUES

Electrocardiogram Monitoring
For sudden heart failure, urgent treatment is required. Monitoring of heart function is therefore quite important. The ECG
is the most convenient method of monitoring the electrical
function of the heart, whereas the mechanical pump function
of the heart is best monitored by examining the patients
blood pressure and cardiac output. An ECG signal can be obtained by attaching electrodes to the body surface. For patient
monitoring, electrodes are always attached to the torso as
shown in Fig. 1(a), whereas the standard lead system in
which electrodes are attached to the limb and chest is used
in ordinary ECG examinations for diagnosis. Disposable ECG
electrodes, as shown in Fig. 1(b), are commonly used for longterm monitoring. A stable ECG can be obtained using these
electrodes for a day or longer.
The ECG waveform thus obtained is always displayed on
a CRT monitoring screen with ordinary sweep speeds, together with other parameters. Unusual waveforms such as
premature ventricular contractions can be identified visually.
However, it is unlikely that someone would be able to watch
the monitor screen all of the time. Most ECG monitors have
a built-in computer that automatically detects abnormal
waveforms and triggers the alarm. To reduce as much as possible the number of false alarms, both false negatives and
false positives, highly intelligent algorithms for detecting abnormal waveforms, such as arrhythmias, have been developed
1
PATIENT MONITORING
Blood Pressure Monitoring
+
(a)
Metal snap
Plastic body
Adhesive pad
Foam with gel

Ag-AgCl
(b)
Figure 1. Typical electrode locations for ECG monitoring (a), and a

cross-section of a disposable foam electrode (b).
and installed in intensive care monitoring systems (1). Most

bedside ECG monitoring systems have a real-time display
and large data storage facility that allows for retrospective
observation. Some of them have a memory capacity that is
able to record an ECG for up to 24 h. Radiotelemetering is
convenient, even in bedside monitoring. Eliminating the cable
connection to the patient is advantageous not only to make
the patient less restricted, but also to attain electrical safety.
However, electromagnetic compatibility should be secured
when it is used together with other instruments.
For ambulatory patients, Holter ECG monitoring is performed in which the ECG is recorded typically for 24 h. The
typical Holter recorder records the ECG on an audio cassette
tape for 24 h, then the tape is brought to the hospital, the
recorded ECG is played back by a scanner at 60 or 120 times
the recording speed, and analyzed automatically so that different kinds of arrhythmias and other abnormalities may be
classified and counted. To detect and record only pathological
waveforms, a digital recorder with solid-state memory can be
used; for example, a system can detect automatically the
change in ECG during transient myocardial ischemia and record up to 18 episodes that are only 6 s each (2). Although
longer time digital recording needs a very large memory capacity, 24 h recording is realized using a small hard disk
drive in a system in which the ECG data is first stored in a
solid-state memory and then transferred to the disk over
short periods of time (3).
Arterial blood pressure monitoring is essential in a patient

whose circulation is unstable, and is commonly performed
during cardiovascular surgery and postoperative care. There
are two methods of blood pressure monitoringdirect and indirect. In the direct method, a catheter is introduced into the
artery as shown in Fig. 2, and a pressure transducer is connected to the proximal end of the catheter. To avoid blood
clotting in the catheter, a small amount of saline is supplied
either continuously or intermittently. Intraarterial pressure
can be measured accurately enough as long as the transducer
is adequately calibrated. Either a strain-gage or capacitive
type of pressure transducer is commonly used for this purpose. Disposable pressure transducers are convenient because
sterilization of the transducer before use is troublesome. In
addition, the performance of disposable pressure transducers
is comparable or even better than that of expensive reusable
pressure transducers (4).
The catheter-tip pressure transducer which has a pressure-sensing element at the tip is sometimes used for intraarterial pressure monitoring. It has many advantages: It
has no time delay and has a flat frequency response over a
wider range; saline injection is unnecessary; and it is less affected by the mechanical motion of the catheter. However, it
is fragile and expensive. Many different principles can be
used in detecting pressure at the tip, such as semiconductor
strain gauges, and capacitive and optical principles. Some
transducers have many pressure-sensing elements near the
tip. For example, a model is available that has up to six pressure sensing elements in an 8F size tip (outer diameter 2.67
mm) (Mikro-Tip, Millar Instruments, Inc., Houston Texas).
While the direct blood pressure measurement method is
accurate and reliable, it is an invasive procedure, and, thus,
an indirect noninvasive method is preferred for less critical
patients. The most common method of indirect blood pressure
measurement is the auscultatory method in which a pressure
cuff is attached to the upper arm. The cuff is deflated from a
position somewhat above the systolic pressure, and both the
systolic and diastolic pressures are determined by monitoring
a sphygmomanometer while listening for the Korotkoff sound
using a stethoscope. While the auscultatory method is the
standard method of clinical blood pressure measurement, and
is actually performed for patient monitoring such as during
anesthesia, it is neither automatic nor continuous. Hence, a
noninvasive continuous blood pressure monitor had been in
demand. Two methods have now become available: the vascular unloading method and the tonometry.
Infusion tube
Stop cock
Intraarterial catheter
Pressure transducer
Figure 2. The conventional method of direct arterial pressure monitoring.
PATIENT MONITORING
Finger cuff
Interface module
IR LED
Pressure
transducer
Pressure
source
Photo cell
Servo valve
Plethysmogram
+
Set point
Cuff pressure setting

(a)
Tonometer
Pressure sensor array
Artery
Skin
Bone
(b)
Figure 3. Indirect methods of instantaneous arterial pressure monitoring: (a) vascular unloading technique, and (b) tonometry.
The vascular unloading method is used to measure instantaneous intraarterial pressure by balancing externally applied pressure to the intravascular pressure using a fast
pneumatic servo-controlled system (5). As shown schematically in Fig. 3(a), a cuff is attached to a finger, and near-infrared light transmittance is measured at the site where the cuff
pressure is affected uniformly. Because absorption at nearinfrared is mainly due to the hemoglobin in blood, the change
in light absorption corresponds to the change of blood volume
at optical pass, thus a pulsatile change in transmitting light
intensity is observed from the pulsation of the artery. It is
possible to compensate for the pulsatile change of arterial
blood volume by introducing a servocontrol in which cuff pressure is controlled by the intensity of the transmitted light so
that an increase of arterial blood increases light absorption
and the signal increases cuff pressure so as to obstruct further increase of arterial flow. If such a servocontrol works fast
enough and with a sufficient loop gain at an adequate level of
light intensity, a condition is realized where the intraarterial
and the cuff pressures are balanced. At this condition, the
circumferential tension of the arterial wall is reduced to zero;
such a situation is called vascular unloading. It has been
shown that accurate blood pressure together with instantaneous arterial pressure waveforms can be obtained when an
adequate servocontrol system is introduced and adjusted correctly (6). A commercial unit that uses this principle has been
developed (Finapress, Ohmeda, Englewood, Colorado). In this
system, the interface module, which has a pneumatic servovalve is attached to the back of the hand so that the connection from the valve to the finger cuff is minimized, thus reducing the time delay.
Tonometry is a method of measuring internal pressure

from the reaction force. When a flat plate is pressed onto a
flexible deformable boundary membrane to which internal
pressure is exerted, internal pressure can be measured from
the outside regardless of the transverse tension developed in
the membrane. This principle has been applied successfully
in intraocular pressure measurement, and it is also applicable
to arterial blood pressure measurement (7). As shown in Fig.
3(b), the tonometry transducer, the tonometer, is applied to
the skin surface so that an artery is just beneath the sensing
element, and a part of the arterial wall is flattened. To detect
the pressure at the center of the arterial flattening, a multiple-element transducer is used, and the value at the center of
the pressure distribution is detected automatically. Measurement is always performed on the radial artery at the wrist. A
tonometer is now commercially available (Jentow, Nihon Colin Co., Komaki-shi Japan).
Sometimes, blood pressure is monitored in an ambulatory
patient. For this purpose, a fully automated portable sphygmomanometry system is used. A pressure cuff is attached to
the upper arm, and is inflated intermittently at selected intervals. The Korotkoff sound is detected by a microphone, and
systolic and diastolic pressures are determined and stored in
a memory. Commercial models are now available (e.g., Medilog ABP, Oxford Medical Ltd., Oxford, UK) (8).
Cardiac Output Monitoring
Cardiac output is the volume of blood ejected by the heart
per unit time. Because the capacity of circulatory transport is
proportional to cardiac output, and the level of metabolism is
limited by the capacity of oxygen transport, cardiac output
has to be maintained above a level corresponding to the oxygen demand. Even if blood pressure is normal, cardiac output
drops when peripheral vascular resistance is increased. Thus,
to establish the state of circulatory function correctly, monitoring both blood pressure and cardiac output is desirable.
However, there is no well approved noninvasive method of
cardiac output monitoring. In practice, the thermodilution
method has been used most commonly in critically ill patients
both during surgery and in intensive care units, although it
is an invasive procedure.
In the thermodilution method, a Swan-Ganz thermodilution catheter is introduced into the pulmonary artery through
the superior vena cava and right heart as shown in Fig. 4.
Approximately 10 mL of cold saline at near 0C is injected
instantaneously into the right atrium, and the temperature
change is recorded by a thermistor placed in the pulmonary
artery. Cardiac output is then obtained by the amount of cold
saline divided by the area of the blood temperature record
that lies under the baseline (9). It has been confirmed by
many studies that the thermodilution method of cardiac output measurement is reliable and accurate enough for most
clinical purposes. Although the monitoring is not continuous,
measurement is repeatable, and the catheter can be placed
for several days while the patient is in an intensive care unit.
The measurement of thoracic impedance has been studied
as a method of assessing cardiac output. According to the
original Kubiceks method (10), four band electrodes are used
so that two are attached around the neck and two around the
upper abdomen. An ac current in the 20 to 100 kHz range at
a current level within the range from 10 A to a few milli-
PATIENT MONITORING
Thermodilution
catheter
Balloon
Thermistor
Aorta
Pulmonary
artery
Superior
vena cava
Left atrium
Injection
port
Right
atrium
Left ventricle
Right ventricle
Figure 4. Thermodilution method for cardiac output measurement.
amps is supplied through the outer electrode pair, and the

induced voltage is measured from the inner electrode pair.
The stroke volume is computed from the slope of the thoracic
impedance change in the ejection phase assuming a homogeneous cylindrical model for large arteries. The cardiac output
is expressed by the stroke volume times the heart rate. Although inconsistencies remain among the reports that have
evaluated the impedance method in contrast to other methods
of cardiac output measurement, its noninvasiveness is a great
advantage and thus further improvements are desirable.
Respiratory Monitoring
Respiratory monitoring involves monitoring of ventilation and
respiratory gases. While ventilation of the lung can be assessed by observing movements of the thorax, it can be measured quantitatively by either gas flow in the airway or volume changes of the lung. During anesthesia or under
artificial ventilation where the patient is intubated, gas flow
in the airway can be monitored by inserting a flowmeter between the endotracheal tube and the breathing circuit. Many
different types of flowmeters such as a rotameter, pneumotachograph, hot-wire anemometer, ultrasound flowmeter, and
vortex flowmeter have been used. Most of them provide instantaneous gas flow rates. Tidal volume can be obtained by
integrating the flow rate for the inspiratory or expiratory
phase. Under artificial ventilation using a volume-limit type
of mechanical ventilator, tidal volume is determined simply
by presetting the ventilator.
Spontaneous breathing in unintubated patients can be
monitored by the respiratory motion of the thorax and abdomen. A simple method of monitoring such motion is to measure the circumferential length or cross-sectional area of the
thorax and abdomen. A flexible belt containing a zigzag-fashioned wire can be used as a transducer. When it is attached
to the thorax or abdomen so as to form a single-turn coil, its
inductance changes with respiratory motion, and tidal volume
can be obtained with considerable accuracy (11). A commer-
cial model of this type is currently available (Respitrace, AMI

Inc., Sedona, Arizona).
Lung volume change can also be monitored by measuring
the electrical impedance across the thorax (12). Impedance is
measured by placing electrodes at both sides of the thorax,
applying an ac current, and detecting the voltage that develops between the electrodes. Although thoracic impedance depends largely upon electrode position, the size and shape of
the body, and body fluid distribution, it can be a quantitative
monitor of lung volume changes when it is calibrated adequately using a spirometer.
Respiratory gas is also a common parameter that is used
for patient monitoring; monitoring the level of carbon dioxide
in expired air is especially important during anesthesia and
in intensive care where artificial ventilation is performed. In
physiological conditions, the carbon dioxide content in the
body fluids, particularly in the arterial blood, is always maintained within a narrow range by the regulatory mechanism
of respiration, but it may vary largely under artificial ventilation when the setting of the ventilator is inadequate. The arterial carbon dioxide partial pressure is related to the carbon
dioxide content in expired air and especially to the value at
the end of the expiratory phase. Carbon dioxide in the expired
air can be monitored beat-by-beat by a carbon dioxide analyzer, called a capnometer, in which carbon dioxide content is
measured by infrared absorption (13). There are two types of
capnometer: the side-stream capnometer and the mainstream
capnometer. In the side-stream capnometer, the sensor is located in the main unit, and a small amount of gas flow
branched from the patients airways is pumped continuously
to it through a fine tubing. In the mainstream capnometer, a
cuvette with an infrared source and a detector is inserted between the endotracheal tube and the breathing circuit as
shown in Fig. 5. Although the mainstream capnometer has
the advantage of no time delay, it has disadvantages, such as
the condensation of water vapor to the window and loading a
weight to the connector.
The mass spectrometer has also been used for continuous
respiratory gas monitoring (14). It can be used to analyze
many gasses simultaneously, not only physiological gasses
such as oxygen, carbon dioxide, and nitrogen but also anesthetic gas such as nitrous oxide, halothane, enflurane, and
isoflurane. In addition, many patients can be monitored with
the aid of a mass spectrometer by using an inlet select unit.
In fact, a single mass spectrometer system is capable of serIR source
Sapphire
windows
Endtracheal
tube
Breathing
circuit
Sapphire
windows
Rotating filters
IR detector
Motor
Figure 5. Configuration of a mainstream capnometer.
PATIENT MONITORING
vicing up to sixteen patients (Lifewatch Monitor, Perkin-Elmer Co., Pomona, California).
I1
LED1
Blood gas always means the oxygen and carbon dioxide contents of the blood. Because most of the oxygen in the blood
exists in combination with hemoglobin, the oxygen content of
the blood can be expressed in terms of the ratio of the amount
of oxyhemoglobin to that of total hemoglobin; this ratio is
called the oxygen saturation. A small amount of oxygen, usually less than 1%, remains in the plasma as dissolved oxygen,
and its amount is expressed in terms of oxygen partial pressure. Although there is a relationship between oxygen saturation and oxygen partial pressure, this relationship is nonlinear so that saturation increases steeply with increasing
oxygen partial pressure when the latter lies in the range 20
to 40 mm Hg (2.7 to 5.3 kPa), but tends to saturate when the
oxygen partial pressure reaches above 60 mm Hg (8 kPa). In
normal arterial blood, oxygen saturation is above 98%, and
oxygen partial pressure is approximately 100 mm Hg (13.3
kPa). The main purpose of monitoring oxygen level is to confirm the oxygen transport which sustains metabolic demand.
Carbon dioxide is highly soluble in body fluids, and it is
also converted, reversibly, to bicarbonate ions. Therefore,
blood plasma as well as interstitial fluids have an apparently
large storage capacity for carbon dioxide. However, changes
in the carbon dioxide content of the body fluids causes a
change in the acid-base balance of those body fluids, which is
expressed by pH. Thus, it is important to maintain an adequate carbon dioxide level in the body fluids. It is therefore
monitored by measuring the partial pressure of carbon dioxide of arterial blood.
Blood gas levels can be measured by taking a blood sample
and analyzing it using a blood gas analyzer which provides
information about the partial pressures of oxygen and carbon
dioxide, and about the pH of the blood. However, in a patient
whose respiration is unstable, blood gas values may fluctuate
so that frequent measurement is required, and hence continuous blood gas monitoring is preferred.
Arterial blood oxygen saturation can be monitored noninvasively using a pulse oximeter (15). Due to the difference in
the spectral absorption of oxyhemoglobin and reduced hemoglobin, the oxygen saturation of a particular blood sample can
be determined by absorption measurements at two wavelengths, typically in a red band between 600 nm and 750 nm
and in an infrared band between 850 nm and 1000 nm. However, the tissue in vivo contains both arterial and venous
blood, and hence light absorption occurs by both components.
To obtain the arterial component selectively, the pulsatile
component is extracted. As shown in Fig. 6(a), light absorption is usually measured in a finger. Two light-emitting diodes of different wavelengths, for example 660 nm and 910
nm, are operated alternately, and the transmitted light is detected by a photocell. The pulsatile components of both wavelengths are then extracted by a bandpass filter. Arterial oxygen saturation is determined from the ratio of these two
components.
Although the pulse oximeter is reliable enough and has
been used successfully for patient monitoring in most cases,
measurement sites of the transmittance measurement are
limited, and thus a reflection-type pulse oximeter in which
Photo cell
Logarithmic
amplifier
Blood Gas Monitoring

LED2
Multiplexer
Bandpass
filter
I2
yyyy
;;;;
;;;;
yyyy
=
log I1
S pO 2
log I2
(a)
Thermistor
Heating element
Cathode
Electrolyte
Anode
Oxygen permeable
membrane
Epidermis
Capillary
Artery
Vein
(b)
Figure 6. Pulse oximetry (a), and transcutaneous oxygen electrode

(b).
backscattered light is measured has been developed (16). In

back-scattered light measurement, a difficulty arises due to
the fact that the optical pathlength may vary when absorption is varied, although it is not changed as much in transmission measurement. In principle, this difficulty can be solved
by using more than three wavelengths, however, a reflectiontype pulse oximeter with comparable performance to the
transmission-type oximeter has not yet been developed. In
some applications, the reflection-type oximeter is highly appreciated. For example, it is applied to fetal monitoring during labor in which the sensor is applied to the skin of the fetal
head (17).
The oxygen content in arterial blood can also be measured
continuously and noninvasively with the aid of a transcutaneous oxygen electrode (18,19). The configuration of the probe is
shown in Fig. 6(b). The principle employed is that of polarographic measurement, by which current drains proportionally
to the amount of oxygen that reaches the cathode by diffusion
through the oxygen permeable membrane. Because the oxygen flux is determined by the gradient in oxygen partial pressure at the membrane, and the oxygen partial pressure at the
electrode surface is reduced to zero by the electrode reaction;
the current that results from the oxygen flux depends upon
the oxygen partial pressure on the outside of the membrane.
When the probe is used for measuring arterial oxygen partial
pressure, the electrode body is heated to approximately 42 or
43 C. At this temperature, arteriovenous shunts in the skin
tissue fully open, thus allowing large amounts of blood to flow
through the tissue, far more than is required nutritionally, so
that the venous blood has almost the same oxygen content as
that of the arterial blood. Consequently, the oxygen partial
pressure in the tissue reaches almost the same level as that
of the arterial blood, and thus the arterial oxygen partial
pressure can be measured transcutaneously.
PATIENT MONITORING
Both the pulse oximeter and the transcutaneous oxygen

electrode can be used for monitoring blood oxygenation; however, each method has advantages and disadvantages. The
pulse oximeter is safe, easy to use, inexpensive, and sensitive
at lower partial pressures. However, for higher oxygen partial
pressure where oxygen saturation is almost 100%, a pulse oximeter can not detect any change in oxygen partial pressure.
Higher oxygen partial pressures may occur during, for example, oxygen therapy. In such a condition, oxygen partial pressure may vary in wider range, and thus a transcutaneous oxygen electrode can be a good monitor of gas exchange in the
lung.
Carbon dioxide partial pressure in the blood can also be
measured transcutaneously using a heated carbon dioxide
electrode, similar to the transcutaneous oxygen electrode. The
carbon dioxide electrode consists of a pH electrode covered
with a carbon dioxide permeable membrane (20). This type of
electrode has been used for neonatal monitoring. The combined oxygen and carbon dioxide electrode which consists of a
transcutaneous carbon dioxide electrode and a transcutaneous oxygen electrode is also available (21).
Balloon
inflation port
Drainage eye
Balloon
Urinary
drainage
Silicone catheter
Thermistor
Cable
(a)
Temperature output
yy
;;
Heater
Thermal
insulator
Thermistors
(b)
Body Temperature Monitoring

The term body temperature usually means the body core temperature, which is the temperature of the central part of the
body. Many different techniques have been used for monitoring body temperature (22). Although different body parts
have different temperatures, such differences are small when
the temperature is stable, so that body temperature can be
monitored fairly well at many measurement sites. Body temperature is usually measured by a clinical thermometer at the
oral cavity. Recently, the tympanic thermometer is also used.
For continuous monitoring, it is measured at the rectum,
esophagus, bladder, auditory canal, tympanum, nasal cavity,
or digestive tract. However, when body temperature varies,
significant differences in observed temperatures may occur
between sites. Thus when rapid changes of body temperature
have to be monitored the measurement site used is important.
Rectal temperature has been used widely in patient monitoring because the rectum is a convenient site into which a
thermometer probe can be inserted far enough to protect it
from heat loss. Rectal temperature is always higher than oral
temperature as well as temperatures of other sites, and has
been considered to be a reliable indicator of body core temperature. However, when body temperature varies, changes in
rectal temperature are delayed comparable to those of other,
more central parts of the body, and thus rectal temperature
cannot be accurate enough for monitoring in such conditions.
The esophagus has been used most frequently as a site for
body temperature monitoring during anesthesia. Esophageal
temperature is measured by inserting a probe through the
mouth or nose so that the sensor tip is positioned at nearheart level. Under stable conditions, esophageal temperature
is intermediate between oral and rectal temperature, and follows internal temperature changes rapidly.
Bladder temperature can be monitored using a thermistortipped bladder catheter as shown in Fig. 7(a). Although bladder temperature is close to rectal temperature in stable conditions, it follows internal temperature changes rapidly. Bladder temperature is recommended as a measurement site for
Figure 7. Two methods of body temperature monitoring: (a) thermistor-tipped bladder catheter, and (b) zero-heat-flow method.
core temperature, particularly for patients in whom bladder

catheterization is indicated.
Body temperature can also be monitored across the skin
using the zero-heat-flow method as shown in Fig. 7(b) (23).
The probe that is used in this method has two thermistors to
detect heat flow across the probe. It also has a heater, and
the heater current is controlled so that the temperatures of
two thermistors are equal, which means that we can compensate for the heat flow from the skin to the outer air. Under
such conditions, the probe can be regarded as an ideal thermal insulator. When the skin surface is insulated, the temperature gradient in the tissue near the surface will vanish,
and finally the temperature of the surface of the skin will
reach that of the deep tissue. A commercial model (Coretemp,
Terumo Co., Tokyo) has now been developed for which discshaped probes of different sizes, from 15 mm to 80 mm in
diameter, are available. By applying a probe to the forehead,
chest, or abdomen, body temperature can be monitored continuously for several days in intensive care units (24). Simultaneous monitoring of body core and peripheral temperatures
by applying probes to the forehead and to the sole of a patients foot, temperature differences between the body core
and the limbs can be observed which can be a useful index of
peripheral circulation (25).
Intracranial Pressure Monitoring
The brain is surrounded by the skull, and the inner pressure
of the skull is almost uniform and is called the intracranial
pressure. Normal intracranial pressure is about 10 mm Hg
(1.3 kPa), referring to the zero-pressure level in upper cervical spine. However, because of the high stiffness of the skull,
a small increase in cranial volume causes a significant increase in intracranial pressure. Increases in intracranial
pressure are serious, because they cause obstruction of the
cerebral blood circulation. Such a situation can occur follow-
PATIENT MONITORING
ing intracranial bleeding, cerebral edema, growth of tumors,

infectious lesions, and parasites. Therefore, intracranial pressure is measured and continuous monitoring is performed in
such patients, especially when increase in intracranial pressure is likely to occur.
To estimate intracranial pressure, spinal measurement
has been used. Because communication exists between the
spinal fluid and the fluid in the ventricles, cerebral pressure
can be measured by puncturing the lumbar vertebra. However, such a technique cannot be used for monitoring intracranial pressure. To monitor intracranial pressure continuously, an invasive method has to be used, except in neonates
who have natural openings of the skull, the fontanelles,
through which intracranial pressure can be measured noninvasively.
Both extracranial measurement using a liquid-filled tube
and intracranial measurement using an implantable transducer are possible, and there are many different approaches
for each method (26). In extracranial measurement, a liquidfilled tube is introduced into a ventricle, subdural space, or
subarachnoidal space via a burr hole made in the skull, as
shown in Fig. 8(a), and a pressure transducer is connected to
the proximal end of the tube. A catheter-tip pressure transducer can also be used in place of the fluid-filled tube.
A pressure transducer that fits into a bore hole made in
the skull, as shown in Fig. 8(b), may also be used. In this
;;;
yyy
;
yy
;
yyyyy
;;;
;;
yyyy
;;;;
Skull
Skin
Pressure catheter
Brain
Dura
(a)
Skull
Skin
Pressure transducer
Brain
Dura
(b)
Fontanometry transducer
Skull
Skin
Brain
Dura
(c)
Figure 8. Methods of intracranial pressure monitoring: (a) placement of a pressure catheter in the subarachnoidal space, (b) implanting a transducer in a bore hole through the skull, and (c) fontanometry.
configuration, the diaphragm of the pressure transducer can

be adjusted to the epidural surface so that the error due to
deformation of the dura mater can be minimized. A telemetry
system is also attempted in which a small transmitter is assembled in the transducer unit and the signal is received by
a coil placed on the skin. The elimination of the cable connection is advantageous to avoid infection.
In the neonate, the skull is not completely formed so that
openings called the fontanelles exist, and intracranial pressure can be monitored noninvasively by placing a transducer
on a fontanelle as shown in Fig. 8(c). The diaphragm of the
pressure transducer has to be in coplanar alignment with the
skin surface so that the tension of the skin tissue does not
affect the measurement. When the transducer is adequately
fixed, intracranial pressure can be monitored for 24 h or more.
ADVANCED TECHNOLOGY FOR PATIENT MONITORING
There are many clinical requirements in patient monitoring
for which the present technology is still insufficient. Many
biochemical parameters are still difficult to monitor continuously, and the use of biosensors has been investigated extensively. The importance of monitoring brain function in critically ill patients has been recognized, and new technniques
are becoming available. Patient monitoring in extreme situations are required, such as in a strong electromagnetic field
during magnetic resonance imaging (MRI) examination or
during hyperthermia therapy using electromagnetic heating.
Home monitoring requires the development of monitoring systems that can be operated without the aid of medical personnel.
Monitoring Biochemical Parameters Using Biosensors
Although blood analysis is the most important clinical examination, it requires the procedure of blood sampling. Frequent
examination is required in some cases. For example, blood
glucose has to be measured many times a day in diabetic patients. If a sensor attached to the body can detect chemical
parameters in the body fluid continuously, frequent blood
sampling becomes unnecessary, and the precise control of
blood glucose will be possible, which is the goal of the artificial pancreas. Biosensors are promising for this purpose. The
biosensor is a device in which components of biological origin,
such as enzymes, antibodies, cells, or even microorganisms,
are used to analize specific chemical species selectively (27).
By immobilizing an adequate quantity of biological components on a sensor, the species to be analyzed is detected by
ordinary electrochemical, optical, or acoustic sensing principles.
However, there are many difficulties when biosensors are
used for patient monitoring. Because the skin is impermeable
for most chemical species except gas, a biosensor has to be
inserted into the body space to make direct contact with the
body fluid. To implant a sensor into the tissue, invasive procedures are required, and it is also necessary that the sensor is
nontoxic and biocompatible. When used in the blood vessels,
its surface should be anticoagulant. Once a sensor is placed
in the body, recalibration is difficult; hence long-term stability
is required. Despite such difficulties, extensive studies have
been done in in vivo chemical measurement using biosensors
(28). Although still not accepted for clinical routine use, moni-
PATIENT MONITORING
toring for fairly long periods of time has been attained. For
example, a ferrocene-mediated type of glucose sensor covered
with a newly designed biocompatible membrane could be used
for 7 days without calibration, and for 14 days with in vivo
calibration by comparison with blood sampling data (29).
As a solution for the difficulty of the in vivo application of
biosensors, ex vivo measurement has been attempted in which
a small amount of body fluid is drained from the body and
perfused through a flow-through sensor cell. An advantage of
this type of ex vivo measurement is the easiness of calibration
and replacement of the sensor. However, continuous drainage
causes loss of body fluids. Microdialysis could be a solution to
this difficulty. The microdialysis probe has a semipermeable
membrane at the tip, and a fluid is perfused through it by a
fine double-lumen catheter at a very low flow rate. When the
probe is placed in the subcutaneous tissue, small molecules,
such as glucose, are able to diffuse through the membrane
from the body fluid to the perfusion solution, and is then analyzed by a biosensor. In microdialysis, the permeability of the
membrane affects the measurement. To realize accurate measurements through the membrane without being affected by
the membrane permeability, the application of a null method
was proposed (30). In this system, the perfusion solution is
adjusted using a servocontrol system so that concentrations
at the inlet and the outlet of the probe are equal. This method
is advantageous, not only for eliminating the effect of the
change in membrane permeability but also for eliminating the
effect of drift and sensitivity change of the sensor.
As a less-invasive chemical measurement method, the effluent fluid analysis has been attempted (31). If the outermost
layer of the skin, the stratum corneum, is removed by stripping with an adhesive tape many times and negative pressure
is then applied to the skin surface, a small amount of fluid,
called effluent fluid, can be collected. It has been shown that
blood glucose can be monitored quasicontinuously using this
method.
Monitoring Brain Function
Monitoring brain function is required during anesthesia, in a
patient who lacks consciousness, and in the neonate. Electroencephalography (EEG) has been used widely for monitoring
the electrical activity of the brain. The responses of the brain
to sensory inputs such as visual and auditory stimulation can
be examined monitoring the resultant slight changes in the
EEG waveform that are known as evoked potentials. While
the amplitude of the evoked potentials is smaller than that of
ordinary EEG activity, it can be extracted by averaging many
responses. The function of the motor system can be examined
by stimulating the mortor cortex. A strong magnetic pulse can
be used for this purpose. The magnetic stimulation induces
eddy currents that cause firing of motor neurons, and visible
muscle contractions or muscular activities visualized in the
form of an electromyograms are induced if the function of the
motor system is normal (32).
Brain function is sustained by the oxygen supply through
the cerebral circulation, and thus a sufficient supply of oxygen
to the brain is of primary importance. When the blood supply
to the brain is decreased, the oxygen partial pressure in the
tissue decreases, and consequently the oxygen saturation of
venous blood returning from the brain will also decrease.
When decreasing cerebral circulation is suspected, jugular ve-
nous oxygen saturation can be measured. For continuous

monitoring, a regional oxygen saturation catheter is placed
in the jugular vein. A fiber-optic regional oxygen saturation
catheter is commercially available for this purpose (Baxter
Healthcare Corp., Edward-Critical Care, Irvine, California).
In the neonate, cerebral oxygen supply and utilization can
be monitored noninvasively using near-infrared spectroscopy.
This technique is based on the measurement of transmitted
light across the head. Even though the intensity of the transmitted light through the head is on order of 1015 of that of
the incident light, it is still detectable using a low-noise photomultiplier tube. This technique provides information about
the level not only of oxygen saturation of hemoglobin but also
of oxidized cytochrome oxidase in the brain tissue (33). Cytochrome oxidase is an enzyme that catalyzes cellular energy
production, and the amount of oxidized cytochrome oxidase is
related to cellular oxygen consumption. An instrument that
monitors oxygen saturation and oxidized cytochrome oxidase
levels is commercially available (NIRO-500, Hamamatsu Photonics Co., Hamamatsu, Japan).
Monitoring under Strong Electromagnetic Fields
Patient monitoring is required even in extreme conditions,
such as under strong electromagnetic fields. Hyperthermia
cancer therapy is a typical case of this kind. Hyperthermia
cancer therapy is based on the fact that cancer cells are
weaker than normal cells at high temperatures near the limit
of survival, that is, in the range of 42 to 45C. To treat the
malignant tumor, local heating by applying a strong radio frequency (RF) field has been used. Heating should affect cancer
cells; however, normal cells have to be protected. Therefore,
precise temperature measurement under strong RF fields is
required. Although conventional temperature sensors, such
as ordinary thermistors and thermocouples, cannot be used,
many different techniques have been attempted. For example,
fiber-optic temperature sensors have been developed for this
purpose. To convert temperature into an optical signal, several techniques have been developed. Among them are liquid
crystal, fluorescence, birefringent crystals with polarizers,
and the semiconductor band-edge absorption shifts. Some of
these are commercially available (e.g., Fluroptic Thermometer, Luxtron Corp., Mountain View, California). However,
such probes can only measure temperature at a particular
point. For precise temperature control while under RF heating, temperature distribution imaging in the strong RF field
is demanded.
Magnetic resonance imaging employs strong magnetic field
and RF pulses. The ferromagnetic material in the field is
magnetized and distorts the image. In conductive material,
eddy currents are induced and the magnetic field produced by
the induced currents also distorts the image. Radio frequency
pulses interfere with electronic instruments, so that ordinary
monitoring systems cannot be used near the scanner. For patient monitoring in such a situation, monitoring devices
should be made using nonmagnetic and nonconductive materials. In fact, there are some techniques that can be used such
as a blood pressure cuff with plastic connectors, chest wall
movement sensors for a respiratory monitor, and pneumatic
pulse monitoring using a finger cuff (34). Fiber-optic probes
would also be applicable.
PATIENT MONITORING
Patient Monitoring at Home

For chronically ill patients, the continuation of medical treatment at home would be more comfortable than to stay in a
hospital apart from their family. Medical treatment at home
is preferable not only for the patient, but it reduces medical
expenses. However, when the treatment involves frequent examinations of physiological parameters, it is difficult to perform such procedures at home in the same way as in the hospital. Even in such patients, medical treatment would be
given if adequate patient monitors were available. For example, in diabetic patients, the treatment of administering insulin can be controlled even at home by measuring blood glucose
frequently; convenient devices for blood glucose measurement
are commercially available. Oxygen therapy using oxygen delivery systems is also possible at home, but monitoring arterial oxygen saturation using a pulse oximeter is recommended
to secure the treatment.
Many kinds of home health care devices have been developed, are commercially available, and are used routinely at
home (35). To secure the function of such devices and to avoid
risks due to malfunctioning or inadequate operation of the
devices, patient monitoring would be helpful. If more reliable
monitoring systems become available, more effective treatments will become acceptable in home care.
The application of health monitoring at home is not limited
to critically ill patients. Even in apparently healthy people,
abnormalities are found quite often during screening in physiological or biochemical examinations. Early diagnoses can be
made using such data, and, consequently, each disease that
is diagnosed earlier can be treated more reliably than cases
in which a desease is diagnosed only after the appearance of
apparent symptoms. If fully automated monitoring instruments are installed in the home, and physiological parameters can be monitored without performing any tedious operations, abnormalities would be identified much easier. The
concept of home health monitoring is considered a new possibility in health care technology (36).
To realize this concept, ordinary clinical examinations such
as the ECG measurement by limb lead are rarely employed.
However, ECG can be observed automatically without any
measurement operation in specific situations. For example,
when one takes a bath in a bathtub, ECGs can be recorded
through water from electrodes attached to the inside wall of
the tub. Also, ECGs can be recorded using conductive cloth
electrodes on a bed or even from a toilet seat. Other paramaters such as body weight can be measured accurately using a
specially designed load cell installed on a toilet. Temperature
sensors installed beneath a bed sheet provides information
about the time a patient spent in the bed and about body
motion during sleep which can provide information on sleep
disturbances. Simple infrared sensors and magnetic switches
installed in each room and on appliances or water taps can
provide a fairly complete record of the daily activities of the
isolated living person. Although this kind of technology is not
yet well developed, there are many possibilities that would
allow for patient monitoring in the home.
Automated home health monitoring would be especially effective for the elderly, because physiological function declines
with age, and gradual changes are rarely recognized. The normal ranges of health parameters for each individual would be
determined accurately from long-term records so that even a
slight change in such parameters can be identified. If a longterm record of health parameters is obtained, it would be utilized for a retrospective analysis when symptoms appear, and
would be utilized not only for accurate diagnosis but also for
epidemiological studies in a population if such data are accumulated for many people. It is expected that this approach
would contribute to a reduction in the need for medical services and, consequently, a reduction in medical expenses.
BIBLIOGRAPHY
1. N. V. Thakor, Computers in electrocardiography. In J. G. Webster (ed.), Encyclopedia of Medical Devices and Instrumentation,
New York: Wiley, 1988, pp. 10401061.
2. B. D. Bertolet et al., Evaluation of a novel miniature digital ambulatory ECG transient myocardial ischemia detection system, J.
Ambulat. Monitoring, 5: 3339, 1992.
3. J. De Maso et al., Ambulatory high-resolution ECG recorder using disk storage, J. Ambulatory Monitoring, 5: 317322, 1992.
4. V. L. Gordon et al., Zero stability of disposable and reusable pressure transducers, Med. Instrum., 21: 8791, 1987.
5. J. Penas, Photoelectric measurement of blood pressure, volume
and flow in the finger, Digest of the 10th International Conference
on Medical and Biological Engineering, Dresden, 1973, p. 104.
6. K. Yamakoshi, H. Shimazu, and T. Togawa, Indirect measurement of instantaneous arterial blood pressure in the human finger by vascular unloading technique, IEEE Trans. Biomed. Eng.,
BME-27: 150155, 1980.
7. J. S. Eckerle, Arterial tonometry. In J. G. Webster (ed.), Encyclopedia of Medical Devices and Instrumentation, New York: Wiley,
1988, pp. 27702776.
8. G. Manning, S. G. Vijan, and M. W. Millar-Craig, Technical and
clinical evaluation of the Medilog ABP non-invasive blood pressure monitor, J. Ambulatory Monitoring, 7: 255264, 1994.
9. W. Ganz et al., A new technique for measurement of cardiac output by thermodilution in man, Am. J. Cardiol., 27: 392396,
1971.
10. W. G. Kubicek et al., Development and evaluation of an impedance cardiac output system, Aerospace Med., 37: 12081212,
1966.
11. T. S. Chadha et al., Validation of respiratory inductive plethysmography using different calibration procedure, Am. Rev. Respirat. Dis., 125: 644649, 1982.
12. L. E. Baker, Electrical impedance pneumography. In P. Rolfe
(ed.), Noninvasive Physiol. Meas., London: Academic Press, 1979,
pp. 6594.
13. K. Bhavani-Shanker et al., Capnometry and anesthesia, Can. J.
Anesth., 39: 617632, 1992.
14. I. E. Sodal, J. S. Clark, and G. D. Swanson, Mass spectrometers
in medical monitoring. In J. G. Webster (ed.), Encyclopedia of
Medical Devices and Instrumentation, New York: Wiley, 1988,
pp. 18481859.
15. J. W. Severinghaus and J. F. Kelleher, Recent development in
pulse oximeter, Anesthesiology, 76: 10181038, 1992.
16. A. Takatani et al., Experimental and clinical evaluation of a noninvasive reflectance pulse oximeter sensor, J. Clin. Monitoring, 8:
257266, 1992.
17. A. C. M. Dassel et al., Reflectance pulse oximetry at the forehead
of newborns: the influence of varying pressure on the probe, J.
Clin. Monit., 12: 421428, 1996.
18. A. Huch et al., Continuous transcutaneous oxygen tension measurement with a heated electrode, Scand. J. Clin. Lab. Invest.,
31: 269275, 1973.
10
PATTERN RECOGNITION
19. P. Eberhard et al., Continuous pO2 monitoring in the neonate by

skin electrodes, Med Biol. Eng., 13: 436442, 1975.
20. J. W. Severinghaus, M. Stafford, and A. F. Bradley, Tc pCO2 electrode design, calibration and temperature gradient problems,
Acta Anesthesiol. Scand. 68 (Suppl): 118122, 1978.
J. G. Webster (ed.), Encyclopedia of Medical Devices and Instrumentation, Vol. 1Vol. 4, New York: Wiley, 1988.
. O
berg, Biomedical Transducers and
T. Togawa, T. Tamura, and P. A
Instruments, Boca Raton, FL: CRC Press, 1997.
TATSUO TOGAWA
21. J. W. Severinghaus, A combined transcutaneous pO2-pCO2 electrode with electrochemical HCO3 stabilization, J. Appl. Physiol.
Respirat. Environ. Exercise Physiol., 51: 10271032, 1981.
Tokyo Medical and Dental

University
22. T. Togawa, Body temperature measurement, Clin. Phys. Physiol.

Meas., 6: 83108, 1985.
23. R. H. Fox et al., A new method for monitoring deep body temperature from the skin surface, Clin. Sci., 44: 8186, 1973.
24. T. Togawa, Non-invasive deep body temperature measurement.
In P. Rolfe (ed.), Non-invasive Physiological Measurements, London: Academic Press, 1979, pp. 261277.
25. T. Tsuji, Patient monitoring during and after open heart surgery
by an improved deep body thermometer, Med. Prog. Technol., 12:
2538, 1987.
26. M. R. Gaab and H. E. Heissler, Monitoring intracranial pressure.
In J. G. Webster (ed.), Encyclopedia of Medical Devices and Instrumentation, New York: Wiley, 1988, pp. 19611983.
27. A. P. F. Turner, I. Karube, and G. S. Wilson (eds.), Biosensors,
Fundamentals and Applications, Oxford: Oxford University
Press, 1987.
28. S. J. Alcock and A. P. F. Turner (eds.), In vivo chemical Sensors,
Recent Developments, Bedford: Cranfield Press, 1993.
29. K. Nishida et al., Development of a ferrocene-mediated needletype glucose sensor covered with newly designed biocompatible
membrane, 2-methacryloyloxyethyl phosphorylcholine-co-n-butyl
methacrylate, Med. Prog. Technol. 21: 91103, 1995.
30. T. Tamura and T. Togawa, Continuous blood pH monitoring by
use of null method, Sensors Actuators, 16: 273285, 1989.
31. N. Ito et al., A novel blood glucose monitoring system based on an
ISFET biosensor and its application to a human 75-g oral glucose
tolerance test, Sens. Actuators B1: 488490, 1990.
32. S. Ueno, T. Matsuda, and M. Fujiki, Functional mapping of the
human motor cortex obtained by focal and vectorial magnetic
stimulation of the brain, IEEE Trans. Magn. MAG-26: 1539
1544, 1990.
33. M. Cope and D. T. Delpy, System for long-term measurement of
cerebral blood and tissue oxygenation on newborn infants by near
infrared transillumination, Med. Biol. Eng. Comput. 26: 289
294, 1988.
34. G. A. Volgyesi et al., Design and evaluation of a pneumatic pulse
monitor for use during magnetic resonance imaging, J. Clin Monitoring, 7: 186188, 1991.
35. M. H. Hafner and R. A. Gottesman, Home health care devices. In
J. G. Webster (ed.), Encyclopedia of Medical Devices and Instrumentation, New York: Wiley, 1988, pp. 15091516.
36. T. Togawa et al., Attempt of fully automated home health monitoring systems, Med. Biol. Eng. Comput. 34 (Suppl. 1): Part 1,
3132, 1996.
Reading List
R. S. C. Cobbold, Transducers for Biomedical Measurements: Principles and Applications, New York: Wiley, 1974.
P. Rolfe, ed., Non-invasive Physiological Measurements, Vol. 1, London: Academic Press, 1979.
P. Rolfe, ed., Non-invasive Physiological Measurements, Vol. 2, London: Academic Press, 1983.
D. W. Hill and A. M. Dolan, Intensive Care Instrumentation, 2nd. ed.,
London: Academic Press, 1982.
PATTERN-BASED ORGANIZATION. See SELFORGANIZING FEATURE MAPS.
PHYSIOLOGICAL MODELS, DEVELOPMENT

Modeling is an important facet of engineering. Models are
simplied representations of real-world systems. For this
reason, models are also an important part of everyday life.
If you wanted to drive from Tucson, Arizona to Madison,
Wisconsin you would use a model of the highway system
called a road map (either on paper or on a computer screen).
Such a map would allow you to understand the highway
system of the United States without driving on every road.
If you wanted a model of the life of Southern aristocrats
at the time of the Civil War, you could use the book Gone
with the Wind. Highway engineers create models to see how
trafc-light synchronization or lane obstructions would affect trafc ow. Ohms law may be a good model for a resistor. F = ma is a simple model for the movement of a
baseball. Models allow us to apply mathematical tools to
real-world systems. We use models to understand things
that are big, complicated, expensive, or far away in space
or time. Certainly, physiological systems fall into some
of these categories and are often modeled in the eld of
biomedical engineering.
Figure 1 shows the relationships between models and
the real world. On the extreme left, people do experiments
on real-world systems. Baseball players often t into this
category. Over the years, there has been a lot of experimentation with the baseball bat. Most of this experimentation was illegal, because the rules say that (for professional players) the bat must be made from one solid piece
of wood. However, to make the bat heavier, George Sisler,
who played rst base for St. Louis Browns in the 1920s,
pounded Victrola phonograph needles into his bat barrel
and in the 1950s Ted Kluszewski of the Cincinnati Reds
hammered in big nails. To make the bat lighter, many players have drilled a hole in the end of the bat and lled it with
cork. Detroits Norm Cash admitted to using a corked bat
in 1961 when he won the American League batting title
by hitting .361. However, the corked bat may have had little to do with his success, because he presumably used a
corked bat the next year when he slumped to .243. Some
players have been caught publicly using doctored bats. In
1974, the bat of Graig Nettles of the Yankees shattered as
it made contact, and out bounced six Super Balls. In 1987
Houstons Billy Hatcher hit the ball and his bat split open,
spraying cork all over the ineld. These are all examples of
experimentation with no concrete models to guide them.
To lessen the waste of time and decrement of performance entailed in such experiments with altered bats, we
made mathematical models of individual humans. Then

we coupled these models to the equations of physics and
predicted the ideal bat weight for each individual (2,3,3a).
This modeling process is shown in the center box of Fig.
1. However, there is another box labeled Computer simulation of model on the right side of Fig. 1. Our model was
composed of mathematical equations that we had to solve
on a computer. If everything goes right, the digital computer simulation should produce the same results as the
mathematical equations. But care must be taken to ensure
that this is true. Things that you must worry about include
(a) the accuracy of the computer code; (b) numerical factors, such as the integration step size truncation errors,
and the integration technique (e.g., quadrature, AdamsMoulton, RungeKutta) [see Yakowitz and Szidarovszky
(4) for details]; (c) implementation considerations such as
using a commercial simulation package that is much bigger
than the model, like using a calculator to add single-digit
numbers (the point is that in some situations the unused
routines could cause problems, like overwriting areas of
memory or forcing pointers out of bounds); and (d) the possibility that the hardware is defective (How often do you
run the diagnostics on your personal computer?). In our
study, we carefully assessed each of these to see how they
would affect our predictions about the real world. In addition, matching the output of the model to the real-world
system can be a useful experimental approach to numerical validation in general.
Finally, at the extreme right side of Fig. 1 we nd pure
mathematicians working in the computer world often with
no regard to the real world. Early studies of fractals t into
this category. For more on the philosophy and practice of
modeling, see Bahill (5).
STEPS IN THE MODELING PROCESS

Figure 1. Relationship of systems and models. [Reprinted with
permission from W. L. Chapman, A. T. Bahill, and A. W. Wymore,
Engineering Modeling and Design, Boca Raton, FL: Copyright
CRC Press, 1992, p. 45 (1).]
Describe the system to be modeled

Determine the purpose of the model
Determine the level of the model
Gather experimental data describing system behavior
Investigate alternative models
Select a tool or language for the simulation
Make the model
Validate the model
Show that the model behaves like the real system
Emulate something not used in the models design
Perform a sensitivity analysis
Show interactions with other models
Integrate with models for other systems
Analyze the performance of the model
Re-evaluate and improve the model
Suggest new experiments on the real system
State your assumptions
Models can be used for many purposes:
Understand or improve an existing system or organization
Physiological Models, Development

Create a new design or system

Control a system
Suggest new experiments
Guide data collection activities
Allocate resources
Identify cost drivers
Increase return on investment
Identify bottlenecks
Help sell the product
Reduce risk
Modeling is not a serial process; some steps can be done

in parallel and it is very iterative. This prescription for
describing processes was developed by Bahill and Gissing
(6).
Frameworks help people organize and assess completeness of integrated models of their enterprises. Several popular frameworks have been used to architect enterprises.
The Zachman framework, like many others, considers multiple perspectives and multiple aspects of an enterprise
(Bahill, Botta and Daniels (6a)).
DESCRIBE THE SYSTEM TO BE MODELED
The control of movement has long been an enigma for scientists as well as for parents who marvel at the miracle of
seeing their children take their rst steps. The control of
muscles that we often take for granted is so complex that
it is difcult to comprehend the intricacies involved. To develop an understanding of such complex movement control
systems, we started with a study of a simple neuromuscular system, developed physiologically realistic models, and
then rened these models.
The eye movement system is a good starting point because of its simplicity, relative to other neuromuscular systems. This system has primarily two degrees of freedom,
namely, horizontal and vertical; and only two muscles are
involved in horizontal eye movements, as compared with
six or more degrees of freedom and about 30 major muscles for each leg involved in locomotion. The study of the
eye movement system is also aided by the ease with which
the movements can be measured. Any knowledge gained
in the control of eye movements will contribute not only
to the understanding of the oculomotor system but also to
the understanding of larger, more complex neuromuscular
systems.
The purpose of the eye movement system is to keep
the fovea, the region of the retina with the greatest visual
acuity in daylight, on the object of interest. To accomplish
this task, the following four types of eye movements work
in harmony: saccadic eye movements, which are used in
reading text or scanning a roomful of people; vestibuloocular eye movements, used to maintain xation during
head movements; vergence eye movements, used when looking between near and far objects; and smooth-pursuit eye
movements, used when tracking a moving object. These four
types of eye movements have four independent control systems, involving different areas of the brain. Their dynamic
properties, such as latency, speed, and bandwidth, are different, and they are affected differently by fatigue, drugs,
and disease.
The specic actions of these four systems can be illustrated by the example of a duck hunter sitting in a rowboat
on a lake. He scans the sky using saccadic eye movements,
jerking his eyes quickly from one xation point to the next.
When he sees a duck, he tracks it using smooth-pursuit eye
movements. If the duck lands in the water right next to his
boat, he moves his eyes toward each other with vergence
eye movements. Throughout all this, he uses vestibuloocular eye movements to compensate for the movement of
his head caused by the rocking of the boat. Thus, all four
systems are continually used to move the eyes.
This section is primarily about developing and validating a model for the human smooth-pursuit eye movement
system. Other systems are only included when they interact with the smooth-pursuit system.
The purpose of this model is to help understand the human smooth pursuit eye movement system. The level of
the model is that of eye movements from a few minutes of
arc to a few dozen degrees, of speeds up to 30 degrees per
second, and for durations up to 20 seconds.
GATHER EXPERIMENTAL DATA DESCRIBING SYSTEM

BEHAVIOR
Experiments with transient target waveforms reveal a
150 millisecond (msec) time delay in the human smoothpursuit eye movement system (7). The effects of this time
delay are apparent during starting and stopping transients, as shown in Fig. 2(8). However, when a human (or
a monkey) tracks a target that is moving sinusoidally, the
subject quickly locks onto the target and tracks with neither latency nor phase lag. It is as if the subject creates an
internal model of the target movement and uses this model
to help track the target. This internal model has been called
a predictor (9-11), a long-term learning process (12), a percept tracker (1316), a neural motor pattern generator (17)
and a target-selective adaptive controller (8,18-20).
The Sinusoidal Target Waveform

The sinusoid is the most common smooth-pursuit target
waveform because it is easy to generate and easy to track.
Our sinusoidal target waveform is given by r(t) = A sin t.
Our normal amplitude, A, was 5 (i.e., 5 from primary
position).
Figure 2 (top) shows tracking of the beginning of sinusoidal movement. Smooth pursuit began 150 ms after
the target started to move. It was followed by a corrective saccade at 200 ms and then by zero-latency, unitygain tracking. Figure 2 also shows a termination of sinusoidal smooth-pursuit tracking. The smooth-pursuit velocity started declining 120 ms after the target velocity
dropped. It reached zero velocity at 220 ms, when a corrective saccade occurred to end the subjects tracking. Thus,
the beginning and ending transients show the effects of the
time delay, whereas steady-state tracking does not.
Figure 2. Typical beginning (top) and ending (bottom) of sinusoidal tracking. When the target (dashed line) started there was
a 150 ms delay before the eye velocity increased; when the target
stopped there was a 120 ms delay before the eye velocity began to
decrease. Target movements were 5 from primary position. The
time axis is labeled in seconds, and upward deections represent
rightward movements. [Reprinted from A. T. Bahill and J. D. McDonald, Smooth pursuit eye movements in response to predictable
target motions, Vision Res., 23: 15731583, 1983, with permission
from Elsevier Science. (8).]
The Cubic Target Waveform

Humans can overcome a large internal time delay and
track sinusoidal target waveforms with unity gain and no
time delay. Moreover, they learn to do this very quickly.
To help determine if humans can easily track every predictable waveform, we created a cubic waveform. The cubic
waveform is simple, but we could not imagine a naturally
occurring cubic visual target.
We used the cubical target waveform (shown in Fig. 3)
because no naturally occurring visual targets move with a
periodic cubical waveform; thus, our results were not inuenced by previous learning; yet, the cubical waveform resembles a sinusoid so it should be possible to track. The cubical target waveform was formed with the following thirdorder polynomial:
where T represents target period and A is the amplitude.

Previous studies have shown that humans track well when
the target has an amplitude of 5 and a frequency of about
0.32 Hz, so these values were used in our experiments. The
target always started with zero phase and zero offset. No
warning was given before the target started. Another benet of the cubic waveform is that it looks like a sinusoid
but the velocity is strikingly different. Therefore, by analyzing the eye velocity records we could tell if the subject
had really learned the cubical waveform or if he had merely
approximated it with a sinusoid.
Figure 3 shows excellent tracking of the cubical target
waveform. Using only smooth-pursuit eye movements, the
subject was able to keep the fovea on the target for over 8 s.
Saccades were not removed or ltered out of the eye position traces; indeed small conjugate saccades can be seen at
the 8.5 s mark.
Figure 3. Binocular eye movements for good tracking of the cubical target waveform. [From D. E. McHugh and A. T. Bahill, Learning to track predictable target waveforms without a time delay,
Invest. Ophthalmol. Vis. Sci., 26: 933, 1985. (21).]
Learning to Track the Cubical Waveform

Figure 3 shows that a human can track the cubic target
waveform very well. But this capability is not inherent.
It must be learned. Our standard learning protocol began with a 6 s square-wave calibration target waveform,
followed by 9 s of the cubical target waveform, 3 s of the
square-wave target waveform, another 9 s of the cubical
waveform, and nally another 6 s of the square-wave calibration target waveform. The subjects were allowed to rest
for 5 min and then the sequence was repeated. This process
continued for about 2 h.
Because the purpose of the eye movement control system is to keep the fovea on the target, we felt that the error
between the eye and the target was the most appropriate measure of the quality of tracking. Our primary metric
therefore was the mean-square error (MSE) between eye
position and target position. The human fovea (specically
the inner foveal pit) has a radius of 0.5 (17, 22). Therefore, a target consistently on the outer edge of the fovea
produces an MSE of 0.252 .
Single and double exponential curves were t to the
MSE data. The best t was usually an exponential of the
form MSE = AeBt + C. The solid lines of Fig. 4 show the
exponential curves t to the data of our four best-tracking
college students. We were trying to quantify the ultimate
capabilities of the human smooth-pursuit system, so we
only report the performance of our best subjects. In this
gure, we only show data of four of 20 college students. The
other students did not demonstrate such low-error tracking.
To narrow in on this exquisite tracking performance we
decided to study optimal humans performing optimally.
Who is an optimal human? For eye tracking capability, we
Figure 4. Time course of learning for seven subjects. Solid lines

are the exponential curves t to the data of our four best-tracking
college students. Circles, asterisks, and squares are data points for
three professional baseball players. [From D. E. McHugh and A.
T. Bahill, Learning to track predictable target waveforms without
a time delay, Invest. Ophthalmol. Vis. Sci., 26: 935, 1985. (21).]
thought professional athletes would t the bill. So we invited some professional baseball players to participate in
our experiments. The MSEs for three members of the Pittsburgh Pirates Baseball Club are represented by circles, asterisks, and squares in Fig. 4. In viewing the target for the
rst time, professional baseball players 1 and 2 had much
smaller MSEs (0.05 and 0.08) than our other subjects. They
had never seen a cubical waveform before, yet they started
out with low MSEs. Baseball players 1 and 2 played in the
major leagues for over 10 years. Player number 3 never
got out of the class A Farm System. These data seem to
indicate that the ability to track the cubical waveform is
correlated with the ability to hit a baseball.
DEVELOPING THE MODEL
An important decision in making a model is determining
its architecture. Some architectural decisions that must
be made are whether it should be state-based or memoryless. Botta, Bahill and Bahill (22a) offer recommendations
to help make this determination. In this case, we decided
that the smooth pursuit system was a stated-based system.
(The state dynamics are modeled with an integrator K/(s
+ 1)). Next, choosing between continuous and discrete, we
decided the smooth pursuit system was discrete. To complete the high-level architecture we choose a closed-loop
feedback control system.
Most physiological systems are closed-loop negativefeedback control systems. For example, consider someone
trying to touch his or her nose with a nger. He or she would
command a new reference position and let the arm start to
move. But before long, sensory information from the visual
and kinesthetic systems would signal the actual nger position. This sensory feedback signal would be compared to
the reference or command signal to create the error signal
that drives the arm to the commanded output position.
In the analysis of such systems, it is difcult to see which

effects in the output are due to elements in the forward
path and which are due to sensory feedback. In order to
understand the contribution of each element, it is necessary to open the loop on the systemthat is, to remove the
effects of feedback. For some systems it is easy to open the
feedback loop, while for others it is exceedingly difcult
since some systems have multiple or even unknown feedback loops. It is easy to open the loop on the human eye
movement system.
Many investigators have studied the human smoothpursuit eye movement system under open-loop conditions;
these studies have helped us understand this system. However, some investigators reported varied and inconsistent
responses; they found open-loop responses idiosyncratic. It
is suggested that the reason for these difculties is that
physiological systems, unlike man-made feedback control
systems, are capable of changing their control strategy
when the control loop is opened. Several specic changes in
eye movement control strategy are shown in this section.
Although the specic system studied was the eye movement system, the technique presented should generalize
to other physiological systems.
Opening the Loop on a System
A linear system can be schematically represented as a
closed-loop system, as shown in Fig. 5(a). In this gure,
R represents the reference input, and Y is the output. The
output is measured with a transducer, H, and the resulting
signal is subtracted from the input to yield the error signal,
E. In many systems (such as the oculomotor systems), the
element in the feedback loop, H, is unity; therefore the output is compared directly with the input, which explains the
reason for calling the resultant the error. This error signal
is the input for the main part of the system represented by
G. This is called a closed-loop system because of the closed
loop formed by G, H, and the summer. This system can be
redrawn as shown in Fig. 5(b). Although the transfer function of this equivalent system describes the inputoutput
relationship of the system, it is not very useful for modeling
physiological systems because it hides specic behavior by
lumping everything into one box. On the other hand, important information about the systems performance can
be gained by techniques that examine components within
the loop. One such technique for studying a system is to
open the loop, as shown in Fig. 5(c), and then study the
response of this open-looped system. The open-loop transfer function is dened as the total effect encountered by a
signal as it travels once around the loop. That is, the openloop transfer function is Gol = GH.
Note that this is not the inputoutput transfer function
of the system with its loop opened (which would be G), nor is
this the transfer function of the equivalent redrawn closedloop system shown in Fig. 5(b). When we open the loop on
a closed-loop system, bizarre behavior often results. In response to a step disturbance, a closed-loop system with its
loop opened will usually vary its output until it is driven
out of its normal operating range. For instance, if R in Fig.
5(c) is a step and G is a pure integrator, the error will be constant and the output will increase until the system reaches
is easy and some problems must be overcome. In many

systems, the difculties lie in trying to isolate one system
so that others do not interfere, as in the previously mentioned pupillary and motor control systems. In other cases,
the difculty lies in opening the loop on the system. For example, if the output of the respiratory system is dened to
be the ventilation rate, then one could study the open-loop
behavior of the system by controlling the concentration of
the gases being breathed while monitoring the ventilation
rate. However, when modeling a different aspect of this system, such that a different quantity is dened as the output,
opening the loop would become difcult: For example, controlling the venous concentration of CO2 would be difcult.
Physiological systems often have several parallel feedback loops (e.g., hormonal and neural) acting simultaneously. One of the greatest challenges in studying a physiological control system is that one may not even be aware
of all the feedback pathways.
Opening the Loop on the Eye Movement Control System
Figure 5. (a) A closed-loop feedback control system, (b) an equivalent representation, and (c) the closed-loop system with its loop
opened. Many analysis techniques require the study of the openlooped system of (c). [From A. T. Bahill, Bioengineering: Biomedical, Medical and Clinical Engineering, 1981, p. 215. Reprinted
with permission of Prentice-Hall, Upper Saddle River, NJ. (Ref.
5.]
its limit of linearity.

Often the success of a systems analysis depends on being able to open the loop on a system. If it is an electrical
circuit, one might merely cut a wire. However, if it is a human physiological system, such an approach is not feasible,
and other techniques must be developed. Such techniques
usually involve manipulating the variable normally controlled by the system, so that the feedback is ineffective in
changing the error signal. For example, in the physiological sciences, some of the earliest examples of opening the
loop are the voltage clamp technique developed by Marmount (23) and Cole (24), wherein they measured the voltage and injected current to keep the voltage constant, and
the light modulation technique used by Stark to study the
human pupil (25). In the voltage clamp technique the experimenters xed the voltage across a neuronal membrane,
the parameter that is normally controlled by the neuron:
Although it struggled to open and close the ionic channels,
the neuron could not regulate the membrane voltage, and
therefore the loop was opened. In the case of the pupil of
the eye, the experimenters controlled the amount of light
falling on the retina: Although it struggled to open and
close the pupil, the pupillary system could not control the
amount of light falling on the retina, and thus the loop
was opened. Similarly, the use of force and length servos
in research on motor systems provides a means of examining components within feedback loops, although setting up
these studies is complicated by the multiplicity of feedback
loops in these systems (see, for example, 26).
We think these open-loop techniques can be used in a
broad range of physiological systems. Of course, nothing
An easy way to open the loop on the eye movement system

is to stabilize an object on the retina. This can be done, for
example, by looking a few degrees to the side of a camera
when someone triggers a ash. There will be an afterimage
a few degrees off your fovea. Try to look at the afterimage:
You will make a saccade of a few degrees, but the image (being xed on the retina) will also move a few degrees. You
will then make another saccade, and the image will move
again. Thus, no matter how you move your eye, you cannot
eliminate the error and put the image on your fovea. This
is the same effect as if someone opened the loop on an electronic system by cutting a wire [as in Fig. 5(c)]. Therefore,
this is a way of opening the loop on the eye movement system. There is also another simple way to study open-loop
saccadic behavior. Gaze at the blue sky on a sunny day and
try to track your oaters (sloughed collagen bers in the
vitreous humor). These hair-like images move when the
eye moves; therefore your initial saccades will not succeed
in getting them on the fovea. However, with a little practice, one can learn to manipulate these images, because
they are not xed on the retina and a human can rapidly
learn to manipulate the system. This latter point often confounds attempts to open the loop on a physiological system.
When the experimenter attempts to open the loop, the human quickly changes control strategy, thereby altering the
system under study.
The most common experimental technique for opening
the loop on the eye movement system, pioneered by Young
and Stark (27), employs electronic feedback as shown in
Fig. 6. In operation, the target is given a small step displacement, say 2 degrees to the right. After about 200 ms,
the eyes saccade 2 to the right. During this movement, the
target is moved 2 farther to the right, so that at the end of
the saccade the target is still 2 to the right. After another
200 ms delay, the eyes saccade another 2 to the right, and
the target is moved another 2 , thereby maintaining the
2 retinal error. The saccadic eye movements are not effective in changing the retinal error; therefore, the loop has
been opened. In this open-loop experiment, the subject produces a staircase of 2 saccades about 200 ms apart, until
Results of Open-Loop Experiments on the Smooth-Pursuit

System
Figure 6. Electronic technique for opening the loop on the human

eye movement system. The position of the eye, E , is continuously
measured and is summated with the input target signal, T . For
the eye movement system H = 1, because if the eye moves 10 , the
image on the retina also moves 10 . If the eye movement monitor
and associated electronics are carefully designed so that H = 1,
then any change in actual eye position, E , is exactly canceled by
. Thus the error signal,
the change in measured eye position, E
E, is equal to the target signal. This is the same effect as if the
feedback loop had been cut, as in Fig. 5(c). The target position
in space (TPS) is the sum of the input signal and the measured
eye position; care must be taken to keep this position within the
linear range of the eye movement monitor. [From A. T. Bahill and
D. R. Harvey, Open-loop experiments for modeling the human eye
movement system, IEEE Trans. Syst. Man Cybern., SMC-15: 241,
1986 IEEE (30).]
Figure 7. Position of the target and eye as functions of time for

typical human open-loop tracking. After the feedback loop was
opened, and the 1-sec mark, the subject made a series of saccades
trying to catch the target. When this strategy did not work, he
seemed to turn off the saccadic system and produce only smoothpursuit eye movements. This subject was experienced in oculomotor experiments. The large open-loop gain appears to be a characteristic of such experienced subjects. [From A. T. Bahill and D.
R. Harvey, Open-loop experiments for modeling the human eye
movement system, IEEE Trans. Syst. Man Cybern., SMC-15: 248,
1986 IEEE (30).]
the measuring system becomes nonlinear. Such a staircase

of saccades is shown in the beginning of Fig. 7.
Electronic feedback has also been used to open the loop
on the smooth-pursuit system. In these experiments, the
target was moved sinusoidally. When the eye moved attempting to track the target, the measured eye position
signal was added to the sinusoidally moved target position
(as shown in Fig. 6). Thus the eye movements became ineffective in correcting the retinal error and the feedback
loop was, in essence, opened. In contrast to open-loop saccadic experiments, open-loop smooth-pursuit experiments
do not stabilize the image on the retina, but rather the target is moved across the retina in a controlled manner by
the experimenter. This is done because the saccadic system
is a position tracking system and retinal position must be
controlled, whereas the smooth-pursuit system is a velocity
tracking system and retinal velocity must be controlled.
Open-loop experiments should provide results that not

only describe the characteristics of elements within the
feedback loop, but also provide a description of the systems
performance under closed-loop conditions. Consequently,
similarity of actual closed-loop behavior with that predicted from open-loop data is indication of the success of
the investigation. Such agreement has been found by Wyatt
and Pola (28, 29) in experiments in which subjects tracked
sinusoidal waveforms. Although idiosyncratic differences
were found between their subjects, agreement was found
between actual and predicted closed-loop behavior for individual subjects. However, subsequent investigators were
not able to replicate their results (16). And in other studies
(30, 31), individualistic behavior was varied enough to obviate any meaningful description of the system using such
data.
Several factors can be identied that possibly contribute
to the differences between individual subjects and between
different experiments. One such factor is the predictability of the target waveform used in testing. While Wyatt
and Pola (28) used predictable sinusoidal waveforms and
obtained consistent results, Collewijn and Tamminga (31)
used a pseudorandom mixture of sinusoids and found great
variability between subjects. However, sinusoids were also
used by Harvey (30) with inconsistent results between subjects. Another factor may be the inuence of prior experience on subject performance. Examining the results from
several studies (2831) reveals that open-loop gains are
larger in subjects with more experience in laboratory tracking tasks.
The one common element shared by these studies is intersubject variability, although the magnitude of this variability differed considerably. It is noteworthy that such
variability is found not only between subjects, but also in
the performance of individual subjects in single trials. Such
variation has been observed by Harvey (30) and also by
Leigh et al. (32), in a subject in which open-loop behavior
was observed by presenting a visual target to a patients
paralyzed eye while monitoring the motion of the normal,
but covered, eye. Each subjects performance also depends
on the instructions given to the subject (33). These ndings
show that the variability inherent in open-loop studies is
attributable not only to differences between subjects but
also to time-varying performance of individual subjects.
Comparing Open-Loop Experiments with Simulations
Insight into the behavior of the smooth-pursuit system
under open-loop conditions was sought by Harvey (30)
through a comparison of experimental results with those
from simulations. The simulations were performed using
the target-selective adaptive control (TSAC) model (19)
shown in Fig. 8. This model has three branches. The top
branch, the saccadic branch, generates a saccade after a
short delay whenever the disparity between target and
eye position is too great. The middle branch, the smoothpursuit branch, produces smooth tracking of moving targets. The input to the smooth-pursuit branch is velocity,
so the rst box (labeled smooth-pursuit processing) con-
Figure 8. The general form of the TSAC model. [From A.

T. Bahill and T. M. Hamm, Using open-loop experiments
to study physiological systems with examples from the human eye movement systems, News Physiol. Sci., 4: 107, 1989
(56).]
tains a differentiator and a limiter. The box labeled smoothpursuit controller and dynamics contains a rst-order lag
(called a leaky integrator), a gain element, a time delay, a
saturation element, and an integrator to change the velocity signals into the position signals used by the extraocular motor system. The bottom branch contains the targetselective adaptive controller that identies and evaluates
target motion and synthesizes an adaptive signal, Rc , that
is fed to the smooth-pursuit branch. This signal permits
zero-latency tracking of predictable visual targets, which
the human subject can do, despite the time delays present
in the oculomotor system. The adaptive controller must be
able to predict future target velocity, and it must know and
compensate for the dynamics of the rest of the system. The
adaptive controller is separate from the smooth-pursuit
system in the model and also in the brain (11). The adaptive
controller sends signals to the smooth-pursuit system and
also other movement systems (34). All of these branches
send their signals to the extraocular motor system, consisting of motoneurons, muscles, the globe, ligaments, and
orbital tissues. And of course, the nal component of the
model is a unity gain feedback loop that subtracts eye position from target position to provide the error signals that
drive the system. The solid lines in this gure are signal
pathways, while the dashed lines are control pathways. For
instance, the dashed line between the saccadic controller
and the smooth-pursuit controller carries the command to
turn off integration of retinal error velocity during a saccade.
In the experiments, many different target waveforms
are used. The step target was presented to the subject to
verify that the technique of opening the loop using electronic feedback was working. Because the step target introduced a position error rather than a velocity error, this
experiment opened the loop on the saccadic system rather
than the pursuit system. A position error with the feedback
loop opened should have elicited a staircase of saccades. If
this expected open-loop response to the step target was
seen, then the electronic feedback was opening the loop
correctly, as between 1.5 s and 2.5 s of Fig. 7.
There was difculty in getting consistent results for sinusoids with the loop opened. The most consistent results
obtained for such presentations came from the rst few seconds after the loop has been opened. This nding suggests
that the difculties with open-loop sinusoids were proba-
bly due to the involvement of high-level processes such as

adaptation. Once the loop was opened, the behavior of the
target changed. Often the subjects would appear to respond
to this change in target behavior by changing their tracking
strategies. Figure 7 shows a presumed example of such a
change in human tracking strategy. Between 1.5 s and 2.5 s
of this record the subject behaved as one would expect for
a subject tracking an open-loop target; there is a saccade
every 200 ms (approximately the time delay before the saccadic system responds to a position error). However, in the
middle of the record, the saccades cease; it seems that the
subject turned off the saccadic system. Such saccade free
tracking was common in these experiments and in other
open-loop experiments (16,28,29,32,33,35). The records are
strikingly devoid of saccades in spite of the large position
errors, a nding that, oddly, received little comment by previous investigators (except for 33), although it is often seen
in the data.
By way of comparison, the model is shown tracking a sinusoid under open-loop conditions in Fig. 9. To simulate the
changes in strategy that are apparent in the human data
of Fig. 7, the model characteristics were changed at intervals. From 2 to 4.25 s there is normal closed-loop tracking. At 4.25 s the loop was opened, the adaptive controller
was turned off, and the smooth-pursuit gain was reduced
to 0.7, thus producing a staircase of saccades similar to
those shown in Fig. 7. At 7.25 s the saccadic system was
turned off, the adaptive controller was turned back on, and
the gain of the smooth-pursuit system was returned to its
normal value; the model tracked with an offset similar to
that of Fig. 7. This type of position offset was often noticed
in human subjects during open-loop tracking. Finally, at
10.5 s the adaptive controller was turned off and the model
tracked without an offset but with a time delay as was seen
in some subjects.
These simulations help explain some confusing data in
the literature by allowing us to suggest that when the loop
on the human smooth-pursuit system is opened, subjects
alter their tracking strategy to cope with altered target
behavior. Some subjects continue to track with all systems (producing a staircase of saccades), some turn off the
saccadic system (producing smooth tracking with an offset), some also turn off the adaptive controller (producing
smooth tracking without an offset), and some change the
gain on the smooth-pursuit system. Thus, each subject ap-
Figure 9. Position of the target (dashed) and model (solid) as

functions of time under a variety of conditions. At the rst arrow, the loop was opened, at the second arrow the saccadic system
was turned off, and at the third arrow the adaptive controller was
turned off. Tracking patterns similar to each of these are common
in human records. [From A. T. Bahill and D. R. Harvey, Open-loop
experiments for modeling the human eye movement system, IEEE
Trans. Syst. Man Cybern., SMC-15: 249, 1986 IEEE (30).]
pears to adapt to the novel tracking task created by opening

the loop by selecting subcomponents of the smooth-pursuit
system and/or changing parameters within those subsystems. All these strategy changes are within the possibilities
provided by the model.
To eliminate these changes in strategy, recent studies
of open-loop smooth pursuit tracking only use the rst
140 msec of target movement (35a).
Such plasticity is common in physiological systems.
Repetitive stimulation of the vergence eye movement system indicates that that the speed of an individual movement depends on the size of the preceding target movement
(37a). Multifaceted control is also common in other physiological systems (see, for example, 36 and 37). Thus, the
potential exists in other physiological control systems for
changes in strategythat is, a change in the balance of control subsystems in different physiological states whether
these states occur naturally or are imposed by an investigator. Such changes may occur in different behavioral
states as observed, for example, for respiratory control in
the newborn (38). Consequently, it should not be surprising that when an investigator attempts to open the loop on
a control system the control strategy changes. This section
demonstrates this principle for the eye movement system.
The technique of opening the loop on a physiological
system in order to better understand its behavior is very
powerful as long as care is taken to acknowledge that the
human is a complex organism and is likely to change its
behavior when the input changes its behavior.
MAKE THE MODEL
The human can overcome a time delay and track visual
targets with zero latency. This is nicely demonstrated by
the smooth-pursuit eye movement system. We found that
if our model was to emulate the human, it had to predict
target velocity and compensate for system dynamics. The
model accomplished this using a prediction algorithm. To
help validate the model, a sensitivity analysis and a parameter estimation study were performed.
Figure 10 shows our model for the human eye movement systems. Like the human, this model can overcome
the time delay and track a target without latency. To do
this, the model must be able to predict future target velocity and compensate for system dynamics. In this section, a
least-mean-square technique for predicting target velocity
is described. After incorporating this prediction algorithm

into the model, the model was studied to learn more about
the model, and hopefully about the human. In particular,
we performed a sensitivity and analysis of the predictor
and then investigated how parameter variations affected
the MSE between the predicted output and the actual target waveform.
The TSAC Model
This section primarily examines the smooth-pursuit eye
movement system. The earliest model for the smoothpursuit system is the sampled data model developed by
Young and Stark (27). Because of later evidence presented
by Robinson (39) and Brodkey and Stark (40), the pursuit
branch is no longer viewed as a sampled data system, but
rather as a continuous one.
There is one physically realizable model capable of overcoming the time delay in the smooth-pursuit branch, the
TSAC model developed by McDonald (18, 19). This model
with the saccadic and smooth-pursuit branches and their
interactions is shown in Fig. 10. The computer simulation
that implements this model was written in the C language
on a Unix computer system.
Referring to Fig. 10, the input to the smooth-pursuit
branch is retinal error, which is converted to velocity by the
differentiator. The limiter prevents any velocities greater
than 70 /s from going through this branch. [The numbers
given in this section are only typical values, and the standard deviations are largefor example, and LaRitz (41)
showed smooth-pursuit velocities of 130 /s for a baseball
player.] The leaky integrator K/(ts + 1) is suggested from
(a) experimental results showing that humans can track
ramps with zero steady-state error (7) and (b) open-loop
experiments that showed a slope of 20 dB per decade for
the pursuit branchs frequency response (30). The gain, K,
for the pursuit branch must be greater than unity, since
the closed-loop gain is almost unity. Currently used values
for the gain are between 2 and 4 (30, 42). The esT term
represents the time delay, or latency, between the start of
the target movement and the beginning of pursuit movement by the subject. A time delay of 150 ms is currently accepted (30, 33). The saturation element prevents the output
of any velocities greater than 60 /s, the maximum velocity
produced by most human smooth-pursuit systems.
The model must be able to overcome the 150 ms time
delay and track with zero latency. Because the smoothpursuit system is a closed-loop system, the models time
delay appears in the numerator and the denominator of
the closed-loop transfer function,
An adaptive predictor using adaptive lters was designed

to overcome the time delay in the numerator. Compensation for the models dynamics overcomes the time delay in
the denominator.
We used several different techniques for predicting target velocity. There are many more to choose from (see the
Adaptive lters and Filtering theory sections of this
encyclopedia). Now we will make a detailed presentation of
Figure 10. McDonalds TSAC model has

three branches: smooth pursuit, saccadic,
and the adaptive predictor.
one of these prediction techniques. The nonmathematical

reader may skip this section (all the way to VALIDATE
THE MODEL) without loss of continuity.
The Least-Mean-Square Adaptive Filter
The least-mean-square (LMS) adaptive lter (4345), is a
self-designing lter composed of a tapped delay line, variable weights, a summing junction to add the weighted signals, and machinery to adjust the weights. Two processes
occur in the adaptive lter: the adaptation process and the
operation process.
The adaptation process handles weight adjustment. The
values of the weights are determined by estimating the
statistical characteristics of the input and output signals.
The heart of the adaptation process is the weight adjustment algorithm. As each new input sample is received, the
weights are updated by the algorithm,
where
W(j + 1) the weight vector after adaptation
W(j) the weight vector before adaptation
ks the proportionality constant controlling stability and
the rate of convergence
E(j) the difference between the desired response and the
lters output, the error
X(j) the vector of input signals
[E2 (j)] the gradient of the error squared with respect to
W(j)
In order to nd the best possible weights, we computed
the gradient (with respect to W) of the squared error, set
this equal to zero, and solved for the optimum weights. The
result is the WeinerHopf equation:
where
WLMS the vector of weights that would give the LMS error
(x, x) autocorrelation matrix of the input signals
(x, d) covariance matrix between the input signal and the
desired output signal
Figure 11. Implementation of the weight adjustment algorithm.

[From D. R. Harvey and A. T. Bahill, Development and sensitivity analysis of adaptive predictor for human eye movement
model, Transaction of the Society for Computer Simulation, December 1985. 1985 by Simulation Councils, Inc., San Diego, CA.
Reprinted by permission. (20).]
To solve the WienerHopf equation it is necessary to

compute the correlation matrices. However, this would require a lot of computer time; furthermore, these matrices
cannot be computed in advance, because this would require
a priori knowledge of the statistics of the input signal.
Because it is difcult to compute the true gradient,
we use an estimate of the gradient, which is equal to
2E(j)X(j). Our algorithm is a form of the method of steepest descent using estimated gradients instead of measured
gradients. Using this estimated gradient, the adjustment
algorithm can be written as
Figure 11 illustrates the implementation of this weight

adjustment algorithm. If the input signals are uncorrelated, then the expected value of the estimated gradient
converges to the true gradient without any knowledge of
the input signals statistics.
During the operation process of the LMS lter, illustrated in Fig. 12, the tapped delay-line input signals are
weighted, using the gains from the adaptation process and
summed to form the output signal. The difference between
the desired output signal and the actual output of the lter is the error that is fed back to the weight adjustment
algorithm.
The speed and accuracy of the lter while converging to
the optimal solution depends on several factors. Because
noise is introduced into the weight vector from the gradient estimates, it follows that if the lter is allowed to
10
Figure 12. The LMS adaptive lter. The boxes labeled Weight
adjustment contain systems like Fig. 11. [From D. R. Harvey and
A. T. Bahill, Development and sensitivity analysis of adaptive predictor for human eye movement model, Transaction of the Society
for Computer Simulation, December 1985. 1985 by Simulation
Councils, Inc., San Diego, CA. Reprinted by permission. (20).]
converge slowly, less noise will be introduced during each

adaptation cycle and the convergence will be smoother. Regardless of the speed with which the lter converges, some
noise will be introduced. This noise prevents the lter from
converging to the minimum MSE. The ratio of the excess
MSE to the minimum MSE gives a measure of the misadjustment of the lter compared to the optimum system. The
misadjustment depends on the time constant of the lters
weights, where the time constant is dened as the time it
takes for the weights to fall within 2% of their converged
value. A good approximate formula for the misadjustment,
M, is
This algorithm shows that M is proportional to the number

of weights, n, and inversely proportional to the time constant, MSE . The time constant MSE can be measured experimentally for each simulation. However, we would prefer to
nd an analytical way to estimate it. We can do that as follows.
To ensure convergence the proportionality constant, ks ,
in the weight adjustment algorithm must be within the
following bounds:
where E[X2j ] is the expected value of the square of the

jth input. For slow and precise convergence, ks should be
within the following more restrictive bounds:
According to Widrow (43, 44), for a lter using tapped delayline input signals, the time constant is related to the proportionality constant by
Figure 13. The adaptive predictor. The boxes labeled Adaptive

lter and Slave lter contain systems similar to those in Fig.
12. [From D. R. Harvey and A. T. Bahill, Development and sensitivity analysis of adaptive predictor for human eye movement
In summary, an adaptive lter is made up of a tapped delay

line, variable weights, a summing junction, and the weight
adjustment algorithm. The lter adjusts its own internal
settings to converge to the optimal solution. Due to noise
from the gradient estimate, the accuracy and speed of convergence depends on the number of weights and the proportionality constant, ks .
The Adaptive Predictor
The adaptive predictor is an application of the LMS adaptive lter. We used this predictor to overcome the 150 msec
time delay in the smooth-pursuit model.
Figure 13 shows the design of the adaptive predictor.
Two lters are used: an adaptive lter and a slave lter. The
adaptive lter determines the appropriate weights. It does
this by predicting the value of the input signal 150 ms into
j + T ). To accomplish this, the input signal,
the future, D(
D(j), is delayed by an amount of time equal to the time to
be predicted, in this case 150 ms. This delayed signal, X(j),
then serves as the input to the adaptive lter. The lters
weights converge to values that give an output signal, Y(j)
j), which ideally matches the undelayed input signal.
or D(
The slave lter is responsible for predicting. The input
to the slave lter is the undelayed signal, D(j). The slave
lter is organized like the adaptive lter except there is no
automatic adaptation processthat is, no weight adjustment. The weights from the adaptive lter are copied into
the slave lter after each adaptation cycle. The output of
j + T ), is the predicted value of the input
the slave lter, D(
signal at the desired future time.
For the TSAC model, the velocity of the target must be
predicted 150 ms into the future to overcome the smooth
pursuit systems time delay. Therefore, the targets velocity, the input signal to the predictor, was delayed by 150 ms
and used as the adaptive lters input. Our adaptive lter used between 15 and 150 weights and a proportionality constant of 0.00001. Figures 14 and Figure 15 show
the behavior of the predictor with 150 weights. 14 show
the behavior of the predictor with 150 weights. Figure 14
j + T ), for various tarshows the output of the predictor, D(
get waveforms superimposed on the signal to be predicted.
11
The predicted target velocity from the adaptive predictor compensates for the effects of the time delay in the numerator of the transfer function of Eq. (6). To overcome the
effects of the time delay in the denominator, compensation
for the models dynamics must be done. This means that
the brain must have a model for itself and the rest of the
physiological system, and that it uses this model to generate the required compensation signal.
Compensating for Plant Dynamics
When linear state-variable feedback notation is used for a
system, its closed-loop transfer function is
where
Figure 14. The predictors output superimposed on the signal it is

predicting for four different target velocity waveforms, which are,
from top to bottom: parabolic, triangular, sinusoidal, and square.
Y system output, in Fig. 10

Ri system input
T time delay
A system matrix
b input coefcient vector

vector transpose operation

k transposed control vector

h transposed output coefcient vector
K the gain
The general method of compensating for model dynamics is complex. It involves computing an adaptive signal
Ra , which, when added to the target position Rs , produces
a system input Ri that will produce zero-latency tracking.
This method is discussed in detail by McDonald (18, 19).
We will now briey show how we used it.
For the human eye movement system the order of the
system, the control vector and the output vector are one,
so that the following values are appropriate:
Figure 15. The learning curve for the adaptive predictor. [From
D. R. Harvey and A. T. Bahill, Development and sensitivity analysis of adaptive predictor for human eye movement model, Transaction of the Society for Computer Simulation, December 1985.
1985 by Simulation Councils, Inc., San Diego, CA. Reprinted by
permission. (20).]
The lters transients die out within 2.5 s of each abrupt

change in velocity.
Figure 15 shows the predictors MSE, E( j)2 , plotted
against the number of iterations of the lter; an iteration was completed every 5 ms. After 450 iterations the
MSE was effectively zero, which corresponds to 2.25 s. This
agrees with the predictors output in Fig. 14. The settling
time of the lter, 450 iterations, is approximately 4 MSE ,
where MSE is the average time constant for the weights.
This gives a MSE of 112.5 iterations. Using Eq. (14), M
= n/4 MSE , the predictor has a misadjustment of approximately 33.3%.
The systems input, ri (t), is the sum of the target reference

signal, rs (t), and the adaptive signal, ra (t), that must be
computed. To obtain zero-latency tracking, y(t) must equal
rs (t). Putting all of this information into Eq. (10) gives
Solving for ra gives
The e+sT term shows that predictions must be made. However, the smooth-pursuit system is a velocity tracking system, not a position tracking system, so the controller must
be able to predict future values of target velocity. For example, if rs (t) is the present target velocity, it must be able to
produce rs (t + T ), where T is the time delay of the smoothpursuit system. And the controller must modify this pre-
12
diction to compensate for the dynamics of the system in

accordance with Fig. 10 becomes
This compensation signal allows the smooth-pursuit system to overcome the time delay. To synthesize this signal
the adaptive controller must be able to both predict future
values of the target velocity and compute rst derivatives.
These are reasonable computations for the neurons of the
human brain. Therefore, Eq. (14) is the algorithm that is
in the box of Fig. 10 labeled Target-selective adaptive controller.
Perform a Sensitivity Analysis
To determine which parameters have the greatest effect
on the model and when they exert their inuence, we comy
puted the semirelative sensitivity function, S , for each parameter (5,46,47):
where y is the output of the system and is the parameter

that is varied. For this study, we used a xed perturbation
size of +5%. In general a 5% step size may be too large, but
in this particular study it worked well (5).
The smooth-pursuit model developed in this study is not
independent of other systems. The saccadic system and
the adaptive predictor interact with the smooth-pursuit
branch. Therefore, we performed the sensitivity analysis
twice: once with the saccadic system and the predictor
turned on, and again with the saccadic system and the predictor turned off. Eliminating the saccadic system and predictor allowed us to isolate the pursuit branch and study
it independently.
The sensitivity of the predictor was studied for three parameters: ks , the proportionality constant; the number of
weights; and the time to be predicted. For ks and the number of weights, the target waveforms were also changed to
determine if the predictor was sensitive to different input
signals.
The effect of ks was found to be the greatest after points
of acceleration discontinuities. We performed a sensitivity analysis for many target waveforms, including the four
shown in Fig. 14. The inuence of ks is most apparent for
the analyses done with the cubic position waveforms. In
Fig. 16, we show the results for the cubical target position waveform, which has the parabolic velocity waveform
shown in this gure. S ks peaks at the turnaround points
and then begins to taper off to a steady-state value.
Similar results were found for the sensitivity analyses
when the number of weights was changed for each target
waveform. Figure 17 shows the results of the sensitivity
occurs a little later for the weights. S n is similar for the two
parameters, but the time of greatest sensitivity occurs a
little later for the weights. This similarity of the two sensitivity functions is reasonable if the misadjustment algorithm of the adaptive lter from Eqs. 6 and 9 is recalled:
Figure 16. Semirelative sensitivity function of the predictor for

changes in the proportionality constant, ks , for a cubic waveform.
Figure 17. Semirelative sensitivity function of the predictor for

changes in the number of weights for a cubic waveform. [From D.
R. Harvey and A. T. Bahill, Development and sensitivity analysis
of adaptive predictor for human eye movement model, Transaction of the Society for Computer Simulation, December 1985.
1985 by Simulation Councils, Inc., San Diego, CA. Reprinted by
permission. (20).]
This equation shows that a 5% change in either the proportionality constant, ks , or the number of weights, n, will
change the misadjustment of the predictor in a similar
manner.
The other parameter changed for the predictor was the
prediction time, the desired time to be estimated. The S
curve for this case also had the same shape as the curve
for the number of weights and the proportionality constant,
13
but its magnitude was smaller.

From these curves, the effect of the predictor can be determined. Changing each parameter by 5% showed that all
of them exert their greatest inuence right after acceleration discontinuities. Therefore, the predictors inuence
will be the greatest at those points.
The Effect of Parameter Changes on the Mean-Square
Error
Our semirelative sensitivity analysis gives a measure of
how changing a parameter affects the model, and it shows
when the parameter exerts its greatest inuence. For our
second sensitivity analysis, we considered the effect on the
models performance of changing each parameter over a
range of values. Each parameter was given values above
and below the nominal values; the velocity MSE between
the models output and the target was computed for each
change. For the predictor, the lters mean-square error
(MSE) was computed between the velocity of the target
150 ms in the future and the velocity predicted by the adaptive predictor. The MSE were then plotted against the parameter values.
Figure 18. The MSE of the predictor as a function of changes in

the proportionality constant. [From D. R. Harvey and A. T. Bahill,
Development and sensitivity analysis of adaptive predictor for human eye movement model, Transaction of the Society for Computer
Simulation, December 1985. 1985 by Simulation Councils, Inc.,
San Diego, CA. Reprinted by permission. (20).]
The Predictors Sensitivity to Changes in Parameters

The effect of changes in the proportionality constant on the
predictor was studied rst. As the proportionality constant
in Fig. 18 became larger, the lters MSE became smaller.
According to the misadjustment algorithm, the larger the
value of ks , the larger the misadjustment. This appears to
disagree with this gure. However, the MSE for the gure
was taken during the rst 12 s of the simulation; therefore,
the start-up transients are inuencing the error. The larger
the value of ks , the faster the lter adapts; for smaller ks
the lter takes longer to converge, but does not converge to
a solution with smaller error. Therefore, in the gure, the
large MSE for a small ks results because the lter takes
longer to converge to the optimal solution. With the larger
ks values, the lter is converging rapidly and appears to
have a smaller error. If ks were increased even more, the
error would also begin to increase. When we made the lters task easier, by eliminating the start-up transient and
only studying the steady-state behavior, we found that the
lters MSE increased with ks as expected.
Our detailed analysis also showed a larger MSE for the
cubic waveform compared to the sinusoidal waveform. This
result is not unexpected since the cubic is a waveform that
is of higher order than the sine wave and because the misadjustment is proportional to the expected value of the input signal.
Referring to Fig. 19, the MSE of the predictor is shown
as function of the number of weights in the adaptive lter.
According to the misadjustment algorithm, as the number
of weights increases, so does the misadjustment of the lter.
The curves here show the lters error decreasing until 15
weights and then rising slightly before falling off after 40
weights. Because the adaptive lters use a tapped delayline input signal, as the number of weights is increased,
the input signals for the adaptive and slave lters begin
to overlap. This improves the predictors performance because the statistics of the two input signals are the same
Figure 19. The MSE of the predictor as a function of changes in

the number of weights for a cubic waveform. [From D. R. Harvey
and A. T. Bahill, Development and sensitivity analysis of adaptive predictor for human eye movement model, Transaction of the
Society for Computer Simulation, December 1985. 1985 by Simulation Councils, Inc., San Diego, CA. Reprinted by permission.
(20).]
since the input signals are the same.

The increase in error between 15 and 40 weights shows
the rise in error predicted by the misadjustment algorithm.
However, after 40 weights the statistics of the input signals
for the two lters begin to get close enough that the error
drops off. The input signals for the two lters begin overlapping after 30 weights, which is approximately where the
curves peak.
The effect of changing the prediction time and the signals frequency were also studied. Figure 20 shows the predictors error as a function of prediction time. The error
appears to be a linear function of the prediction time. The
further into the future that is to be predicted, the worse the
predictor does. We also computed that for changes in frequency, the faster the target moves the poorer the predictor
does.
14
Figure 20. The change in the MSE of the predictor as the prediction time is changed. [From D. R. Harvey and A. T. Bahill, Development and sensitivity analysis of adaptive predictor for human eye
movement model, Transaction of the Society for Computer Simulation, December 1985. 1985 by Simulation Councils, Inc., San
Diego, CA. Reprinted by permission. (20).]
Summarizing, the predictors performance is poorer as

the proportionality constant is increased, although the error is a function of the time when the measurements were
taken. For instance, in this study the start-up transients
have not died down so the reverse statement appears true.
For the weights, as the number of weights increased, the
error also increased. The exception, seen in this work, occurs when a tapped delay-line input signal is used and the
statistics of the input signals to the adaptive and slave
lters are similar. The error of the predictor increases as
the prediction time increases and as the input signals frequency increases.
Discussion of Model and Least-Mean-Square Predictor
To create a model, we rst determine the form and then derive parameter values. When possible we use physiological
data to derive these values. A sensitivity analysis shows
which parameters are the most and the least important so
we can focus our efforts appropriately. In one of our nal
modeling stages we run a function minimization routine to
adjust parameter values so that we get the least-squared
error between the human and the model outputs.
Our model shown in Fig. 10 approximates the human
smooth-pursuit system. Similarly, our simulation is only
an approximation of the model in Fig. 10. For example, the
model of Fig. 10 should be stable for any gain up to 2.3.
But our simulation started to oscillate at 1.8. We found
that we were getting 5 to 10 of articial phase shift from
the differentiators, the integrators, and even the summers.
A smaller simulation step size would have obviously solved
the problem; however, just being aware of the problem was
also sufcient.
Our LMS predictor worked well except when discontinuities in the target waveform were present. For any desired accuracy, trade-offs could be made between the predicted gain and the number of weights. When this predictor was incorporated into our full eye movement model,
the model was able to overcome its 150 ms delay and track
Figure 21. Position as a function of time for the TSAC model

tracking a target with only the smooth-pursuit branch (top),
smooth-pursuit and saccadic branches (middle), and all three
branches turned on (bottom). Only the bottom trace resembles
tracking of a normal human. Target movement was 5 ; time is
in seconds. [From A. T. Bahill and J. D. McDonald, Model emulates human smooth pursuit system producing zero-latency target
tracking, Biol. Cybern., 48: 218, 1983 (19).]
targets with no latency, just like the human.

For optimal performance, 150 weights were used. Because the model gets a new target position every 5 ms, this
means it uses the previous 750 ms of data for each calculation. We are not sure that the human uses this large a
data window. Therefore, we also ran the model with only
15 weights. Even with this reduced number of weights, the
model still performed as well as the human.
VALIDATE THE MODEL
The model tracks targets just as humans do. But in addition, we can do things with the model that we cannot
do with humans. In Fig. 21 the top trace shows the model
tracking with only the smooth-pursuit branch turned on,
that is the saccadic branch and the adaptive predictor were
turned off. The middle trace shows the model tracking with
smooth pursuit and saccades only. Finally, the bottom trace
shows the model tracking with smooth-pursuit saccades
and the adaptive predictor. Only the bottom trace matches
the tracking of normal humans.
Overcoming a Time Delay
To overcome a time delay and produce zero-latency tracking, you must (a) predict future target positions and (b)
compensate for system dynamics, shown in Fig. 10. The
latter means that you must have a model of the system
that is updated when the system is changed by exercise,
fatigue, or temperature variations.
INVESTIGATE ALTERNATIVE MODELS
We have just shown the development of the LMS adaptive
predictor. It worked well, but we also compared it to alternative predictors. In our models we used the following
predictors: (a) difference equations, for example, r(n + 1)
= Ar(n) + Br(n 1), (b) menu selection, (c) LMS adaptive
lters, (d) recursive least-square (RLS) lters, (e) Kalman

lters, (f) adaptive lattice lters, and (g) a recursive leastsquare lter in conjunction with menu selection.
Difference equations were the simplest and least accurate. In the menu selection technique, the system has a
menu of waveforms to choose from. In our simple models,
we allowed sinusoidal, parabolic, cubic, and pseudorandom
waveforms. The model then tracked the target and tried to
identify the frequency, amplitude, and target waveform. It
then used an equation for that waveform to help predict
target motion. The seventh technique used a RLS lter
to identify the waveform and then used equations off the
menu to track the target. The other four techniques are
typical lters described in digital signal processing literature.
When we rst searched for literature on prediction we
found very little. Then we realized that any digital lter
could also be used for prediction. In fact, if you can either
model a system, identify a system, lter a signal, or predict
a signal, then you can do the other three operations with no
additional effort. All of our predictors allowed zero-latency
tracking, just like the human. But, as will be discussed
later, some matched other aspects of human behavior better than others.
The principle of Ockhams razor (48) states that if
two models are equal in all respects except complexity,
then the simpler model is the better one (see also pespmcl.vub.ac.be/OccamRaz.html). This is one reason why
we like the menu selection predictors. They are simpler
than the digital lters, which require complex matrix manipulation. Such calculations are ne for serial processing
digital computers, but are not likely to be used by parallel
processing analog computers such as the brain. This is one
of the reasons that articial neural networks are becoming
so popular among physiological systems modelers (11).
15
Figure 22. Human tracking of a sinusoidal target waveform. The

top trace shows target position (dashed) and eye position (solid),
the middle trace shows target velocity, and the bottom trace shows
eye velocity. The eye velocity waveform does not match the target
velocity waveform. [From A. T. Bahill and J. D. McDonald, Model
emulates human smooth pursuit system producing zero-latency
target tracking, Biol. Cybern., 48: 220, 1983 (19).]
EMULATE SOMETHING NOT USED IN THE MODELS

DESIGN
A powerful technique for validating a model is to use it
to simulate something that was unknown when the model
was developed. Figure 22 shows some human tracking was
that was noted to be unusual when the data were collected.
The target position was a sinusoidal waveform, but the eye
velocity waveform looks like that of a parabola. This behavior had not been seen before, and no explanation was apparent. But then we ran the menu selection model forcing it
to choose the wrong waveform. Figure 23 shows the model
tracking a sinusoidal waveform using a wrong guess of the
parabolic waveform. These waveforms look very much like
the human tracking of Fig. 22. This is another reason that
we favor the menu selection predictors.
The Science of Baseball
To help validate the model, we used it to simulate something that was not used in the design of the model. Ted
Williams, arguably the best hitter in the history of baseball, has described hitting a baseball as the most difcult
single act in all of sports (49). The speed of the ball approaches 100 mph (45 m/s) (baseball is a game of inches,
Figure 23. TSAC model with menu selection predictor tracking a

sinusoidal target with an incorrect (parabolic) adaptive signal. The
top trace shows target position (dashed) and model eye position
(solid), the middle trace shows target velocity, and the bottom trace
shows model eye velocity. [From A. T. Bahill and J. D. McDonald,
Model emulates human smooth pursuit system producing zerolatency target tracking, Biol. Cybern., 48: 219, 1983 (19).]
so the SI units come second in this section), producing angular velocities greater than 500 /s as the ball passes the
batter. Humans cannot track targets moving faster than
70 /s (50) or perhaps 100 /s (51); yet, professional batters
manage to hit the ball with force consistently and are able
to get a piece of the ball on an average of more than 80% of
their batting attempts. In this section we investigate how
they do this by examining a professional baseball player
tracking a pitched ball, and we demonstrate the superiority of his eye movements and headeye coordination to
16
those of our other subjects.

Why did we want to study a batter tracking a baseball?
We wanted to learn more about how the brain controls
movement, and we therefore were searching for a situation
in which a human was performing optimally. This condition is fullled by a professional baseball player tracking
a pitched baseball.
In addition to the four basic eye movement systems, the
batter can also use the head-movement system. Does he?
Earlier studies by Bahill and LaRitz (41) have suggested
several strategies for tracking a baseball: Track the ball
with head movements and smooth-pursuit eye movements
and fall behind in the last 5 ft (1.5 m) of ight; track with
eyes only, or with head only, and fall behind in the last 5
ft (1.5 m); track the ball over the rst part of its trajectory
with smooth-pursuit eye movements, make a saccadic eye
movement to a predicted point ahead of the ball, continue
to follow it with peripheral vision, and nally, at the end of
the balls ight, resume smooth-pursuit tracking with the
balls image on the fovea, the small area in the center of the
retina that has ne acuity. We will examine each of these
strategies.
The Simulated Fastball. To discover how well a batter
tracked the ball, we had to be able to determine the position of the ball at all times, and thus we could not use a
real pitcher or a throwing machine. Instead, we simulated
the trajectory of a pitched baseball. We threaded a shing
line through a white plastic ball and stretched this line between two supports, which were set 80 ft (24 m) apart in
order to accommodate the 60.5 ft (18 m) between pitcher
and batter; a string was attached to the ball and wrapped
around a pulley attached to a motor, so that when the motor
was turned on, the string pulled the ball down the line at
speeds between 60 mph (27 m/s) and 100 mph (45 m/s). The
ball crossed the plate 2.5 ft (0.8 m) away from the subjects
shoulders, simulating a high-and-outside fastball thrown
by a left-handed pitcher to a right-handed batter. This, like
all our constraints, was designed to give our subjects the
best possible chance of keeping their eyes on the ball. A low
curve ball thrown by a right-handed pitcher would have
been much harder to track.
By controlling the speed of the motor and counting the
rotations of the shaft, we could compute the position of the
ball at every instant of time, and thus compare the position
of the ball to the position of the batters gaze. We dene both
positions in terms of the horizontal angle of the ball: the
angle between the line of sight pointing at the ball and a
line perpendicular to the subjects body (see Fig. 24). This
angle is slightly more than 0 when the pitcher releases
the ball, and it increases to 90 when the ball crosses the
plate.
Tracking of a Professional Baseball Player. Figure 25
shows the tracking of a professional ballplayer Brian
Harper, then a member of the Pittsburgh Pirates. He
tracked the ball using head and eye movements, keeping
his eye on the ball longer than our other subjects did. Our
best-tracking student fell behind when the ball was 9 ft
(2.7 m) in front of the plate. This professional baseball
player was able to keep his position error below 2 until
Figure 24. The horizontal angle of the ball, , as dened in this

study, ranges from near 0 degrees when the pitcher releases the
ball to 90 degrees when the ball crosses the plate. [From A. T. Bahill
and T. LaRitz, Why cant batters keep their eyes on the ball, Am.
Sci., 72: 250, 1984 (41).]
the ball was 5.5 ft (1.7 m) from the plate. The peak velocity of his smooth-pursuit tracking was 120 /s; at this point,
his head velocity was 30 /s, thus producing a gaze velocity of 150 /s. In three simulated pitches to the professional
baseball player, at speeds of 60 mph (27 m/s), 67 mph (30
m/s), and 70 mph (31 m/s) the overall tracking patterns
were the same; his maximum smooth-pursuit eye velocities were 120, 130, 120 /s (52).
The gaze graph also takes into account the side-to-side
and front-to-back movements of the head; such translations of the head can produce changes in the gaze angle
(53). The data show that the contribution of the translation angle was slight until the ball was almost over the
plate.
The professional baseball player had faster smoothpursuit eye movements than our other subjects. In fact,
he had faster smooth-pursuit eye movements than any reported in the literature. He also had better headeye coordination, tracking the ball with equal-sized head and eye
movements, whereas our other subjects usually had disproportionately large head or eye movements.
Keep Your Eye on the Ball. Although the professional
baseball player was better than the college students at
tracking the simulated fastball, it is clear from our simulations that batters, even professional batters, cannot keep
their eyes on the ball. Our professional baseball player was
able to track the ball until it was 5.5 ft (1.7 m) in front of
the plate. This could hardly be improved on; we hypothesize that the best imaginable athlete could not track the
ball closer than 5 ft (1.5 m) from the plate, at which point it
would be moving three times faster than the fastest human
could track. This nding runs contrary to one of the most
often repeated axioms of batting instructorsKeep your
eye on the balland makes it difcult to account for the
widely reported claim that Ted Williams could sometimes
see the ball hit his bat.
If Ted Williams were indeed able to do this, it could only
be possible if he made an anticipatory saccade that put his
eye ahead of the ball and then let the ball catch up to his
eye. This was the strategy employed by the subject of Fig.
26; this batter observed the ball over the rst half of its
trajectory, predicted where it would be when it crossed the
plate, and then made an anticipatory saccade that put his
eye ahead of the ball. Using this strategy, the batter could
see the ball hit the bat.
17
Figure 26. In order to see the ball hit his bat, this subject made
an anticipatory saccade, indicated by the jump in the gaze angle
(thick line) that put his eye ahead of the ball (thin line); as a result,
the ball was on the fovea at the point of contact. The subject did
not move his head until after the ball crossed the plate. [From A.
T. Bahill and T. LaRitz, Why cant batters keep their eyes on the
ball, Am. Sci., 72: 251, 1984 (41).]
ipatory saccade early in the trajectory.
Figure 25. The success of a professional baseball player in tracking a simulated 60-mph (27 m/sec) pitch is shown in these graphs.
The thin line in the top graph represents the horizontal angle of
the ball, , as it would be seen by a right-handed batter facing a
left-handed pitcher; the thick line represents the actual horizontal
angle of gaze of the subject trying to track this ball. This gaze angle curve is generated by combining the horizontal head angle, the
horizontal eye angle, and the head-translation angle, which represents the eye movement necessary to compensate for side-to-side
and front-to-back movement of the head. Movements to the right
appear as upward deections. [From A. T. Bahill and T. LaRitz,
Why cant batters keep their eyes on the ball, Am. Sci., 72: 251,
1984 (41).]
But why would a batter want to see the ball hit the bat?
Because of his slow reaction time, he could not use the information gained in the last portion of the balls ight to
alter the course of the bat. We suggest that he uses the
information to discover the balls actual trajectory; that is,
he uses it to learn how to predict the balls location when it
crosses the platehow to be a better hitter in the future.
The anticipatory saccade must be made before the end of
the trajectory, because saccadic suppression prevents us
from seeing during saccades (54, 55). This suppression of
vision extends about 20 msec after the saccade. So if you
want to see the ball hit the bat, you must make your antic-
Head Movements and the Vestibulo-Ocular System. The

vestibulo-ocular system is little used when tracking a baseball. However, in monitoring the eyes of our professional
ball player, we did detect a small vestibulo-ocular movement to the left during the early part of the balls trajectory,
as the head was moving to the right; this appears as the
slight dip between 0.5 s and 0.7 s in the eye position trace in
Fig. 25. At this point, the head position was changing faster
than the angular position of the ball, and the vestibuloocular eye movement compensated for the premature head
movement. Why would the batter want to start his head
movement early? The answer is that the head is heavier
than the eye and consequently takes longer to get moving;
therefore, in the beginning of the movement, as the head
starts turning to the right ahead of the ball, the vestibular
system in the inner ear signals the ocular system to make
a compensating eye movement, thus giving his head a head
start.
However, this vestibulo-ocular compensation must soon
stop. In the end, the eye and head must both be moving
to the right, and the batter must therefore suppress his
vestibulo-ocular reex so that the tracking head movement
does not produce compensating eye movements that would
take his eye off the ball. The professional baseball player
was very good at suppressing his vestibulo-ocular reex.
Some of our student subjects did not make head movements until after the ball crossed the plate; others moved
their heads very little. Perhaps they did this because they
could not suppress the vestibulo-ocular reex very well.
The fact that our professional baseball player used his
head to help track the ball seems to violate another often repeated batting axiom, Dont move your head. The
professional made small tracking head movements in the
range of 10 to 20 . He was able to suppress the vestibuloocular reex for these movements, which were probably
small enough to go unnoticed by a coach. However, body
movements could produce head movements of 90 or more;
it may be difcult to suppress the vestibulo-ocular reex
for these large body-induced movements, which along with
correlated poor performance would be noticed by a coach.
Therefore, we think the axiom should be protracted: Dont
18
Figure 27. The model trying to track a baseball with the predictor turned off. The top trace is the angular position of the ball (dotted) and gaze (solid) and the bottom trace is velocity. The record is
1 s long. The model kept its eye on the ball until the ball was 9
ft (2.7 m) in front of the plate. This tracking resembles that of our
best-tracking college students.
let your body move your head, but its okay to move your
head a little in order to track the ball.
Batters do not use vergence eye movements. This is reasonable, since vergence eye movements are not needed to
track the ball between 60 ft (18 m) and 6 ft (1.8 m) from
the plate and since there is not sufcient time to make such
movements between 6 ft (1.8 m) and the point of contact;
indeed, our data contained no vergence eye movements. So
any claim that a batter actually saw the ball hit the bat
must be based on monocular vision; only the dominant eye
tracks the ball.
Strategies. Sometimes our subjects used the strategy of
tracking with head and eyes and falling behind in the last 5
ft (1.5 m), and sometimes they used the strategy of tracking
with head and eyes but also using an anticipatory saccade.
It has been speculated that baseball players might use the
latter strategy when they are learning the trajectory of a
new pitch and use the former strategy when hitting home
runs.
The professional baseball player tracked our simulated
pitch better than any other subjects did. This superior
tracking was due to (a) his use of both head and eye movements, (b) real fast smooth-pursuit eye movements, and (c)
giving his head a head start.
Modeling Baseball Players. The eye movements of baseball players were not used in the development of the TSAC
model. So if the model could simulate such eye movements,
it would be a strong validation of the model. First, the limiter in the TSAC model was increased from its nominal
value of 70 /s to the 130 /s that the professional baseball
player exhibited. Figure 27 shows the model with the predictor turned off trying to track a baseball. It fell behind
when the ball was nine feet from the plate. Figure 28 shows
the model with the predictor turned on tracking a baseball.
It was able to track the ball until the ball was 5.5 feet from
the plate. The predictor makes a big difference. With the
predictor, the model does as well as the professional baseball player whose data are shown in Fig. 25. The ball and
gaze traces of Fig. 25 look very much like those of Fig. 28.
Models Are Simplications
Remember that a model is a simplied representation of
some particular aspect of a real-world system. The real
Figure 28. The model tracking a baseball with the predictor

turned on. The model kept its eye on the ball until the ball was
5.5 ft (1.7 m) in front of the plate. This tracking resembles that of
our professional baseball player.
baseball moves in three dimensions. The right-hand rules

explain the spin-induced deection of balls (57)in this three
dimensional space. In the studies reported in this chapter,
we controlled the movement of the ball so that vertical eye
movements were not required to track the ball.
What Was the Purpose of this Section on Baseball?
These data were not used in the development of the model.
Therefore, trying to make the model match these human
data is a test of the model. The model was able do things
it was not designed to do. Of course, some things had to
be modied. For example, the literature on smooth pursuit eye movements says that humans cannot track targets moving faster than 70 degrees/sec. Yet the professional baseball player of Figure 25had smooth pursuit eye
movements up to 130 deg/sec. So the Limiter in the model
should be adjustable for different levels of performance.
Thus the models architecture did not have to change, it
only required parameter tuning. The models validation is
also enhanced by the models realistic interactions with the
saccadic and vestibulo-ocular systems as shown in Figures
24 to 28.
SUMMARY
The data presented in this section prompt the following
summary about eye movements and baseball. You cant
keep your eye on the ball. Our best students could only
track the ball until it was 9 feet in front of the plate. At
that point, its annular velocity was so high that they fell
behind. However, professional baseball players have superior eye movements. They have better coordination of
head and eye movements. They have faster smooth pursuit eye movements than any reported in the oculomotor
literature. And they have better suppression of vestibuloocular eye movements. Because of these superiorities, the
professional baseball player of this study was able to keep
his eye on the ball until it was 5.5 feet in front of the plate.
At this point, its annular velocity was so great that no one
could track it farther. However, many people have reported
that they have seen the ball hit their bat. This is possible
using a different strategy: track the ball over the rst half
of its ight, and then make a saccade that takes the eye
off of the ball and aims the eye at the future site of the
bat-ball collision. With this strategy, you can see the ball
hit the bat.
The data presented in this section prompt the following summary about modeling human eye movements. Humans can overcome the time delays of the eye movement
systems and track predictable visual targets with no latency or phase lag. To do the same, the TSAC model had to
compensate for system dynamics and predict target velocity. Therefore, we think humans must use mental models of
their eye movement systems to help compensate for system
dynamics. These mental models must be adaptive, so that
they can change due to muscular activity, fatigue, temperature, and so on. One good way to predict target velocity is
menu selection. The baseball players menu contains fastball, curveball, and slider.
ACKNOWLEDGMENT
BIBLIOGRAPHY
1. W. L. Chapman, A. T. Bahill, A. W. Wymore, Engineering Modeling and Design, Boca Raton FL: CRC Press, 1992.
2. A. T. Bahill, W. J. Karnavas, The ideal baseball bat, New Sci.,
130 (1763): 2631, 1991.
3. A. T. Bahill, M. Morna Freitas, Two methods for recommending
bat weights, Ann. Biomed. Eng., 23 (4): 436444, 1995.
4. A. T. Bahill, The ideal moment of inertia for a baseball or softball bat, IEEE Transactions on Systems, Man and Cybernetics,
Part A: Systems and Humans, 34(2): 197204, 2004.
5. S. J. Yakowitz, F. Szidarovszky, An Introduction to Numerical
Computations, New York: Macmillan, 1989.
6. A. T. Bahill, Bioengineering: Biomedical, Medical and Clinical
Engineering, Englewood Cliffs, NJ: Prentice-Hall, 1981.
7. A. T. Bahill, B. Gissing, Re-evaluating systems engineering
concepts using systems thinking, IEEE Trans. Syst. Man Cybern., part C,28: November 1998.
8. A. T. Bahill, R. Botta and J. Daniels, The Zachman framework
populated with baseball models, Journal of Enterprise Architecture, November 2006.
9. C. Rashbass, The relationship between saccadic and smooth
tracking eye movements, J. Physiol., 159: 326338, 1961.
10. A. T. Bahill, J. D. McDonald, Smooth pursuit eye movements
in response to predictable target motions, Vision Res., 23:
15731583, 1983.
11. G. Westheimer, Eye movement responses to a horizontally
moving visual stimulus, AMA Arch. Ophthalmol., 52: 932941,
1954.
12. L. Stark, G. Vossius, L. R. Young, Predictive control of eye
tracking movements, IRE Trans. Hum. Factors Electron.,
HFE-3: 5257, 1962.
13. R. E. Kettner et al., Prediction of complex two-dimensional
trajectories by a cerebellar model of smooth pursuit eye movement, J. Neurophysiol., 77: 21152130, 1997.
14. P. J. Dallos, R. W. Jones, Learning behavior of the eye xation
control system, IEEE Trans. Autom. Control, AC-8: 218227,
1963.
15. S. Yasui, L. Young, Perceived visual motion as effective stimulus to pursuit eye movement system, Science, 190: 906908,
1975.
16. L. Young, Pursuit eye movementswhat is being pursued, inR. Baker andA. Berthoz (eds.), Control of Gaze by
Brain Stem Neurons, pp. 2936, Amsterdam: Elsevier/NorthHolland, 1977.
19
17. M. Steinbach, Pursuing the perceived rather than the retinal

stimulus, Vision Res., 16: 13711376, 1976.
18. A. Mack, R. Fendrich, E. Wong, Is perceived motion a stimulus
for smooth pursuit?, Vision Res., 22: 7788, 1982.
19. R. Eckmiller, A Model of the Neural Network Controlling
Foveal Pursuit Eye Movements, inA. F. Fuchs andW. Becker
(eds.) Progress in Oculomotor Research, pp. 541550, New
York: Elsevier/North-Holland, 1981.
20. J. D. McDonald, A. T. Bahill, Zero-latency tracking of predictable targets by time-delay systms, Int. J. Control, 38:
881893, 1983.
21. A. T. Bahill, J. D. McDonald, Model emulates human smooth
pursuit system producing zero-latency target tracking, Biol.
Cybern., 48: 213222, 1983.
22. D. R. Harvey, A. T. Bahill, Development and sensitivity analysis of adaptive predictor for human eye movement model,
Trans. Soc. Comput. Simul., 2: 275292, 1985.
23. D. E. McHugh, A. T. Bahill, Learning to track predictable target waveforms without a time delay, Invest. Ophthalmol. Vis.
Sci., 26: 932937, 1985.
24. H. Davson, The Physiology of the Eye, New York: Academic,
1976.
25. R. Botta, Z. Bahill, and A. T. Bahill, When are Observable States
Necessary? Systems Engineering, 9(3): 228240, 2006.
26. G. Marmount, Electrode clamp for squid axon: Studies on the
axon membrane, J. Cell. Comp. Physiol., 34: 351382, 1949.
27. K. S. Cole, Dynamic electrical characteristics of the squid axon
membrane, Arch. Sci. Physiol., 3: 253258, 1949.
28. L. Stark, Neurological Control Systems, Studies in Bioengineering, pp. 308312, New York: Plenum, 1968.
29. R. B. Stein, What muscle variable(s) does the nervous system
control in limb movements?, Behav. Brain Sci., 5: 535577,
1982.
30. L. R. Young, L. Stark, Variable feedback experiments testing a
sampled data model for eye tracking movements, IEEE Trans.
Hum. Factors Electron., HFE-4: 3851, 1963.
31. H. J. Wyatt, J. Pola, Smooth pursuit eye movements under open-loop and closed-loop conditions, Vision Res., 23:
11211131, 1983.
32. H. J. Wyatt, J. Pola, The role of perceived motion in smooth
pursuit eye movements, Vision Res., 19: 613618, 1979.
33. A. T. Bahill, D. R. Harvey, Open-loop experiments for modeling the human eye movement system, IEEE Trans. Syst. Man
Cybern., SMC-15: 240250, 1986.
34. H. Collewijn, E. P. Tamminga, Human xation and pursuit in
normal and open-loop conditions: Effects of central and peripheral retinal targets, J. Physiol. (Lond.), 379: 109129, 1986.
35. R. J. Leigh et al., Visual following during stimulation of
an immobile eye (the open loop condition), Vision Res., 22:
11931197, 1982.
36. J. Pola, H. J. Wyatt, Offset dynamics of human smooth pursuit
eye movements: Effects of target presence and subject attention, Vision Res., 37: 25792595, 1997.
37. P. Van Donkelarr, C. Fisher, R. G. Lee, Adaptive modication
of oculomotor pursuit inuences manual tracking responses,
Neuroreport 5: 22332236, 1994.
38. M. F. W. Dubois, H. Collewijn, Optokinetic reactions in man
elicited by localized retinal motion stimuli, Vision Res., 19:
11051115, 1979.
39. S. N. J. Watamaniuk, S. J. Heinen, Perceptual and oculomotor evidence of limitations on processing acceleration motion,
Journal of Vision, 3: 698709, 2003.
20
40. C. J. Gordon, J. E. Heath, Integration and control processing

in temperature regulation, Annu. Rev. Physiol., 48: 595612,
1986.
41. J. Ludbrook, Reex control of blood pressure during exercise,
Annu. Rev. Physiol., 45: 155168, 1983.
42. T. L. Alvarez, M. Bhavsar, J. L. Semmlow, M. T. Bergen C.
Pedrono, Short-term predictive changes in the dynamics of
disparity vergence eye movements, J of Vision, 5(7), 640649,
2005.
43. D. J. C. Read, D. J. Henderson-Smart, Regulation of breathing in newborn during different behavioral states, Annu. Rev.
Physiol., 46: 675685, 1984.
44. D. A. Robinson, The mechanics of human smooth pursuit eye
movements, J. Physiol., 180: 569591, 1965.
45. J. S. Brodkey, L. Stark, New direct evidence against intermittency or sampling in human smooth pursuit eye movements,
Nature, 218: 273275, 1968.
46. A. T. Bahill, T. LaRitz, Why cant batters keep their eyes on
the ball, Amer. Sci., 72: 249253, 1984.
47. L. Young, Pursuit eye tracking movements, inP. Bach-y-Rita,
C. C. Collins, andJ. Hyde (eds.), The Control of Eye Movements,
New York: Academic, 1971, pp. 429443.
48. B. Widrow, Adaptive lters, inR. E. Kalman, andN. DeClaris
(eds.), Aspects of Network and System Theory, pp. 563587,
New York: Holt, Rinehart and Winston, 1971.
49. B. Widrow et al., Stationary and nonstationary learning
characteristics of the LMS adaptive lter, Proc. IEEE, 64:
11511162, 1976.
50. B. Widrow, S. D. Stearns, Adaptive Signal Processing, Englewood Cliffs, NJ: Prentice-Hall, 1985.
51. A. T. Bahill, J. R. Latimer, B. T. Troost, Sensitivity analysis
of linear homeomorphic model for human movement, IEEE
Trans. Syst. Man Cybern., SMC-10: 924929, 1980.
52. W. J. Karnavas, P. Sanchez, A. T. Bahill, Sensitivity analyses
of continuous and discrete systems in the time and frequency
domains, IEEE Trans. Syst. Man Cybern., SMC-23: 488501,
1993.
53. W. H. Jefferys, J. O. Berger, Ockhams razor and Bayesian analysis, Am. Sci., 80: 6472, 1992.
54. T. Williams, J. Underwood, The Science of Hitting, New York;
Simon and Schuster, 1982.
55. L. Schalen, Quantication of tracking eye movements in normal subjects, Acta Otolaryngol., 90: 404413, 1980.
56. G. Zaccara et al., A new method for analyzing smooth-pursuit
eye movements description of a microcomputer program and
evaluation in healthy volunteers, Ital. J. Neurol. Sci., 12:
5375445, 1991.
57. R. G. Watts, A. T. Bahill, Keep Your Eye on the Ball: Curve Balls,
Knuckleballs, and Fallacies of Baseball, New York: Freeman,
2000.
58. J. D. McDonald, A. T. Bahill, M. B. Friedman, An adaptive control model for human head and eye movements while walking,
IEEE Trans. Syst. Man Cybern., SMC-13: 167174, 1983.
59. L. Stark, J. A. Michael, B. L. Zuber, Saccadic suppression:
A product of the saccadic anticipatory signal, inC. R. Evans
andT. B. Mulholland (eds), Attention in Neurophysiology, London, Butterworths, 1969, pp. 281303.
60. E. Matin, Saccadic suppression and the dual mechanism theory of direction constancy, Vision Res., 22: 335336, 1982.
61. A. T. Bahill, T. M. Hamm, Using open-loop experiments to
study physiological systems with examples from the human
eye movement systems, News Physiol. Sci., 4: 104109, 1989.
62. A. T. Bahill, D. G. Baldwin, Describing baseball pitch movement with right-hand rules, Computers in Biology and
Medicine, 2007.
A. TERRY BAHILL
University of Arizona, Tucson,
AZ, 85721
NEURAL PROSTHESES
339
Visual prosthesis
Auditory prosthesis
Respiratory prosthesis
Breathing assistance
Cough
Upper-extremity function
Grasp and release
Reaching
Genitourinary prosthesis
Bladder control
Bowel control
Erection and ejaculation
Lower-extremity function
Standing
Transfers
Stepping and walking
Figure 1. Examples of applications of electrical stimulation of the

nervous system to restore function to individuals with neurological
impairment.
FUNDAMENTALS OF ELECTRICAL STIMULATION

OF THE NERVOUS SYSTEM
The fundamental unit of communication in the nervous system is the action potential, an electrochemical signal that
propagates along neurons as a flux of ionic current between
the extracellular and intracellular spaces. Artificially generated action potentials can be initiated by electrical stimuli,
and will propagate from the site of stimulation in the same
way as, and have the same effect as, naturally generated ac-
Neural prostheses are a developing technology that use electrical activation of the nervous system to restore function to
individuals with neurological impairment. Applications have
included stimulation in both the sensory and motor systems
(Fig. 1) and range in scope from experimental trials in single
individuals, as in the case of the visual prosthesis, to commercially available devices placed in thousands of individuals, as
in the case of auditory prostheses (Fig. 2). Neural prostheses
function by initiation of action potentials in nerve fibers
which carry the signal to an endpoint where chemical neurotransmitters are released, either to affect an end organ or another neuron. Thus, neural prostheses are all devices that enable selective and graded control of neurotransmitter release
and, in principle, any end organ under neural control is a
candidate for neural prosthetic control.
1200
Freehand
Vocare
Clarion
Nucleus
20000
15000
1000
800
10000
600
400
5000
200
0
1975
1980
1985
1990
1995
Cumulative number implanted
NEURAL PROSTHESES
Cumulative number implanted
1400
0
2000
Year
Figure 2. Cumulative numbers of devices implanted for restoration
of hand-grasp (FreeHand), bladder and bowel function (Vocare), and
restoration of hearing (Clarion, Nucleus). Data for Freehand and Vocare provided by NeuroControl Corp., Cleveland, OH. Data for Clarion provided by Advanced Bionics Corp., Sylmar, CA. Data for Nucleus (right axis) provided by Cochlear Corp., Denver, CO.
NEURAL PROSTHESES
tion potentials. Figure 3 shows the responses of a model neuron to two different amplitude stimuli. In response to the
smaller stimulus, the neural membrane responds like a parallel RC circuit. However, above a critical amplitude (threshold)
the membrane initiates an action potential as a result of flux
of sodium ions from the extracellular space to the intracellular space. This action potential is then propagated down the
fiber to its terminal where neurotransmitter is released to affect the end organ or another neuron. Artificial generation of
action potentials is the basis for all neural prostheses.
The threshold for excitation (i.e., initiation of an action potential) depends on both the amplitude and the duration of
the stimulus. As shown in Fig. 4(a), the stimulus amplitude
necessary for excitation, Ith, increases as the duration of the
stimulus, PW, is decreased. This relationship is termed the
strengthdurations relationship and is given by Eq. (1).
10
Threshold current amplitude
340
Irh
1
10
Tch
100
1000
Pulse duration ( s)
(a)
(1)
The parameter Irh is the rhoebase current, and is defined as

the current amplitude necessary to excite the neuron with a
pulse of infinite duration. The parameter Tch is the chronaxie
and is defined as the pulse duration necessary to excite the
neuron with a pulse amplitude equal to twice the rheobase
current.
The amount of charge necessary for excitation can be determined directly from the strength duration relationship and
is given by Eq. (2).
Qth =
PW Irh
1 exp (PW /Tch )
(2)
100
Threshold charge
Ith
Irh
=
1 exp{PW/[ln(2)Tch ]}
10
1
10
The charge required for excitation decreases as the duration

of the pulses decreases [Fig. 4(b)]. Thus, short pulses are
more efficient at generating excitation. Short pulses also have
the benefits of increasing the threshold difference between
different diameter nerve fibers, thereby decreasing the gain
between stimulus magnitude and the number of nerve fibers
activated (1), and increasing the threshold difference between
nerve fibers lying at different distances from the electrode,
thereby increasing spatial selectivity of stimulation (2). Min-
100
1000
Pulse duration ( s)
(b)
Figure 4. Strength-duration (a) and charge-duration (b) relations for

threshold excitation of a nerve fiber.
imizing charge, by use of short pulses, is an important consideration for preventing tissue damage, preventing electrode
corrosion, and minimizing power consumption.
The current required for extracellular stimulation of axons
also depends on the spatial relationship between the electrode
and the nerve fiber and the nerve fiber diameter (3). Transmembrane potentials generated by extracellular current are
largest in the fibers close to the stimulating electrode, thus
less current is required to stimulate neurons in the proximity
of the electrode (Fig. 5). As the distance between the electrode
and the fiber increases the threshold, Ith, increases, and for
excitation of myelinated nerve fibers with a point source electrode, this relationship is described by Eq. (3).
Ith = Io + k r2
Figure 3. Subthreshold and suprathreshold responses of a nerve fiber to a stimulus current pulse. The traces show the transmembrane
voltage as a function of time in response to stimuli of subthreshold
and suprathreshold amplitude. In the subthreshold regime, the membrane behaves as a parallel RC circuit, while in the suprathreshold
regime the membrane generates an action potential.
(3)
The offset, Io, determines the absolute threshold and the

slope, k, determines the threshold difference between fibers
at different distances, r, from the electrode (2).
Similarly, in response to an externally applied stimulus,
nerve fibers with a larger spacing between the nodes of Ranvier experience transmembrane potential changes that are
NEURAL PROSTHESES
trons, while in the body current is carried by ions. Therefore,

at the electrodetissue interface, charge transfer from electrodes to ions must occur. Certain interfacial reactions may
be reversed by using a charge balanced, biphasic stimulus
pulses, and thus stimulation may occur without tissue damage. However, irreversible reactions may occur even with the
use of biphasic stimuli and waveforms must be designed
within established safety limits (5,6)
Threshold current amplitude

(mA)
341
10 m
20 m
MOTOR SYSTEM NEURAL PROSTHESES

1
Restoration of Limb Movement Function

0
0
0.5
1.5
Electrode to fiber distance (cm)

Figure 5. Relationship between the threshold current and the distance between a point source electrode and a nerve fiber in a homogeneous medium ( 55 cm) with the electrode positioned directly
over a node of Ranvier.
larger than those in fibers with a smaller internodal spacing.

Under normal conditions the larger diameter nerve fibers
have larger internodal spacings. Thus when conventional
rectangular stimuli are used larger diameters fibers are activated at smaller stimulus amplitudes than the smaller diameter fibers [(Fig. 5(a)].
FUNDAMENTALS OF ELECTRODES
AND STIMULUS WAVEFORMS
There are several factors that must be taken into consideration when selecting stimulus waveforms and electrodes for
use in a neural prosthesis. The fundamental requirements are
that a sufficient change in transmembrane potential be induced to generate action potentials and that the underlying
tissue not be damaged by the electrode or stimulus.
Neural prosthesis electrodes require a conductor for current delivery, an insulating carrier for the stimulating element, a conductive lead, and lead insulation. Electrodes must
provide selective, graded, and maximal activation of the targeted tissue in a stable and repeatable manner with minimal
activation of neighboring tissue. Implantable electrodes must
be both passively and actively biocompatible. Passive biocompatibility refers to the tissue reaction to the composition,
shape, and mechanical properties of the electrodes materials,
while active biocompatibility refers to the performance of the
device under stimulation. A polarization voltage develops
across the electrodetissue interface when current is passed
through the electrode. Therefore, a regulated current waveform should be used with implanted electrodes to ensure that
the electric field generated in the tissue is independent of
electrode polarization. However, regulated voltage waveforms
should be used when surface electrodes are used to minimize
the possibility of skin burns if an electrode becomes dislodged.
Passage of current through the electrode may cause electrochemical reactions that lead to the formation of toxic chemical species around the electrode, corrosion of the electrode
material, or activation of tissue to a level that causes neuronal damage (4). Metal electrodes carry current via elec-
A number of diseases and injuries can impair the ability of

the nervous system to control movements of the limbs, preventing affected individuals from performing many routine
tasks requiring movements of the limbs, such as standing,
walking, and manipulation of objects by the hands and arms.
Conventional rehabilitation methods provide good solutions
to some problems (e.g., wheelchairs for routine ambulation),
but individuals with movement disabilities must rely on assistance from others for many routine tasks. Neural prostheses based on functional neuromuscular stimulation can restore some of these movements, offering users increased
independence in daily activities and the potential economic
benefits of lower attendant costs and improved occupational
opportunities.
Neural Control of Movement and Disorders. The nervous system normally controls movement using a number of parallel
mechanisms, including voluntary commands produced at a
conscious level, locomotion commands produced at a semiconscious level, and reflex responses (e.g., the knee jerk) at an
unconscious level. However, all neural control systems ultimately activate the same muscles in order to produce their
effect, and therefore all converge upon the final common
pathway represented by the several hundred motoneurons
which typically innervate a muscle. Disorders which result in
impaired movement function can occur at a number of sites
in the chain from the brain down to the muscle fibers. All
current movement neural prostheses substitute artificial electrical activation of the peripheral end of this chain (usually of
the motoneurons themselves) for the normal activation by the
nervous system, and are thus applicable only to lesions that
spare the muscles and motoneurons. Thus, neural prostheses
are not currently used to restore functions lost by diseases
or injuries of the muscles (e.g., muscular dystrophy) or the
motoneurons themselves (e.g., brachial plexus injury). Although neural prostheses could potentially improve movement function in a number of neurological disorders (e.g., cerebral palsy), the two primary applications to date are stroke,
with approximately 60,000 new cases per year with moderate
to severe permanent impairment, and spinal cord injury
(SCI), which affects 7,500 to 10,000 individuals per year. Both
of these conditions tend to be stable after the initial insult,
although the impairments due to stroke and SCI are somewhat different. Stroke normally affects one side of the body
only (producing hemiplegia) and is often accompanied by cognitive, speech, and other disorders. Sensory function is usually spared in individuals with stroke, and muscles in different parts of the limb can be paralyzed, weak, or simply
difficult to activate in a coordinated manner. SCI tends to oc-
342
NEURAL PROSTHESES
cur in younger individuals and the level of movement impairment depends upon the location of the injury. SCI at thoracic
(mid-back) or lower levels results in paraplegia (paralysis of
the legs and pelvis), while SCI at cervical levels results in
tetraplegia (also known as quadriplegiaparalysis of the
legs, trunk, and arms). Within either paraplegia or tetraplegia, the level of impairment increases as the level of spinal
cord injury progresses toward the head. SCI can be complete
(i.e., interrupting all communication with the brain) or incomplete (i.e., some communication with the brain is retained).
Both motor and sensory functions are typically affected.
Design Challenges for Movement Neural Prostheses. The design of neural prostheses must take into account a number of
physiological changes that accompany neurological disorders,
as well as limitations in the current technology for artificially
exciting the nervous system. Following stroke or SCI, paralyzed muscles undergo disuse atrophy, characterized by a
rapid and marked decrease in muscle mass, a decrease in
force-generating capacity, and increased susceptibility to fatigue (7,8). In SCI, denervation occurs when direct physical
damage due to the injury or its subsequent consequences
(swelling, release of various chemical factors, etc.) leads to the
death of motoneurons in and near the area of injury (8). As
the motoneurons degenerate, the muscle fibers they innervate
can no longer be activated by the nervous system, and electrical activation is difficult or impossible. The disuse of limbs
because of paralysis often leads to a rapid reduction in bone
density (osteoporosis), which is especially of concern in lower
extremity neural prostheses because of weight bearing and
the large muscle forces required. Disuse of a limb also often
results in an increase in the passive resistance of joints to
movement (contractures), which may make it difficult for already weakened muscles to move the limb through its needed
range. Paralyzed or paretic muscles are often spastic, that is,
have hyperactive stretch reflexes causing inappropriate muscle contractions or spasms. All currently available neural
prostheses recruit the motor units within the muscle in an
order reverse that of natural recruitment. In many muscles,
full activation via electrical stimulation cannot be achieved
without undesirable spillover to other muscles, limiting the
forces available. Usually, the number of stimulation channels
available is far less than the number of muscles normally participating in the movement function, so simpler alternate
strategies for completing functional tasks must be developed.
Many tasks also involve the simultaneous movement of several different joints, all of which may be impaired. Providing
the user with reasonably natural control over all these functions simultaneously can be a significant challenge.
At least partial solutions to each of these problems have
been developed. Exercise of paralyzed muscle by electrically
stimulated contractions for 2 to 8 hours per day has been
found (7,9) to make the muscle more resistant to fatigue, although increases in force appear to occur consistently in some
muscles but not in others. The effects of denervation following
SCI may be partially compensated by sprouting, a process by
which surviving motoneurons in a muscle reinnervate nearby
denervated muscle fibers and thus maintain their ability to
produce force when electrically stimulated. In the upper extremity, muscle tendon transfer of a nondenervated (either
voluntary or paralyzed) muscle can sometimes be performed
to replace the function of a denervated muscle. Appropriate
weight bearing and exercise can arrest and perhaps even reverse bone demineralization. Joint and muscle contractures
can often be prevented by appropriate therapy, including
movement through the range of movement. Surgical procedures can also be performed in some cases to release tight
joints (10). Spasticity can often be controlled pharmacologically.
Movement Neural Prostheses, Past and Current. Neural prostheses have been developed for restoring specific movements
of both the upper and lower extremities, as summarized in
Table 1. Applications for upper extremity movements have
historically focused on hand grasp and release, primarily in
individuals with cervical SCI. Several systems are currently
available for providing these functions, including two that are
based upon surface electrodes, one using percutaneous electrodes and an external stimulator, and one using a totally
implanted stimulator and implanted electrodes. The surface
systems are relatively inexpensive and are noninvasive since
no surgery is required. However, they require accurate electrode placement before each use, and individuals with denervation may not be able to use the system. The Handmaster
(Ness Ltd.) (11,12) device has been used in individuals with
C4 to C7 SCI, and in hemiplegia. The Tetron Glove (Neuromotion, Inc.) (13) utilizes voluntary wrist function of the user to
control the stimulation, so its applications are primarily limited to C6 to C7 individuals and those with hemiplegia. The
percutaneous electrodes used by Handa, Hoshimiya, and associates at Tohoku University (12,14) offer higher selectivity
and can reach deeper muscles inaccessible from the surface,
and are relatively inexpensive because the electrodes are implemented without open surgery. Muscle-stimulation patterns
are based on templates of natural muscle activation, so separate templates must developed and stored for each task to be
performed. This system has been applied to individuals with
cervical SCI (C4 to C6) and hemiplegia to produce hand, forearm, elbow, and shoulder function. The Freehand (NeuroControl, Inc.), developed originally by Peckham and associates
(7,15) uses 7 to 8 implanted epimysial stimulating electrodes
and a pacemaker-like stimulator implanted in the upper chest
to restore two grasp patterns (key grip and palmar grasp) for
individuals with C5 to C6 level SCI. Power and stimulus commands are transmitted electromagnetically via a skinmounted antenna to the implanted stimulator. Stimulus patterns are controlled voluntarily by the user via a joystick-like
device mounted on the opposite shoulder or on the ipsilateral
wrist. This implanted technology is more expensive and invasive than the other alternatives, but it is highly reliable, has
few external components, and is thus easy to put on and take
off. Furthermore, the implant procedure is usually performed
simultaneously with reconstructive surgeries such as muscle
tendon transfers (10) to maximize voluntary and stimulated
contractions, and to release passive constraints. Continuing
research with this system is extending its functionality to include stimulation of intrinsic hand muscles, wrist function,
forearm function, elbow function, shoulder function, and bilateral hand function. The use of implanted sensors is being
investigated, and the use of movement neural prostheses is
also being extended to individuals with hemiplegia and different levels (C3 to C4, C7) of SCI.
A number of neural prostheses for lower extremity function have also been developed. As noted in Table 1, several
NEURAL PROSTHESES
343
Table 1. Summary of Past and Current Movement Neural Prostheses

Group or Company
Stimulator Type
User-Control Method
11
Surface
External
Preprogrammed
13
Surface
External
Write motion
14
Percutaneous
External
Preprogrammed
NeuroControl, Inc.
Freehand
7,15
Implanted
Implanted
Contralateral shoulder motion
Cleveland VA/CWRU
research
7,39
Implanted, percutaneous
Implanted and
external
Implanted sensors,
voluntary function
Univ. Aalborg research

Various research surface systems
Elmetec A/S Footlifter
NeuroMotion, Inc.
Walkaide
Medtronic/Rancho
implant
Ljubljana implant
Univ. Aalborg
Research
Ljubljana hemiplegia
systems
Ljubljana SCI
systems
Sigmedics, Inc.
ParaStep
Cleveland VA/CWRU
research
Shriners Hospital
(Philadelphia)
Davis et al.
37,38
Percutaneous
External
Nerve recording
early work
Surface
External
commercial
commercial
Surface
Surface
19
17,18
38
Ness, Ltd.
Handmaster
Neuromotion, Inc. Tetron Glove
Tohoku Univ.
research
References
Electrode Type
Functions
Disorders
# Systems
SCI
Foot switch
Hand-grasp and release

Hand-grasp and release
Hand grasp and release, arm movements
Hand grasp and release, elbow extension, surgical reconstruction
Improved hand grasp,
bilateral function,
proximal arm functions
Hand grasp with slip
compensation
Foot drop
HP
250
External
External
Foot switch
Tilt sensor
Foot drop
Foot drop
HP, MS
HP, MS
3800
40
Implanted
Implanted
Foot switch
Foot drop
HP
Epineural
Surface
Implanted
External
Foot switch
Nerve recording
Foot drop
Foot drop
HP
HP, MS
16
Surface
External
2500
Surface
External
SCI
250
22
Surface
External
Standing, walking, foot

drop
Standing, walking, foot
drop
Standing, walking
HP
20,21
Foot switches, hand

switches
Foot switches, hand
switches
Hand switches
SCI
300
23,24,26,27
Implanted
Hand switches
50
Implanted
Hand switches
Standing, walking,
stair climbing
Standing, walking
SCI, HP
Betz and associates

28
External and
implanted
External, implanted
Implanted
SCI, HP
20
Laboratory
Standing
SCI
Implanted
Laboratory
Basic research
SCI
10
Implanted
External
External
Hand switches
Hand switches
Automatic sensorbased
Hand switches
Standing, walking
Standing, walking
Standing, walking
SCI
SCI
SCI
4
70
2
Standing, walking
SCI
LARSI
29
Vienna group
LSU-RGO II hybrid
Andrews et al. hybrid
30
31
32
Cleveland VA/CWRU
research hybrid
23,24
Implanted epineural
Implanted spinal
root
Epineural
Surface
Surface
Percutaneous, implanted
External, implanted
HP, SCI
120
HP, SCI
47
HP, SCI
25
SCI
85
SCI, HP
65
31
50
2
The numbers of systems listed are cumulative and not limited to current users. Some studies have been discontinued. Some subjects have used more than
one system and thus may be counted more than once. References may not contain most recent number of systems, which were in many cases obtained via
personal communication.
HP hemiplegia; SCI spinal cord injury; MS multiple sclerosis.
LARSI Lumbosacral anterior root stimulator implant; LSU-RGO Louisiana State University Reciprocating Gait Orthosis; CWRU Case Western Reserve
University; VA Dept. of Veterans Affairs.
groups have developed systems for overcoming foot drop, a

condition which often accompanies stroke or incomplete SCI
due to the inability of the ankle to dorsiflex and raise the toes
and feet off of the ground during the swing phase of gait. In
all of these systems, an external sensor (usually a contact
switch in the shoe) detects when the foot is off the ground,
triggering stimulation of the common peroneal nerve through
surface (16) or implanted (1719) electrodes. The peroneal
stimulation activates the muscles which produce ankle dorsiflexion, and also evokes a flexion withdrawal reflex (due to
excitation of sensory fibers within the nerve) which causes the
ankle, knee, and hip all to flex and further raise the foot off
the ground. At least two commercial systems are currently
available. The Footlifter (Elmetec A/S) is a two-channel surface system using a foot switch. The Walkaide system (Neuromotion, Inc.) uses a tilt sensor built directly into the stimulation unit. More than 5000 individuals have used foot-drop
neural prostheses.
Stimulation of the peroneal nerve has also been used as
part of a system for restoring standing and walking in individuals with hemiplegia or thoracic SCI (2022). The user
stands up using stimulation of the quadriceps muscles to lock
the knees, in combination with upper-body exertion. During
walking, the flexion withdrawal reflex is evoked alternately
in each leg by two channels of stimulation to allow the nonweight-bearing foot to clear the ground, substituting for the
344
NEURAL PROSTHESES
swing phase of gait. Quadriceps stimulation is used to lock

the knee during the weight support phase of gait. All forward
progression is provided by voluntary actions of the upper extremities, not by lower-extremity contractions, and a rolling
walker or crutches are required for support and stability.
Hand or foot switches are used by the subject to elicit the
stimulus pattern for each leg to produce the stepping movements. A commercial system based on these principles, the
Parastep device (Sigmedics, Inc.), is currently available (22).
Neural prostheses based upon surface stimulation cannot
access anatomically deep muscles, challenge the tolerance of
the user for applying the electrodes daily, and require repeatable electrode placement from day to day. The flexion reflex
also tends to decline in strength (accommodate) during the
repeated activation required during gait. Several groups,
most notably the Cleveland VACase Western Reserve University group (2327) and the satellite Shriners group have
developed lower-extremity neural prostheses based upon permanent percutaneous and/or implanted electrodes to address
these concerns. Forward progression is aided by active contractions in lower-extremity muscles, although all users employ a walker or crutches for stability and safety. Lower-extremity FNS systems employing percutaneous intramuscular
electrodes have been shown to restore standing, walking, and
stair climbing, as well as side- and back-stepping motions to
individuals with paraplegia. The percutaneous electrodes
used in these systems can be introduced in a minimally invasive manner via hypodermic needles (23). The use of implanted stimulators and electrodes surgically implanted
within or on the muscle, or on spinal roots or peripheral
nerves has been investigated by several groups (2530).
Restoration of gait using electrical stimulation is metabolically inefficient, and the number of motions that must be controlled during any lower-extremity action is typically more
than can be produced by the available number of stimulation
channels. All current lower-extremity neural prostheses use
external support of some kind, usually a rolling walker, but
several groups (3133) have combined the actions of FNS
with more extensive external braces such as reciprocating
gait orthoses (RGOs) to produce hybrid systems. The external
braces provide stability, while stimulated contractions of
lower-extremity muscles provide the power for walking and
standing up without assistance. Although RGOs (and other
braces) used with or without electrical stimulation are relatively inexpensive and are often successful in allowing users
to stand and walk with fairly low energy consumption, their
long-term usage rate has been low because they are difficult
to put on and take off, they are often cosmetically unacceptable to the users, and they work well only over flat, even surfaces.
Future Directions. Future clinically available movement
neural prostheses will incorporate the most useful features
that emerge from current research projects. Useful features
will be those that extend the benefits of neural prostheses to
individuals with different disabilities (e.g., higher level SCI,
incomplete SCI, stroke, cerebral palsy), provide additional
functions not currently available (weight shifting, transfers,
more dexterous hand function, proximal arm function, unassisted standing), enhance reliability, incorporate simpler and
more natural user interfaces, and reduce costs. Reliability of
movement neural prostheses will be enhanced by implanta-
tion of most of the system components (e.g., electrodes, stimulator, sensors for control). Lost-cost surface systems will be
effective in some individuals, but the effectiveness of neural
prostheses will continue to be greatly enhanced by reconstructive surgeries. Limitations in current stimulation technology, such as spillover and incomplete activation, may be
addressed by specialized nerve cuff electrodes (34) or other
approaches that selectively target motoneurons within the
nerve trunks rather than in the muscle. Routing leads to an
ever-increasing number of electrodes may be addressed by
leadless injectable electrodes (35) or other approaches which
do not require a separate lead wire from the stimulator to
each electrode. Control methods will integrate with and make
full use of retained voluntary control. Signals recorded from
natural sensory receptors in the paralyzed limbs (3638) and/
or from external or implanted artificial sensors will be used
both as command signals from the user and to implement
closed-loop or feedforward controllers (33,3943) to compensate for internal (e.g., fatigue) and external (e.g., unexpected
loads) disturbances.
Bladder and Bowel Function
Loss of control of bladder and bowel function occurs after spinal injury, and is one of the leading causes of morbidity and
mortality. Complications include frequent urinary tract infections, incontinence, damage to the upper urinary tract, and
constipation. A number of approaches using electrical stimulation to restore bladder function in individuals with spinal
cord injury have been attempted and are outlined in Fig. 6
(44,45). Many attempts have been hampered by direct or reflex activation of the urethral sphincter, which closes the outlet from the bladder, at the same time as the bladder is con-
Sacral spinal cord
(1)
(2)
Bladder
;
(4)
(5)
(3)
External
urethal
sphincter
Figure 6. Approaches to neural prosthetic control of the bladder include stimulation of the spinal cord (1), stimulation of the intradural
(2), and extradural (3) sacral nerve roots, stimulation of the pelvic
nerve (4), and direct stimulation of the bladder wall (5).
NEURAL PROSTHESES
tracting trying to force urine out, thus preventing the bladder

from emptying. However, significant clinical benefit to large
numbers of individuals has been achieved by electrical stimulation of the sacral nerves innervating the lower urinary
tract (46).
The first approach to bladder emptying is direct stimulation of the bladder wall. This technique met with limited success and has virtually been abandoned, although it may be
useful in cases of denervation of the bladder. The failure of
this approach was primarily due to the small region of the
bladder activated by direct stimulation and the difficulty of
creating stable and reliable electrical interfaces in contact
with the bladder wall. The second approach is direct stimulation of the pelvic nerves. While this would seem to be the
most logical method to generate selective activation of the
bladder, this approach has been hindered by difficult surgical
access to the nerve, difficulty in interfacing with the small,
branching pelvic nerve, and unwanted cocontraction of the
urethra, presumably due to activation of pelvic afferent fibers.
The third approach that has been attempted is direct stimulation of the spinal cord using penetrating electrodes. Pairs of
electrodes were implanted into the gray matter of the sacral
spinal cord to stimulate the preganglionic parasympathetic
innervation of the bladder. Good results were achieved in 16
of 27 patients followed for as long as 10 years, but no further
implants have been performed (47). Intraspinal microstimulation for control of bladder function is an active area of research and development (48,49).
The location where electrical stimulation has produced the
most success is the sacral spinal nerve roots, either intradurally on the ventral (motor) roots or extradurally on the combined root. The sacral roots contain the small-diameter preganglionic parasympathetic axons innervating the bladder via
the pelvic nerve and the larger diameter somatic motor axons
innervating the external urethral sphincter via the pudendal
nerve (Fig. 6). Since large fibers have a lower threshold for
excitation (Fig. 5), sacral root stimulation at low amplitudes
results in activation of the urethral sphincter, while higherstimulus amplitudes lead to contraction of both the bladder
and the urethral sphincter, leading to little or no voiding.
Several methods have been tested to overcome the coactivation of bladder and external sphincter caused by sacral root
stimulation including surgical transection of the pudendal
nerve, electrical block of pudendal nerve transmission, stimulation induced fatigue of the sphincter, and intermittent stimulation. Each approach has shortcomings, but intermittent
stimulation has achieved widespread success in emptying the
bladder via poststimulus voiding (44). This technique takes
advantage of the difference in the speed of contraction and
relaxation of the bladder and external urethral sphincter. The
bladder consists of smooth muscle and thus contracts and relaxes slowly, while the external urethral sphincter consists of
striated muscle and contracts and relaxes quickly. Intermittent stimulation (3 s to 6 s on, 9 s off) leads to sustained
contraction of the bladder but relaxation of the sphincter between stimuli. Thus urine passes in the interburst interval
and the bladder is emptied in spurts.
The technique of intermittent stimulation has been combined with transection of the sacral dorsal (sensory) roots to
provide effective bladder control in large numbers of individuals (46). In addition to the ability to urinate when desired,
this treatment also produces reductions in posturination re-
345
sidual volumes, urinary tract infections, bladder trabeculation, and vesicoureteral reflux, and increases bladder capacity and continence (44). Furthermore, since the lower bowel
also receives efferent innervation from the sacral roots, many
patients using the stimulator have an increased frequency of
defecation, a reduction in constipation and fecal impaction,
and a reduction in time spent defecating (44). Penile erection
is also achieved by stimulation in some male users.
The other technique under active investigation to prevent
coactivation of the bladder and external urethral sphincter is
selective stimulation of the small fibers innervating the bladder (50). Selective stimulation of small fibers may be achieved
by arresting action potentials in large fibers (51) or by elevating the threshold of large fibers above that of the small fibers
(52). These techniques should enable selective contraction of
the bladder or lower bowel without contraction of the external
sphincters, and thereby produce better emptying.
Restoration of Respiratory Function
Maintenance of respiratory function is essential for life.
Breathing provides the lungs with fresh air for the exchange
of oxygen with carbon dioxide in the blood so that all cells
of the body can function. Coughing is used to expel foreign
substances and normal secretions from the lungs and thus
prevents obstructions and infection. If respiratory muscle
function is inadequate, a mechanical respirator is often used
to force air into and out of the lungs. Although mechanical
ventilation can maintain life, the individual is continuously
dependent on the respirator, and its use can lead to infection
and bleeding around the tracheotomy site, trauma to the
bronchi within the lungs, and impaired speech and sense of
smell. Impaired cough can lead to obstruction of portions of
the lung and/or pneumonia.
If respiratory impairment results from inadequate activation of the motoneurons of the respiratory muscles, neural
prostheses based upon electrical stimulation are often a viable option for long-term respiratory maintenance. Such systems allow users to decrease or even discontinue the use of
mechanical ventilation, reducing its side effects and significantly enhancing their independence. More than 1000 individuals worldwide have been provided with neural prostheses
that control paralyzed diaphragm function via stimulation of
the phrenic nerve (5356). The primary applications for these
devices have been individuals with high-level cervical (C3 or
above) spinal cord injury, in whom the diaphragm is paralyzed while the phrenic motoneurons remain intact, and in
individuals with central alveolar hypoventilation, where the
brain fails to activate the muscles of respiration due to a
deficit in the chemoreceptors in the carotid body.
All commercially available phrenic pacing systems work
using the principals illustrated in Fig. 7. Electrodes are implanted upon the phrenic nerves, with lead wires from the
electrodes running under the skin to an implanted pacemaker-like device that generates the electrical stimuli. The
stimulators receive power and commands from an external
controller via an electromagnetic link. In all of these systems,
stimulus parameters are set by the clinician to achieve adequate and smooth recruitment of the diaphragm. Stimulation
of the diaphragm (see Fig. 7) acts to expand the volume of the
chest cavity, lowering chest cavity pressure below atmospheric pressure and resulting in air flow into the lungs. The
346
NEURAL PROSTHESES
Glottis
Spinal cord
Phrenic nerve electrode
Trachea
Lead wires
Figure 7. A schematic diagram of the respiratory system and a typical phrenic nerve pacing system. Inhalation is provided to individuals with paralyzed diaphragms by stimulation of the phrenic nerve (usually on
both sides) via an electrode implanted onto the nerve.
An implanted stimulator receives power and stimulus
commands from a small external unit via an electromagnetic link across the skin. Stimulated contractions of the
diaphragm pull it down into the abdomen, drawing air
in through the mouth and nose. Exhalation occurs passively when the diaphragm stimulation is withdrawn
and the elastic properties of the lungs and chest wall
force air out of the lungs.
Implanted stimulator
Skin-mounted antenna
External controller/transmitter
elasticity of the stretched lung tissue and surrounding chest

wall passively force the air back out during expiration when
the diaphragm is relaxed. System users are typically given
control over the number of breaths per minute and duration
of each breath to adjust for different levels of exertion. The
different phrenic nerve pacing systems vary mainly in the
type of electrode used and in the stimulation methods used to
rotate or alternate stimulation among portions of the diaphragm to reduce the danger of fatigue (55).
Current research is aimed at improving the function of
currently available neural prostheses, making them safer and
cheaper to install, and expanding the range of individuals
who might benefit. Work is in progress to make phrenic pacing systems fully implantable to reduce the danger of accidental uncoupling from the external control unit (55). Intramuscular electrodes inserted into the diaphragm (57) have been
proposed as alternatives to phrenic nerve pacing that have a
lower potential for damaging the phrenic nerve and may be
less invasive to implant. Intercostal (58) and/or abdominal
(59) muscle stimulation has been investigated as a method of
providing respiration in individuals with very weak or completely denervated diaphragm muscles, and as a method for
improving cough (60,61). Future work will likely focus on
making respiratory neural prostheses automatically responsive to metabolic demands and on using remaining nonparalyzed respiratory function to trigger stimulation in a natural
and volitional manner.
SENSORY NEURAL PROSTHESES
Auditory Prostheses
The neural prosthesis having the most widespread use and
largest numbers of users is the auditory prosthesis. The vast
majority of auditory prostheses are cochlear implants which
electrically activate the auditory nerve via electrodes implanted within the peripheral organ of the auditory system,
the cochlea. Cochlear implants are intended for individuals
who are sensorineurally deafthat is, the hair cells of the
cochlea, which transduce sound in the form of vibrations into
Diaphragm
(at rest)
Diaphragm
(during inspiration)
Ribs
Phrenic nerves
Left lung
Abdomen
neural action potentials, are impaired, while the nerve fibers

innervating the hair cells are largely preserved. The surviving auditory neurons, or spiral ganglion cells, can be stimulated to fire action potentials by application of stimuli of the
appropriate magnitude, duration, and orientation, and these
artificial action potentials evoke the perception of sound. For
individuals without an intact cochlear nerve, a central device,
with electrodes placed in the cochlear nucleus has also been
developed.
Modern cochlear implants consist of three primary components: (1) an external processor and transmitter, (2) an implanted receiver stimulator, and (3) an implanted electrode
array. The external processor and transmitter collects sound
signals with a microphone, processes them with an algorithm
to convert the most salient features of the sounds into a pattern of electrical stimuli, and transmits the appropriate command information to the implant. The implanted stimulator
receives the information, decodes it, and applies the appropriate stimulus current to the implanted electrode array.
Early cochlear implants used a single-channel electrode
and provided functional benefits to most users (62). Perceptual experiments, in which speech was simulated by means of
a small number of amplitude modulated single-tone generators, suggested that at least six channels were required for
speech comprehension (63). Therefore, modern cochlear implants use multiple-channel electrodes designed to take advantage of the tonotopic organization of the cochlea (64). The
cochlea is organized such that the basal region responds to
high-frequency sounds, while the apical region responds to
low-frequency sounds. This tonotopic arrangement converts
place information to frequency (or pitch) information and is
referred to as the place-pitch theory. Therefore, with a multiple electrode array, the perceived pitch is related to where
within the cochlea the stimulus was applied.
Multichannel cochlear implants, combined with modern
speech processors, have been remarkably successful, and are
currently implanted in over 20,000 individuals worldwide. In
the most successful cases these devices enable open speech
recognition without the aid of lip reading, and even use of
the telephone (65). However, the performance with the same
NEURAL PROSTHESES
devices implanted in different individuals varies widely

across individuals, and performance is also very dependent on
the users environment (66). The reasons for these differences
are not clear and are an area of active research. Future advances in cochlear implants will likely include new speechprocessing strategies, new arrays that place electrodes as
close as possible to the excitable fibers, and new stimulation
techniques that enable selective stimulation of discrete
groups of cochlear nerve fibers. Although the cochlea is organized tonotopically, the ability to stimulate selectively discrete groups of nerve fibers and thereby evoke different frequency percepts, is difficult due to current spread within the
cochlea. Bilateral implants are also likely to produce increases in function, as in hearing individuals the second ear
makes it possible to distinguish multiple sound sources by
their relative locations.
For individuals without an intact cochlear nervefor example due to removal of a tumora central device has also
been developed. This auditory prosthesis uses electrodes
placed on the surface of the cochlear nucleus, the first-order
projection site of auditory nerve fibers. These devices have
been used in far fewer individuals than cochlear implants, but
apparently provide similar performance and benefits (67,68).
Visual Prostheses
Electrical activation of neurons within the visual system is
currently the only technique to restore vision to profoundly
blind individuals. Experimental work on developing a visual
prosthesis was pioneered by Brindley and Lewin (69), who
implanted an array of 80 electrodes on the surface of the visual cortex in a blind volunteer. These experiments demonstrated that electrical stimulation of the cortical surface produced phosphenes (bright spots of light) in the visual field of a
blind volunteer, and that there was a topographical mapping
between the location of the stimulus on the cortical surface
and the location of the phophene in visual space. These experiments were subsequently replicated and extended to enable
a blind volunteer to recognize Braille letters spelled out using
sets of stimulation-evoked phosphenes (70,71). However,
these early experiments also indicated that a prosthesis of
only limited function could be produced with this technique
because of interactions between phosphenes produced by different electrodes, flicker of phosphenes, and persistence of
phosphenes following cessation of the stimulus, all attributed
to the poor selectivity and large numbers of neurons activated
by surface stimulation.
Therefore, an alternative approach, employing intracortical microstimulation with penetrating microelectrodes is being pursued. Fundamental studies on microstimulation of the
human visual cortex (72) were conducted in three sighted volunteers undergoing surgery to remove epileptic foci. The results of these studies demonstrated that thresholds to evoke
phosphenes were approximately two orders of magnitude
lower for intracortical stimulation than for surface stimulation, and that phosphenes evoked by depth stimulation were
steady, while surface evoked phosphenes tended to flicker.
Further studies on intracortical microstimulation for restoration of vision were conducted using an array of 38 microelectrodes implanted in the cortex of a blind volunteer for four
months (73). The results of these studies demonstrated that
intracortical microstimulation produced small, constant phos-
347
phenes in an individual who was blind for 22 years. Further,

intracortical electrodes separated by 500 m could evoke distinct phosphenes, a significant advance in spatial selectivity
as compared to surface stimulation.
An important element of a cortical visual prosthesis is a
high-density microelectrode array suitable for chronic implantation in the brain. Several investigators have recently demonstrated high-density arrays of electrodes, fabricated from
silicon using methods borrowed from integrated circuit manufacturing, which might be appropriate for creating a highdensity interface with the visual cortex (7476).
In addition to the ongoing work on a cortical visual prosthesis, there are also efforts to develop a retinal prosthesis
(77). These devices are intended to be implanted on the inner
surface of the retina and stimulate retinal ganglion cells (78)
to restore vision in individuals who have intact retinal ganglion cells, but damaged or diseased receptors, for example,
due to macular degeneration.
Electrocutaneous Stimulation
The visual and auditory prostheses described above used electrical stimulation to activate portions of the nervous system
devoted to these functions, and therefore directly produce the
sensations of sight or sound. In some conditions, however,
damage to the nervous system is such that direct stimulation
of the sensory neurons is either not possible or does not lead
to sensory perception. For example, blindness caused by damage of the visual cortex cannot be addressed by stimulation of
this compromised structure. In congenital blindness (where
vision was never present), the usual development of the visual cortex may not occur and electrical stimuli applied there
may not evoke sensations that can be interpreted in a visual
manner. It may also be advantageous to provide an individual
with information about the function of an artificial device,
such as the grip force produced by a myoelectric artificial arm
or the output commands from a hand-grasp neural prosthesis.
In such cases, electrical (or mechanical) stimulation of tactile
sensors in the skin has been investigated as a means to convey information about a different sensory modality, an approach called sensory substitution (79,80). Electrocutaneous
stimulation (i.e., electrical stimulation of tactile sensors in the
skin) in an area of the skin with intact sensation can be modulated by the variable of interest (e.g., grasp force) so that the
user interprets the stimulation in terms of this variable
rather than as tactile information. The information can be
coded through single electrodes as changes in stimulus amplitude, stimulus timing, or both. Multiple electrodes can be activated progressively as the variable of interest changes, or
arrays of electrodes can be used to provide information on
inherently multidimensional modalities such as audition and
vision.
Natural perception of tactile stimuli depend upon individual receptor properties, how the stimuli are spatially distributed across the skin, and how these stimuli change with time.
Current systems for electrocutaneous stimulation do not activate receptors within this normal context, so individuals using this approach must learn to interpret the tactile information provided in terms of the modality of interest. This has
proven to be problematic, and electrocutaneous stimulation
has been successfully applied only to a few problems. Blamey
and Clark (81) used electrocutaneous stimulation to provide
348
NEURAL PROSTHESES
auditory information to profoundly deaf individuals. Sabolich

and Ortega (82) used electrocutaneous stimulation to provide
center of pressure (a variable related to standing balance)
feedback to individuals with artificial lower limbs. The FreeHand hand-grasp neural prosthesis described previously is
implemented with one stimulus channel, providing an electrocutaneous signal related to the user command signal.
Future work will likely focus upon the factors that have
limited success to date. To maximize the transformation of
tactile information into the modality being restored, stimulating electrodes must provide more consistent and repeatable
inputs to the tactile system, perhaps by implantation. More
closely spaced electrodes may allow the tactile system to be
activated in a more natural spatial manner. Electrodes and
stimulus parameters also need to be optimized to reduce the
sensation of pain that often accompanies the tactile sensation
produced by electrocutaneous stimulation.
BIBLIOGRAPHY
1. P. H. Gorman and J. T. Mortimer, Effect of stimulus parameters
on recruitment with direct nerve stimulation, IEEE Trans. Biomed. Eng., 30: 407414, 1986.
2. W. M. Grill and J. T. Mortimer, Effect of stimulus pulse duration
on selectivity of neural stimulation, IEEE Trans. Biomed. Eng.,
43: 161166, 1996.
3. D. R. McNeal, Analysis of a model for excitation of myelinated
nerve, IEEE Trans. Biomed. Eng., 23: 329337, 1976.
4. L. S. Robblee and T. L. Rose, Electrochemical guidelines for selection of protocols and electrode materials for neural stimulation,
in W. F. Agnew and D. B. McCreery (eds.), Neural Prostheses:
Fundamental Studies, Englewood Cliffs, NJ: Prentice-Hall, 1990,
pp. 2566.
5. S. B. Brummer and M. J. Turner, Electrochemical considerations
for the safe electrical stimulation of the nervous system with
platinum electrodes, IEEE Trans. Biomed. Eng., 24: 5963, 1977.
6. W. F. Agnew et al., Histologic and physiological evaluation of
electrically stimulated peripheral nerve: Considerations for the
selection of parameters, Ann. Biomed. Eng., 17: 3960, 1990.
7. P. H. Peckham and M. W. Keith, Motor prostheses for restoration
of upper extremity function, in R. B. Stein, P. H. Peckham, and
D. B. Popovic (eds.), Neural Prostheses: Replacing Motor Function
After Disease or Disability, New York: Oxford Univ. Press, 1992,
pp. 162187.
8. T. Gordon and J. Mao, Muscle atrophy and procedures for training after spinal cord injury, Phys. Ther., 74: 5060, 1994.
9. R. B. Stein et al., Optimal stimulation of paralyzed muscle after
human spinal cord injury, J. Appl. Phys., 72: 13931400, 1992.
10. M. W. Keith et al., Tendon transfers and functional electrical
stimulation for restoration of hand function in spinal cord injury,
Hand. Surg. (Am.), 21: 8999, 1996.
11. R. H. Nathan, Control strategies in FNS systems for the upper
extremities, CRC Crit. Rev. Biomed. Eng., 21: 485568, 1993.
12. R. J. Triolo et al., Challenges to clinical deployment of upper limb
neural prostheses, J. Rehabil. Res. Dev., 33: 111122, 1996.
13. A. Prochazka et al., The bionic glove: An electrical stimulator garment that provides controlled grasp and hand opening in quadriplegia, Arch. Phys. Med. Rehabil., 78: 608614, 1997.
14. Y. Handa et al., Functional electrical stimulation (FES) systems
for restoration of motor function of paralyzed muscles-versatile
systems and a portable system, Front. Med. Biol. Eng., 4: 241
255, 1992.
15. B. Smith et al., An externally powered, multichannel, implantable stimulator for versatile control of paralyzed muscle,
16. A. R. Kralj et al., FES gait restoration and balance control in
spinal cord-injured patients, Prog. Brain Res., 97: 387396, 1993.
17. M. Kljajic et al., Gait evaluation in hemiparetic patients using
subcutaneous peroneal electrical stimulation, Scand. J. Rehabil.
Med., 24: 121126, 1992.
18. P. Strojnik et al., Treatment of drop foot using an implantable
peroneal underknee stimulator, Scand. J. Rehabil. Med., 19: 37
43, 1987.
19. D. R. McNeal, R. Waters, and J. Reswick, Medtronic implanted
system in 31 subjects. Experience with implanted electrodes,
Neurosurgery, 1: 228229, 1977.
20. A. Kralj and T. Bajd, Functional Electrical Stimulation: Standing
and Walking After Spinal Cord Injury, Boca Raton, FL: CRC
Press, 1989.
21. A. Kralj, R. Acimovic, and U. Stanic, Enhancement of hemiplegic
patient rehabilitation by means of functional electrical stimulation, Prosthetics & Orthotics Int., 17: 107114, 1993.
22. D. Graupe and K. H. Kohn, Transcutaneous functional neuromuscular stimulation of certain traumatic complete thoracic
paraplegics for independent short-distance ambulation, Neurol.
Res., 19: 323333, 1997.
23. E. B. Marsolais and R. Kobetic, Implantation techniques and experience with percutaneous intramuscular electrodes in the lower
extremities, J. Rehabil. Res. Dev., 23: 18, 1986.
24. E. B. Marsolais and R. Kobetic, Development of a practical electrical stimulation system for restoring gait in the paralyzed patient, Clin. Orthop. Relat. Res., 233: 6474, 1988.
25. R. Kobetic, R. J. Triolo, and E. B. Marsolais, Muscle selection and
walking performance of multichannel FES systems for ambulation in paraplegia, IEEE Trans. Rehabil. Eng., 5: 2329, 1997.
26. R. J. Triolo et al., Implanted Functional Neuromuscular Stimulation systems for individuals with cervical spinal cord injuries:
Clinical case reports, Arch. Phys. Med. Rehabil., 77: 11191128,
1996.
27. R. J. Triolo, R. Kobetic, and R. Betz, Standing and walking with
FNS: Technical and clinical challenges, in G. Harris (ed.), Human
Motion Analysis, New York: IEEE Press, 1996, pp. 318350.
28. R. Davis, W. C. MacFarland, and S. E. Emmons, Initial results
of the nucleus FES-22-implanted system for limb movement in
paraplegia, Stereotact. Funct. Neurosurg., 63: 192197, 1994.
29. D. N. Rushton et al., Lumbar root stimulation for restoring leg
function: Results in paraplegia, Artif. Organs, 21: 180182, 1997.
30. H. Kern et al., Functional electrostimulation of paraplegic patients1 years practical application. Results in patients and experiences (in German), Z. Orthop. 123: 112, 1985.
31. M. Solomonow et al., Reciprocating gait orthosis powered with
electrical muscle stimulation (RGOII), part I: Performance evaluation of 70 paraplegic patients, Orthopedics, 20: 315324, 1997.
32. B. J. Andrews et al., Hybrid FES orthosis incorporating closed
loop control and sensory feedback, J. Biomed. Eng., 10: 189
195, 1988.
33. D. B. Popovic, Functional electrical stimulation for lower extremities, in R. B. Stein, P. H. Peckham, and D. B. Popovic (eds.), Neural Prostheses: Replacing Motor Function After Disease or Disability, New York: Oxford Univ. Press, 1992, pp. 233251.
34. C. Veraart, W. M. Grill, and J. T. Mortimer, Selective control of
muscle activation with a multipolar nerve cuff electrode, IEEE
Trans. Biomed. Eng., 40: 640653, 1993.
35. T. Cameron et al., Micromodular implants to provide electrical
stimulation of paralyzed muscles and limbs, IEEE Trans. Biomed.
Eng., 44: 781790, 1997.
NEURAL PROSTHESES
349
36. J. A. Hoffer et al., Neural signals for command control and feedback in functional neuromuscular stimulation: A review, J. Rehabil. Res. Dev., 33: 145157, 1996.
58. A. F. DiMarco et al., Efficacy of combined inspiratory intercostal

and expiratory muscle pacing to maintain artificial ventilation,
Am. J. Respir. Crit. Care Med., 156: 122126, 1997.
37. M. Haugland et al., Restoration of lateral hand grasp using natural sensors, Artif. Organs, 21: 250253, 1997.
59. J. Sorli et al., Ventilatory assistance using electrical stimulation

of abdominal muscles, IEEE Trans. Rehabil. Eng., 4: 16, 1996.
38. T. Sinkjaer, M. Haugland, and J. Haase, Natural neural sensing

and artificial muscle control in man, Exp. Brain Res., 98: 542
545, 1994.
60. S. H. Linder, Functional electrical stimulation to enhance cough

in quadriplegia. Chest, 103: 166169, 1993.
61. A. F. DiMarco, J. R. Romaniuk, and G. S. Supinski, Electrical
activation of the expiratory muscles to restore cough, Amer. J.
Respir. Crit. Care Med., 151: 14661471, 1995.
62. W. F. House and K. I. Berliner, Safety and efficacy of the House/
3M cochlear implant in profoundly deaf adults, Otolaryngol. Clin.
North Amer., 19: 275286, 1986.
63. C. W. Parkins, Cochlear implant: A sensory prosthesis frontier,
IEEE Eng. Med. Biol., 2: 1827, 1983.
64. G .M. Clark et al., A multiple-electrode hearing prosthesis for
cochlear implantation in deaf patients. Med. Prog. Technol., 5:
127140, 1977.
65. I. Hongo and H. Takahashi (eds.), Cochlear implants and related
science update. Adv. Oto-Rhino-Laryngol., 52: 1997.
66. B. S. Wilson, The future of cochlear implants, Br. J. Audiol., 31:
205225, 1997.
67. R. V. Shannon et al., Auditory brainstem implant: II. Postsurgical issues and performance, Otolaryngol. Head Neck Surg., 108:
634642, 1993.
68. S. R. Otto et al., The multichannel auditory brainstem implant:
6-month coinvestigator results, Adv. Oro-Rhino-Laryngol., 52:
17, 1997.
69. G. S. Brindley and W. S. Lewin, The sensations produced by electrical stimulation of the visual cortex, J. Physiol. (London), 196:
479493, 1968.
70. W. H. Dobelle and M. G. Mladejovsky, Phosphenes produced by
electrical stimulation of human occipital cortex, and their application to the development of a prosthesis for the blind, J. Physiol.
(London), 243: 553576, 1974.
71. W. H. Dobelle et al., Braille reading by a blind volunteer by visual cortex stimulation, Nature, 259: 111112, 1976.
72. M. J. Bak et al., Visual sensations produced by intracortical microstimulation of the human occipital cortex, Med. Biol. Eng.
Comput., 28: 257259, 1990.
73. E. M. Schmidt et al., Feasibility of a visual prosthesis for the
blind based on intracortical microstimulation of the visual cortex,
Brain, 119: 507522, 1996.
74. D. J. Anderson et al., Batch fabricated thin-film electrodes for
stimulation of the central auditory system, IEEE Trans. Biomed.
Eng., 36: 693698, 1989.
75. P. K. Campbell et al., A silicon-based three dimensional neural
interface: Manufacturing processes for an intracortical electrode
array, IEEE Trans. Biomed. Eng., 38: 758767, 1991.
76. A. C. Hoogerwerf and K. D. Wise, A three-dimensional microelectrode array for chronic neural recording, IEEE Trans. Biomed.
Eng., 41: 11361146, 1994.
77. W. Roush, Envisioning an artificial retina, Science, 268: 637
638, 1995.
78. M. V. Narayanan et al., Development of a silicon retinal implant:
Cortical evoked potentials following focal stimulation of the rabbit retina with light and electricity, Invest. Ophthalmol. Visual
Sci., 35: 1380, 1994.
79. K. A. Kaczmarek et al., Electrotactile and vibrotactile displays
for sensory substitution systems, IEEE Trans. Biomed. Eng., 38:
116, 1991.
80. C. L. Van Doren, R. R. Riso, and K. Milchus, Sensory feedback for
enhancing upper extremity neuromuscular prostheses, J. Neurol.
Rehabil., 5: 6374, 1991.
39. P. E. Crago et al., New control strategies for neuroprosthetic systems, J. Rehabil. Res. Dev., 33: 158172, 1996.
40. H. J. Chizeck, Adaptive and nonlinear control methods for neural
prostheses, in R. B. Stein, P. H. Peckham, and D. B. Popovic
(eds.), Neural Prostheses: Replacing Motor Function After Disease
or Disability, New York: Oxford Univ. Press, 1992, pp. 298328.
41. J. J. Abbas and H. J. Chizeck, Feedback control of coronal plane
hip angle in paraplegic subjects using functional neuromuscular
42. A. Kostov et al., Machine learning in control of functional electrical stimulation systems for locomotion, IEEE Trans. Biomed.
Eng., 42: 54151, 1995.
43. P. H. Veltink, Control of FES-induced cyclical movements of the
lower leg, Med. Biol. Eng. Comput., 29: NS8N12, 1991.
44. G. H. Creasey, Electrical stimulation of sacral roots for micturition after spinal cord injury, Urol. Clin. North Amer., 20: 505
515, 1993.
45. N. J. Rijkhoff et al., Urinary bladder control by electrical stimulation: Review of electrical stimulation techniques in spinal cord
injury, Neurourol. Urodyn., 16: 3953, 1997.
46. G. S. Brindley, The first 500 patients with sacral anterior root
stimulator implants: General description, Paraplegia, 32: 795
805, 1994.
47. B. S. Nashold, H. Friedman, and J. Grimes, Electrical stimulation of the conus medullaris to control the bladder in the parplegic patient: A ten year review, Appl. Neurophysiol., 44: 225
232, 1981.
48. R. R. Carter et al., Micturition control by microstimulation of the
sacral spinal cord of the cat: Acute studies, IEEE Trans. Rehabil.
Eng., 3: 206214, 1995.
49. W. M. Grill and N. Bhadra, Genitourinary responses to microstimulation of the sacral spinal cord, Neurosci. Abstr., 22: 1842,
1996.
50. N. J. Rijkhoff et al., Selective detrusor activation by electrical
sacral nerve root stimulation in spinal cord injury, J. Urol., 157:
15041508, 1997.
51. Z.-P. Fang and J. T. Mortimer, Selective activation of small motor
axons by quasitrapezoidal current pulses, IEEE Trans. Biomed.
Eng., 38: 168174, 1991.
52. W. M. Grill and J. T. Mortimer, Inversion of the current-distance
relationship by transient depolarization, IEEE Trans. Biomed.
Eng., 44: 19, 1997.
53. W. W. L. Glenn et al., Twenty years of experience in phrenic
nerve stimulation to pace the diaphragm, Pacing Clin. Electrophysiol., 9: 780784, 1986.
54. J. I. Miller et al., Phrenic pacing of the quadriplegic patient, J.
Thorac. Cardiovasc. Surg., 99: 3540, 1990.
55. G. Creasey et al., Electrical stimulation to restore respiration, J.
Rehabil. Res. Dev., 33: 123132, 1996.
56. D. Chervin and C. Guilleminault, Diaphragm pacing for respiratory insufficiency, J. Appl. Neurophysiol., 14: 369377, 1997.
57. D. K. Peterson et al., Long-term intramuscular electrical activation of the phrenic nerve: Efficacy as a ventilatory prosthesis,
350
NEUROCONTROLLERS
81. P. J. Blamey and G. M. Clark, A wearable multiple-electrode electrotactile speech processor for the profoundly deaf, J. Acoust. Soc.
Amer., 77: 16191621, 1985.
82. J. A. Sabolich and G. M. Ortega, Sense of feel for lower-limb amputees: A phase-one study, J. Prosthetics Orthotics, 6: 3641,
1994.
WARREN M. GRILL
ROBERT F. KIRSCH
BIOIMPEDANCE
As an introduction to electrical impedance and conductance in biology, a review of the relevant terminology is
given and the scope of the discipline is presented. The
area of bioimpedance is broad, including, for example,
impedance cardiography, electrode impedance, impedance
spectroscopy, intraluminal conductance, and impedance tomography. The eld of bioimpedance deals with the electrical conduction properties of biological materials as a
response to the injection of current. It has been known
for more than two centuries that biological structures display the phenomenon of electrical conduction. Later, it
was found that the precise electrical properties of tissues
depend on their cellular composition and their coupling.
These characteristics imply that the voltage changes at a
particular site may provide valuable information regarding the biological materials and processes concerned.
However, to date, our understanding of the electrical
impedance of biological tissues and their changes, as far
as they are associated with physiological activity, is still
limited. This article discusses the application of electrical
impedance in medicine. Briey, bioimpedance can be used
to quantitate extracellular uid, to assess volume changes,
and as an imaging tool similar to ultrasonography. Reviews
can be found in (13), and (4).
In order to meet the requirements of specic applications, electrodes for delivering or recording electrical potentials in biological structures appear in a variety of materials, sizes, and shapes. The interface between electrode
and tissue has been studied extensively (5, 6).
IMPEDANCE CARDIOGRAPHY
Impedance cardiography (ICG) is the noninvasive measurement of physiologically and clinically relevant parameters of the heart and circulation, based on electrical
impedance measurements of the thorax during the cardiac cycle. Recent reviews have been presented by (7) and
(8). The technique uses a low-current (0.5 mA to 4 mA),
high-frequency (50 kHz to 100 kHz), alternating current
across the thorax, not perceivable to the subject. The resulting impedance changes associated with the cardiac cycle (impedance decreases by about 0.2 from diastole to
systole) provide information on stroke volume, cardiac output, pulmonary capillary wedge pressure (9), and systolic
time intervals (by calculating the rst time derivative of
the impedance waveform). In 1969, (10) suggested an index
of cardiac function based on particular calculations applied
to the impedance tracing. This so-called Heather index was
shown to correlate with the severity of cardiac pathology
(7).
Among other techniques to measure cardiac size (e.g.,
X ray, ultrasound, magnetic resonance imaging), external
impedance cardiography has the advantages of being noninvasive, requiring only relatively low-cost equipment, and
permitting continuous monitoring of signals originating
from the beating heart, even during exercise (11). Major
shortcomings result from the fact that a sound physical
model and a comprehensive theory still need to be devel-
oped. (12, 13), and (14) were among the rst researchers
to study the feasibility of the method. Chest impedance
is determined by the relatively constant electrical conduction properties of all tissues concerned, plus a modulated component caused by the combination of respiration, thoracic dimensional changes, and a cardiovascualr
size-related factor. The latter component is due to cyclic
changes of size (i.e., the geometry changes with contraction) of the four compartments of the heart and the major
blood vessels, as well as to the periodic alignment and deformation of the erythrocytes in the ow. Ejection of blood
from the heart distends the walls of the arteries, thus increasing their blood volume and resulting in an impedance
decrease, besides a decrease of lung resistivity owing to
blood perfusion. During relaxation of the cardiac ventricles, blood in the systemic circulation travels downstream,
causing a reduction of the arterial diameter while the erythrocytes lose their orientation owing to the lower velocity, all leading to an increase of the thoracic impedance.
Using suitable ltering techniques, it turns out that the
contributions derived from breathing and locomotion can
be eliminated. Obviously, respiratory components are simply removed if patients are instructed to temporarily hold
their breath, but in animals this approach would not be
feasible. Various ltering procedures have been developed,
including Fourier linear combiner (FLC) and event-related
transversal types. A major problem is that many physiological signals are quasiperiodic; that is, they have a mean
period with a small random variation around this mean at
each interval. (15) introduced a scaling factor to enhance
exibility when choosing lter parameters in the FLC approach. They successfully applied their method to ICGderived stroke volume (SV) in a volunteer during exercise.
Alternatively, one may apply an ensemble averaging technique to 20 beats, thus eliminating respiratory inuences
(16). The number and position of the electrodes employed
may vary depending on the specic purpose of the study
or the particular geometrical model assumed. Basically, a
four-electrode (tetrapolar) arrangement is employed: two
current injecting electrodes form the outer pair, and the
inner two are the sensing electrodes. This setup overcomes
not only impedance problems related to electrode polarization but also the relatively high skin impedance. A typical
conguration for ICG is illustrated in Fig. 1.
The amount of blood pumped per minute by one side
of the heart is termed cardiac output (CO) and equals the
product of heart rate (HR) and SV. In ICG it is the passage of blood in the major arterial vessel (called the aorta)
during the ejection phase of the ventricle that mainly determines the changes in electrical impedance. In contrast,
the electrocardiogram (ECG) is a recording of the electrical wavefront as it spreads over the cardiac tissues, and
this signal provides no information on the amount of blood
pumped nor does it give any insight into the size of the
ventricle or the strength of contraction. Therefore it is important to emphasize that the actual (internal) source of
electricity in the heart, namely, the action potentials that
cause a periodic voltage change on the order of 100 mV
over the cell membranes of the heart, are unrelated to the
electrical impedance variations resulting from the external stimulation electrodes and as recorded by the sensing
Bioimpedance
Figure 1. Electrode conguration for thoracic impedance cardiography: V, the voltage recording electrodes on the lateral base of
the neck and another pair on the chest at the level of the xyphoid;
C, constant current-injecting electrodes on forehead and at abdomen placed 15 cm caudally from the voltage electrodes.
electrodes. It may be concluded that the ECG and the thoracic impedance signal provide complementary information on the activity of the heart: the ECG on the internally
generated electricity, and the thoracic impedance on the
hemodynamic changes owing to the mechanical action of
the heart.
The electrical impedance signal is obtained from a special arrangement of disposable spot or band electrodes
placed on the skin of the head, neck, and chest. A set of
current-injecting electrodes is driven by a constant sinusoidal current of less than 1 mA root mean square at a
frequency ranging from 50 kHz to 100 kHz, while another
set of electrodes senses the resulting voltage from which
the impedance signal is calculated. The peripheral ECG is
usually recorded from the limb leads (Fig. 2).
Thoracic impedance ( Z) has a baseline component Z0
and a time-variable component ( dZ). Z0 depends on posture, tissue composition, and the volume of uids within
the chest. Cardiac edema causes a decrease of the value
for Z0 (17). In noncardiac edema (e.g., in the adult respiratory distress syndrome), Z0 can increase or decrease owing
to capillary leakage of proteins (18). The component dZ corresponds to the volume change in the thoracic aorta during
the cardiac cycle. The maximum value of the time derivative of Z is proportional to the peak ascending aortic blood
ow. (13) developed a formula to derive SV (in mL) from the
thoracic ICG:
where is the specic resistivity of blood (about 135 cm

and 150 cm for women and men, respectively), Tet the
duration of the ejection period (s), L the distance between
both recording electrodes (cm), and Z0 the baseline com-
Figure 2. Left ventricular (LV) volume as obtained by intraventricular impedance catheter in the open-chest dog. Also shown is
the ow in the aorta resulting from these volume changes during
an episode of irregular heart rhythm. (From Ref. (35) with permission.)
ponent of the thoracic impedance (). This approach has

been widely applied, sometimes after modication of the
expression to account for body build. Underlying assumptions for Kubiceks equation and subsequent modications
are that the tissues are modeled as a homogeneous electrical conductor with the shape of a cylinder or truncated
cone, and that the impedance variations observed in synchrony with the heartbeat exclusively reect the timevarying volumetric changes of the cardiovascular system
in the specic model considered. For the cylinder model,
the length is related to the height of the chest, while the
cross-sectional area reects the thoracic circumference; the
values of these parameters are obviously characteristic for
each individual at a given time. (19) determined the inaccuracy of Kubiceks one-cylinder model, or more generally
of any geometrical model with a uniform cross section, both
in a theoretical study and using in vivo experiments. The
results indicate that the one-cylinder model is not valid
and must be replaced by a model with two serially placed
cylinders, which appropriately exhibit a length-dependent
behavior corresponding to differences in body height and
mass. It was concluded that corrections for the Kubicek
equation are required, along with the incorporation of a
two-cylinder model. Interestingly, (20) found for the truncated cone model no gender effect on SV when normalized
for body surface, because the differences in the component
Bioimpedance
Z0 were counterbalanced by changes in dZ. However, both

in females and in males they documented a decline of normalized SV with age (range 20 years to 69 years).
As an extension of the routine approach to estimating
SV by thoracic impedance and using the Kubicek equation,
(9) also found a high correlation with the clinically important pulmonary capillary wedge pressure (PCWP), which
corresponds to left atrial pressure and thus reects left
ventricular lling. They also compared the noninvasively
obtained SV with values determined by thermodilution.
The study population consisted of 24 patients with cardiac
problems, including coronary artery disease and various
types of valvular defects. The PCWP as determined with a
pulmonary artery catheter correlated ( r = 0.92, p < .0001)
with the ratio obtained by dividing the amplitude of the
ICG during diastole (the O-wave) by the maximum value
during systole. It was hypothesized that this ratio reects
efciency of the left ventricle (LV) because the O-wave corresponds with preload, whereas the peak of dZ/dt is related
to afterload. A comparison between the two methods (i.e.,
impedance versus thermodilution) to estimate SV was less
encouraging ( r = 0.69, p < .05), particularly for patients
with valvular disease. As an extension, (21) assessed right
ventricular diastolic function in patients with chronic obstructive pulmonary disease by means of region of interest analysis applied to electrical impedance tomography,
as will be described later.
Of crucial importance, of course, is a quantitative validation of the thoracic impedance method, preferably using
a gold standard for the determination of SV. Unfortunately,
an ideal reference technique does not exist, and therefore researchers often rely on methods that are feasible
and minimally traumatic. For example, (22) recently performed a comparison of hemodynamic parameters derived
from transthoracic electrical bioimpedance with those obtained by thermodilution and ventricular angiography.
These investigators studied 24 human patients with coronary artery disease while employing the three independent methods and concluded that impedance cardiography should not replace invasive hemodynamic monitoring.
However, an accompanying editorial moderated this pessimistic conclusion by pointing to the expected impact from
emerging noninvasive echocardiographic Doppler technology.
As mentioned before, the physical size and shape of an
individual are related to the dimensions considered in any
geometrical model such as the cylinder or the truncated
cone. For a given adult these values may be expected not to
change very much, unless a disproportionate dimensional
change occurs within a relatively short time frame. In pregnancy, for example, maternal blood volume increases by
about 40%, while CO reportedly rises by some one-third
within nine months. Therefore, in longitudinal studies involving pregnant women, certain precautions may be indicated. (23) addressed the question of whether thoracic electrical bioimpedance is suitable for monitoring SV during
pregnancy. It was demonstrated that the Kubicek formula
for SV needs to be modied by a multiplication factor, which
for each woman is to be determined at the beginning of the
pregnancy period. Subsequently, (24) explored the possibility of relating CO to body surface area in women during the
course of their pregnancy. Based on the poor correlations

found for both impedance and Doppler echocardiographic
data, the authors concluded that normalized CO does not
offer any additional information relevant for comparative
evaluations during pregnancy and may even be misleading.
The availability of ICG as a measurement technique
that is noninvasive and permits continuous registration evidently opens the way to investigations of circadian effects.
(25) took advange of these properties of ICG and studied
sleep and circadian inuences on the cardiac autonomic
nervous system activity. They found that parasympathetic
activity (derived from respiratory sinus arrhythmia observed in the ECG) is mostly inuenced by the circadian
system, whereas sympathetic nervous system activity (assessed with the pre-ejection period of the impedance signal)
is predominantly affected by the sleep system.
Unfortunately, there exists no gold standard for cardiac size determination, because most techniques require
barely veried assumptions concerning anatomical geometry. A similar shortcoming also applies to SV, which refers
to the difference between diastolic and systolic volumes.
Another approach would be to compare SV as measured in
the aorta using a ow probe with the ICG signal; such as
experiment will be feasible in the chronically instrumented
animal and therefore deserves due attention.
Two types of problems related to application of the ICG
will be discussed. First is the type of electrodes employed,
that is, band (or tape type, consisting of a 6 mm conductive
foil strip) versus spot (or regional) electrodes. The band
signal is made up of different signals from various regions
and may reect changes in SV (rather than the absolute
value), whereas spot electrodes may be important in measuring regional physiological activities of the central circulation (11, 26). Also, the anatomical position of the electrode and movement can cause large artifacts that may
override the signal of primary interest (27). Second, the
commonly made assumption that aortic distension during
the ventricular ejection phase is the predominant component is not justied, because concomitant changes of lung
resistivity, ventricular contraction, and other factors may
interfere with comparable magnitude (26, 28). Obviously,
the current limitations preclude wide clinical acceptance
and certainly require further measurement renements of
the ICG method, but recently progress has been made in
this eld (1929).
CONDUCTANCE MEASUREMENTS
Besides the method of measuring external impedance
changes caused by the mechanical activity of the heart, it
is also technically possible to record conductance changes
within the lumen of each cardiac compartment. Since the
beginning of this century, scattered reports were published
about the recording of cardiac volume changes derived
from impedance measurements by placing electrodes on
the heart. (30) placed electrodes on the inside of the ventricular wall, but practical problems concerning the electrodes delayed progress in this eld. In 1961 the Brazilian dentist A. Mello-Sobrinho started experiments with a
Bioimpedance
bipolar arrangement, where electrodes mounted on pointed

rods were passed through the myocardium and sutured to
the apex and base of the left ventricle of a dog; he succeeded in producing intraventricular pressureimpedance
loops (31). Tremendous advances were made when multiple electrodes were mounted on a catheter, which can be positioned along the long axis of the ventricle. Because such a
catheter usually is introduced via a peripheral artery, there
is no longer a need to open the chest in order to carry out
the internal impedance measurements. The catheter technique has been pioneered by the research group of Baan
at Leiden University (3235) with applications in both animals and humans. Obviously, this approach implies an invasive procedure (called cardiac catheterization) but has
the advantage that geometrical variation associated with
the cardiac cycle can be measured at a segmental level.
The theoretical background of the impedance catheter, a
term that was coined in the early phases, and its application to cardiac volumetry have been described by (32).
In their model, the left ventricular cavity and the myocardial wall are represented by two confocal spheroids. The
wall is included because the impedance signal is permanently affected by a certain amount of current that leaks
into the ventricular wall and the surrounding extracardiac tissues. This offset component, the so-called parallel
conductance (34), can be determined by a relatively simple intervention, the injection of chglucose or hypertonic
saline (35). A new technique, using dual-frequency excitation, eliminates the need for such an intervention and
offers the additional advantage of providing continuous information on this offset component without physiological
interference (36). The method exploits the fact that muscle
and blood exhibit different conductivities depending on the
frequency employed. Muscle tissue is more conductive at
frequencies above 12 kHz when the current is applied in a
direction normal to the ber orientation (37), while the conductivity of blood is essentially constant within the range
of 2 kHz to 100 kHz. While this technique elegantly solves
problems inherent in parallel conductance for a particular
heart under investigation, the question whether parallel
conductance is independent of left ventricular size remains
controversial. (38) used the conductance catheter for the
measurement of left ventricular volume in the intact dog
to study whether parallel conductance is independent of
left ventricular size, that is, if it is constant throughout the
cardiac cycle. In contrast to the ndings of other investigators, these authors concluded that parallel conductance
varies between animals, but within any given animal parallel conductance does not decrease with left ventricular
size, but rather remains constant.
APPLICATIONS OF CONDUCTANCE METHODS IN

VARIOUS SPECIES
While the dog is historically employed as a standard animal
in cardiovascular research, today we witness conductance
applications in a variety of species. Regarding the canine
heart an example is given in Fig. 3.
(39) applied the conductance catheter to the left ventricle of newborn lambs and demonstrated a clear interaction
between afterload and contractility. Preload and afterload

were varied by inating a balloon catheter positioned in the
posterior vena cava and the thoracic aorta, respectively. In
these anesthetized neonates the authors found evidence of
homeometric autoregulation, enabling the heart to maintain SV at different levels of afterload. Both (40) and (41)
developed a microconductance device ( 8 mm 2 mm 0.5
mm) and a 3F catheter, respectively, so as to accommodate
the standard number of six to eight electrodes of a probe
suitable for use in the rat heart.
As a variant to the dilution technique, (42) measured
CO in small laboratory animals using recordings of blood
conductivity. A 5% glucose solution was injected as a bolus via the femoral vein of mice and rats. The changes of
blood conductivity resulting from this perturbation were
recorded by an intra-aortic probe with platinum electrode
combined with another one positioned within the rectum,
and this signal was used to calculate CO. Other investigators previously applied the method in dogs, and for these
small animals an excellent correlation ( r = 0.97, p < .001)
with the classical type of the indicator dilution method was
observed.
(43) were the rst investigators to simultaneously measure left ventricular (LV) pressure ( P) and volume ( V)
in the horse. Thus far, the conductance catheter was applied to obtain V for clinical evaluations in human patients,
and also for experimental investigations in animals. While
there is a tendency to miniaturize catheters and study
small species such as rats, larger animals like the horse
have not been evaluated with this instrument. They investigated the PV relationship in three anesthetized ventilated horses, with body mass ranging from 250 kg to 600
kg. Introduced via the carotid artery, they employed a 7F
Millar P-catheter and a large-size conductance catheter
(outer electrodes 20 cm apart, with four equidistant sensing electrodes) built at their institution. Respiration in
terms of chest wall excursion was externally recorded using
a mercury-in-silastic gauge placed around the chest. Reliable measurements were obtained including LV volume at
segmental levels, while PV loops could be constructed. An
example is given in Fig. 4. In these preliminary experiments calibration procedures were not yet carried out, and
therefore V data are presented as arbitrary units (A.U.).
Yet, the relative values and the time relationships with
left ventricular pressure and the pneumogram are still applicable and informative.
This pilot study demonstrates the feasibility of performing P and V measurements in the adult equine model. The
slow heart rate and the large LV dimensions in the horse
permit detailed analysis of the hemodynamic consequences
of asynchronous contraction patterns, while fully employing the potential of the conductance catheter to measure
LV contraction at segmental levels.
AORTIC CROSS-SECTIONAL AREA MEASUREMENT

Arterial transverse dimensions change with the cardiac cycle, periodically resulting in extension during the ejection
phase followed by relaxation during the diastolic phase of
the ventricle. These varying dimensions can potentially be
Bioimpedance
Figure 3. Left ventricular pressurevolume loops recorded in the anesthetized horse. (From Ref.
(43), with permission.)
Figure 4. Variations of left ventricular and atrial cross-sectional area in the magnetic resonance imaging (MRI) and the electrical
impedance tomography (EIT) images made with the person in the supine position. (From Ref. (29), with permission.)
measured using the electrical impedance technique with

electrodes positioned within the lumen of the blood vessel. Actually, various methods have been developed to assess continuously the aortic cross-sectional area (CSA), including the intravascular ultrasound (IVUS) method. (44)
reported in 1979 the feasibility of the impedance method
for determining CSA. They noted that movement of the
catheter toward the vessel wall consisted of a calculated
error (maximally 7%), which is considered to be of minor
importance. (45) rened the method and compared the ndings with results obtained from IVUS (range 35 mm2 to 70
mm2 ), yielding a correlation coefcient of 0.97 for 53 data
points collected from experiments in ve piglets.
tachometer have the disadvantage that they require insertion into the airway. The four-electrode arrangement uses
an electrode on each wrist, to which a constant 10 kHz
low-intensity current is applied, and an electrode on each
arm to record the impedance changes. As an alternative,
one may measure variations in electrical resistance of a
mercury-in-silastic-rubber gauge mounted around the thorax. Using a Wheatstone bridge it is rather easy to follow
the periodic alterations caused by respiratory movements
of the chest (Fig. 5). For further details see the paper by
(47).
APNEA MONITORING
IMPEDANCE PNEUMOGRAPHY
The plethysmographic technique employs a device that
records respiratory excursions from movements of the
chest surface on the basis of electrical impedance variations. Originally, the method was applied to the detection of
apnea (i.e., suspension of respiration) in newborns, and for
tracking changes in intrathoracic uid accumulation (46).
Other approaches such as the spirometer and the pneumo-
Sleep apnea is dened as temporary interruption of airow

to the lungs during the sleeping period and lasting for more
than ten seconds. This abnormality usually results either
from upper airway collapse (obstructive type) or, in 10% of
the cases, from the absence of diaphragmatic contraction
due to the lack of neural input (the central type) from the
brain, although mixed types do occur. Such conditions are
usually associated with loud snoring, while general symptoms and signs include fatigue, hypersomnolence during
Bioimpedance
Figure 5. ColeCole diagram for an impedance with a single time constant.
Figure 6. Variations of left ventricular and atrial cross-sectional area in the magnetic resonance imaging (MRI) and the electrical
impedance tomography (EIT) images made with the person in the supine position. (From Ref. 29, with permission.)
Bioimpedance
the day, cardiac problems, and hypertension. Snoring may

lead to soft tissue damage, notably an edematous uvula.
Sudden death due to sleep apnea is often related to comorbidity and is probably rare. The standard measurement
in sleep apnea includes the polysomnography repertoire,
which consists of recording ECG, electroencephalogram,
electro-oculogram, electromyogram of the legs, measurement of oral airow with the use of thermistors, esophageal
pressure, recording of snoring sound intensity, and plethysmography to estimate movements of chest and abdomen,
besides the determination of arterial oxygen saturation.
Typical costs associated with these studies amount up to
$1000 or even more. Treatment may include simple advice
to lose weight and avoid taking alcoholic drinks or sleeping
pills but may also consist of various surgical procedures or
continuous positive airway pressure (CPAP).
Apnea monitoring is based on either the measurement
of bioelectric impedance or on estimation of biopneumatic
impedance. Obviously, the time pattern of respiration is of
primary interest for monitoring in these cases, and quantitative volume changes are of less importance, thus permitting simple instrumentation. The bioelectric technique
is virtually identical to the ICG method, but in the present
case the periodic circulatory components are regarded as
perturbations superimposed on the slower respiratory signal. This signal is recorded with electrodes in the midaxillary line, and it may typically amount to 1 per liter lung
volume change, increasing with inspiration. The cardiac
contribution (usually about 20% of the total amplitude)
can (at least in humans) be determined during voluntary
temporary respiratory interruption. Movement of the body
causes artifacts, which may induce impedance changes in
excess of the signal under investigation. An area that may
be explored in this eld concerns measurement of the motion of the diaphragm: with well-positioned electrodes the
motion of this good conductor is easily assessed, while the
changes of the signal are clearly related to respiration.
TISSUE CHARACTERIZATION: TWO- AND

FOUR-ELECTRODE SYSTEMS
Electrical parameters of biological material are found
to correlate with tissue structure and their (patho)physiological changes (48). As mentioned before, muscle
and blood exhibit different conductivities depending on the
frequency employed. Muscle tissue is more conductive at
frequencies above 12 kHz when the current is applied in a
direction normal to the ber orientation (37). Either two- or
four-electrode systems have been employed for this type of
tissue characterization, in combination with a bridge technique or phase sensitive detector method (using a lock-in
amplier). Although simple in design, the two-electrode
system is less useful because of the presence of a charge
distribution at the interface of the metal electrode and the
tissue. However, this undesired effect of electrode polarization can be eliminated by the use of the four-electrode
technique, with separate pairs of electrodes for current injection and for sensing the resulting potential difference
(49). To calculate resistivity of a particular sample volume, it is necessary to select an electrical volume conduc-
tor model that sufciently describes the geometry and conductive properties of the medium under study. Often, the
medium can be approximated as an innite or semi-innite
structure, or as a thin layer bounded by air on both sides.
Anisotropy is an important but complex aspect of the dielectric properties of tissue (2651), which may be accommodated by considering a multilayered volume conductor
model. (52) determined local anisotropic resistivity of canine epicardial tissue (i.e., the outer muscle layer of the
ventricular wall) in vivo in two orthogonal directions, with
special attention to sample volume. The effective sample
volume for a simple homogeneous isotropic medium primarily depends on the distance between the current electrodes, but for real anisotropic media these researchers
found that in addition both longitudinal and transverse
resistivity (cm) varied not only during the cardiac cycle
but also depended on the driving frequency studied (which
was between 5 kHz and 60 kHz). Finally, it must be emphasized that electrode construction affects accuracy (51).
IMPEDANCE TOMOGRAPHY
This is a technically advanced approach whereby the imaging of an object is realized from measurements in multiple
directions. Usually, 16 to 32 electrodes are placed equidistantly in a plane around the patients. This yields anatomical slices or sections, which can be viewed from various angles. Reasonably good soft tissue contrast can be achieved
by impedance imaging, because of the different electrical
resistivities of the various tissues. Impedance images are
inferior to alternative techniques such as computed tomography and magnetic resonance imaging (MRI). Due to the
three-dimensional spread of current into the object, the
slice thickness cannot be conned to 1 mm or 2 mm. The
strength of impedance tomography, however, resides in its
functional imaging capabilities. Functional imaging is possible if variations in tissue resistivity are associated with
particular physiological events. The rst in vivo impedance
tomography images were produced in 1983 at the University of Shefeld, and the theoretical background as well
as illustration have been summarized by (53). A newer example has already been mentioned, when the right atrium
was selected as the region of interest, and compared with
results from MRI. (21) noninvasively assessed right ventricular diastolic function in patients with chronic obstructive pulmonary disease and in controls by means of region
of interest analysis applied to electrical impedance tomography. Comparison with MRI data showed a correlation of
r = 0.78 (n = 15), while pulmonary artery pressure (measured by right-sided heart catheterization) yielded an exponential relationship with r = 0.83 (p < .001). The same
authors (29) also improved cardiac imaging in electrical
impedance tomography by means of a new electrode conguration, whereby the traditional transversal positioning
at the level of the fourth intercostal space on the anterior
side was replaced by attachment at an oblique plane at the
level of the ictus cordis anteriorly and 10 cm higher posteriorly. Comparison with MRI ndings gave good results
(Fig. 6), while the reproducibility coefcient was 0.98 at
rest and 0.85 during exercise.
Bioimpedance
BIBLIOGRAPHY
Figure 7. ColeCole diagram for an impedance with a single time

constant.
IMPEDANCE SPECTROSCOPY
As stated before tissues can be considered as composites
of cells surrounded by extracellular uids. Each cell has
a cell membrane that encloses intracellular uid and consists of a thin layer of lipoproteins (6 nm). At low frequencies (<10 kHz) the cell membranes have relative high
resistances, and the current is conducted mainly by the
extracellular uid. At high frequencies the membrane capacity causes a decrease of membrane impedance so that
the current ows through the cells. The electrical behavior
of a cell can be modelled by a membrane capacitance parallel to a membrane resistance in series with pure passive
resistive elements, representing extra- and intracellular
uids. Coupling of several such modelled cells results in 2or even 3-dimensional models of tissue. For the measurement of the frequency dependency of tissues two methods
are often used, a resistance R in series with a reactance X
or a conductance G parallel to a capacitance C. The complex impedance Z in the rst approach is given by Z = R +
jX. In the second approach the admittance Y is given by Y
= G + j C. As the complex impedance is the reciprocal of
the admittance the following relations hold R = G/(G2 + 2
C2 ) and X = C/(G2 + 2 C2 ).
The complex series impedance (R + jX) can be visualized
in a diagram, in which the real component R is plotted
versus the imaginary component X. In the literature this
plot is often called the ColeCole diagram (54). In Fig. 7
the ColeCole diagram is given for a single time constant,
where the impedance as a function of frequency is given by
with R0 the resistance at f = Hz and = CR a time

constant.
The diagram is a semicircle with a radius ( r), r = (R0
R )/2 and points of intersection with the horizontal axis
R0 and R .
Although the complex series impedance of tissue is
often plotted as a ColeCole diagram, in practice most
tissues are best described with a semicircle with center
slightly under the horizontal axis. Changes in tissue can be
caused by changes in the amount of extra- and intracellular
uid, changes in tissue composition (e.g., tissue growth, ischemia, infarction, tumors, increase of adipose tissue) and
can be measured and visualized in such plots (55, 56). This
type of research is called impedance spectroscopy.
1. B. C. Penney Theory and cardiac applications of electrical

impedance measurements, CRC Crit. Rev. Biomed. Eng., 13:
227281, 1986.
2. J. J. Ackman A. Seitz, Methods of complex impedance measurements in biologic tissue, CRC Crit. Rev. Biomed. Eng., 11:
281311, 1984.
3. K. R. Foster P. Schwan, Dielectric properties of tissues and
biological materials: A critical review, Crit. Rev. Biomed. Eng.,
17: 25104, 1989.
4. L. A. Geddes L. E. Baker Principles of Applied Biomedical
Instrumentation, 3rd ed., New York: Wiley, 1989.
5. H. P. Schwan Electrode arization impedance and measurements in biological materials, Ann. N.Y. Acad. Sci., 148:
191209, 1968.
6. Y. Kim P. H. Schimpf Electrical behavior of debrillation and
pacing electrodes, Proc. IEEE, 54: 446456, 1996.
7. H. D. Fuller Evaluation left ventricular function by impedance
cardiography: A review, Prog. Cardiovasc. Dis., 36: 267273,
1994.
8. E. Raaijmakers et al meta- analysis of published studies concerning the validity of thoracic impedance cardiography, Crit.
Care Med., 27: 12031213, 1999.
9. H. H. Woltjer et, Prediction of pulmonary capillary wedge
pressure and assessment of stroke volume by noninvasive
impedance cardiography, Am. Heart J., 134: 450455, 1997.
10. L. W. Heather A parison of cardiac output values by the
impedance cardiograph and dye dilution techniques in cardiac
patients, NASA CR. 10 N70-10015: 247258, 1969.
11. R. P. Patterson L. Wang S. B. Raza Impedance cardiography
using band and regional electrodes in supine, sitting, and during exercise, IEEE Trans. Biomed. Eng., 38: 393400, 1991.
12. F. H. Bonjer Circulatieonderzoek door impedantiemeting (in
Dutch), Thesis, University of Groningen, The Netherlands,
1950.
13. W. G. Kubicek, et al. Development and evaluation of
an impedance cardiac output system, Aerospace Med, 37:
12081212, 1966.
14. J. Nyboer Electrical Impedance Plethysmography, Springeld,
IL: Charles C. Thomas, 1970.
15. A. K. Barros N. Ohnishi MSE behavior of biomedical eventrelated lters, IEEE Trans. Biomed. Eng., 44: 848855, 1997.
16. D. W. Kim C. G. Song M. H. Lee, A new ensemble averaging
technique in impedance cardiography for estimation of stroke
volume during treadmill exercise, Frontiers Med. Biol. Eng.,
4: 171178, 1992.
17. E. Raaijmakers, et al. he inuence of extracellular lung water on cadiac output measurements using thoracic impedance
cardiography, Physiol. Meas., 19: 491499, 1998.
18. E. Raaijmakers, et al. Estimation of non-cardiogenic pulmonary oedema using dual-frequency electrical impedance,
Med. Biol. Eng. Comp., 36: 461466, 1998.
19. E. Raaijmakers, et al. The inaccuracy of Kubiceks onecylinder model in thoracic impedance cardiography, IEEE
Trans. Biomed. Eng., 44: 7076, 1997.
20. G. Metry, et al. Gender and age differences in transthoracic
bioimpedance, Acta Physiol. Scand. 161: 171175, 1997.
Bioimpedance
21. A. Vonk Noordegraaf, et al. Noninvasive assessment of right
ventricular diastolic function by electrical impedance tomography, Chest, 111: 12221228, 1997.
22. P. E. Marik J Pendelton R. Smith, A comparison of hemodynamic parameters derived from transthoracic electrical
bioimpedance with those parameters obtained by thermodilution and ventricular angiography, Crit. Care Med., 25:
15451550, 1997.
23. R. M. Heethaar, et al. Thoracic electrical bioimpedance: Suitable for monitoring stroke volume during pregnancy? Eur. J.
Obstet. Gynecol. Reprod. Biol., 58: 183190, 1995.
24. A. C. C van Oppen, et al. Use of cardiac index in pregnancy: Is
it justied? Am. J. Obstet. Gynecol., 173: 923928, 1995.
25. H. J. Burgess, et al. Sleep and circadian inuences on cardiac autonomic nervous system activity, Am. J. Physiol., 42:
H1761H1768, 1997.
26. L. Wang R Paterson Multiple sources of the impedance cardiogram based on 3-D nite difference human thorax models,
27. M. Qu et al. Motion artifact from spot and band electrodes
during impedance cardiography, IEEE Trans. Biomed. Eng.,
33: 10291036, 1986.
28. D. W. Kim Detection physiological events by impedance, Yonsei
Med. J., 30: 111, 1989.
29. A. Vonk Noordegraaf, et al. Improvement of cardiac imaging in
electrical impedance tomography by means of a new electrode
conguration, Physiol. Meas., 17: 179188, 1996.
30. R. F. Rushmer, et al. Intracardiac impedance plethysmography, Am. J. Physiol., 174: 171, 1953.
31. M. E. Valentinuzzi J. C. Spinelli Intracardiac measurements
with the impedance technique, IEEE Eng. Med. Biol. Mag., 8
(1): 2734, 1989.
32. G. Murand J Baan Computation of the input impedance of a
catheter for cardiac volumetry, IEEE Trans. Biomed. Eng., 31:
448453, 1984.
33. J. Baan et al. Continuous stroke volume and cardiac output
from intraventricular dimensions obtained with impedance
catheter, Cardiovasc. Res., 15: 328334, 1981.
34. J. Baan et al. Ventricular volume measured from intracardiac
dimensions with impedance catheter: Theoretical and experimental aspects, in T. Kenner, R. Busse, and H. HingkeferSzalkay (eds.), Cardiovascular System Dynamics, New York:
Plenum, 1982, pp. 569579.
35. P. L. M. Kerkhof End-systolic volume and the evaluation of
cardiac pump function, Thesis, Leiden University, 1981.
36. T. J. Gawne K Gray R. E. Goldstein Estimating left ventricular
offset volume using dual-frequency conductance catheters, J.
Appl. Physiol., 63: 872876, 1987.
37. E. Zheng S. Shao J. G. Webster Impedance of skeletal muscle
from 1 Hz to 1 MHz, IEEE Trans. Biomed. Eng., 31: 477481,
1984.
38. R. S. Szwarc, et al. Conductance catheter measurement of left
ventricular volume in the intact dog: Parallel conductance
is independent of left ventricular size, Cardiovasc. Res., 28:
252258, 1994.
39. R. J. M. Klautz, et al. Interaction between afterload and contractility in the newborn heart, J. Am. Coll. Cardiol., 25:
14281435, 1995.
40. P. Schiereck, et al. Direct recording of EDP-EDV relationship
in isolated rat left ventricle: Effect of diastolic crossbridge formation, Cardiovasc. Res., 28: 715719, 1994.
41. H. Ito, et al. Left ventricular volumetric conductance catheter
for rats, Am. J. Physiol., 270: H1509H1514, 1996.
42. J. Vogel Measurement of diac output in small laboratory animals using recordings of blood conductivity, Am. J. Physiol.,
273: H2520H2527, 1997.
43. P. L. M. Kerkhof Combination of Millar and conductance
catheter for the estimation of left ventricular function in the
equine heart, Proc. 21st Int. Conf., IEEE Eng. Med. Biol. Soc.,
1999.
44. J. E. Axenborg B. Olsson, An electrical impedance method
for measurement of aortic cross-sectional areas, Proc. XII Int.
Conf. Med. Biol. Eng., Jerusalem, 1979, P. 48.1.
45. L. Kornet Extension and imrovements of the electrical conductance method, Thesis, Erasmus University, Rotterdam, The
Netherlands, 1996.
46. L. E. Baker Applications the impedance technique to the respiratory system, IEEE Eng. Med. Biol. Mag., 8 (1): 5052, 1989.
47. P. L. M. Kerkhof Beat-to-beat analysis of high-delity signals
obtained from the left ventricle and aorta in chronically instrumented dogs, Automedica, 7: 8390, 1986.
48. I. Giaever C. R. Keese A morphological biosensor for mammalian cells, Nature, 366: 591592, 1993.
49. R. Plonsey R. C. Barr The four-electrode resistivity technique
as applied to cardiac muscle, IEEE Trans. Biomed. Eng., 29:
541546, 1982.
50. B. R. Epstein K. R. Foster Anisotropy in the dielectric properties of skeletal muscle, Med. Biol. Eng. Comput., 21: 5155,
1983.
51. Y. Wang, et al. Geometric effects on resistivity measurements
with four electrode probes in isotropic and anisotropic tissues,
52. P. Steendijk et al. The four-electrode resistivity technique
in anisotropic media: Theoretical analysis and application
to myocardial tissue in vivo, IEEE Trans. Biomed. Eng., 40:
11381148, 1993.
53. B. M. Eyuboglu B. H. Brown D. C. Barber In vivo imaging
of cardiac related impedance changes, IEEE Eng. Med. Biol.
Mag., 8 (1): 3945, 1989.
54. K. S. Cole Membranes, Ions and Impulses, Ions and Impulses,
Berkeley: Univ. California Press, 1972.
55. M. Gheorghiu E. Gersing E. Gheoghiu Quantitative analysis
of impedance spectra of organs during ischemia, Proc. 10th
Int. Conf. Electrical Bio-impedance, 1998, pp. 7376.
56. H. Schafer,
et al. Dielectric properties of skeletal muscle during ischemia in the frequency range from 50 to 200 Hz, Proc.
10th Int. Conf. Electrical Bio-impedance, 1998, pp. 7780.
Further reading:
S. Grimnes & . G. Martinsen:Bioimpedance and Bioelectricity
Basics, Academic Press (2000). ISBN 0-12-303260-1
D. S. Holder:Electrical Impedance Tomography.Institute of
Physics Publishing (2005). ISBN 0-7503-0952-0
Relevant URLs:
Oslo Bioimpedance Group: http://www.fys.uio.no/elg/bioimp/
Thoracic electrical bioimpedance: http://www.aetna.com/cpb/data/
CPBA0472.html
Clinical applications and guidelines: http://www.regence.com/
trgmedpol/medicine/med33.html
10
Bioimpedance
Animation:
Thorax cross section: http://www.cplire.ru/html/tomo/eitimage.
html
Software for impedance measurement:

http://www.elin.ttu.ee/BME-Wg/R&D/BImpMeas/SOFTWARE.
HTM
http://www.sciencemag.org/cgi/reprint/224/4655/1355.pdf
PETER L. M. KERKHOF
ROBERT M. HEETHAAR
Vrije Universiteit Medical
Center, Amsterdam,
the Netherlands
Vrije University, Amsterdam,
the Netherlands
J. Webster (ed.), Wiley Encyclopedia of Electrical and Electronics Engineering

c 1999 John Wiley & Sons, Inc.
Copyright
HYPERTHERMIA THERAPY
The normal range of body temperature of human beings is maintained at a relatively stable temperature
near 37 C. Organs and tissues function most efficiently at this range. Temperature elevation even a few
degrees above this norm is associated with varying levels of biological responses. Hyperthermia is the term
used to describe significant departure of tissue temperature from the usual limit (40 C) encompassed by
thermoregulatory activity. Its use for therapeutic purposes has expanded in recent years to include a variety
of abnormal conditions. Investigations to date have shown that while hyperthermia can produce whole-body
(regional) and local tissue modifications for effective therapy, temperatures at which the desired tissue response
occurs vary over a wide range. Moreover, final tissue temperature is a complex function of energy deposition,
blood flow, and heat conduction in tissue.
Hyperthermia has been used therapeutically very early in human history. However, aside from a few
well-established medical applications, hyperthermia is still in a relatively early stage of development. Current medical applications fall into three broad categories: musculoskeletal conditions, cancer treatment, and
coagulative ablation therapy. An important aspect of its development is the production of adequate temperature distribution in the target tissue, superficial or deep-seated. Moreover, successful hyperthermia therapy
requires not only a suitable energy source for heat production, but also an understanding of the underlying
pathological condition being treated to define the critical target temperature as well as the ability to reach that
tissue with the heating modality. Energy sources that can be used for hyperthermia include ultrasonic wave
and electromagnetic field and radiation as well as conducted heat or convection.
Diathermy for Musculoskeletal Conditions

Therapeutic heating of musculoskeletal tissues by conversion of electromagnetic and ultrasonic energies in
deep-lying tissues without excessive heating of the skin is known as diathermy (through heat). It has been
the dominant clinical application until the 1970s. Diathermy modalities in use include spot focus ultrasonic
transducers that operate at 1 MHz. In the United States, the most prevalent electromagnetic modalities are the
shortwave inductive-coil diathermy operating at 27.12 MHz and the microwave (corner reflector or aperture)
diathermy operating at 2450 MHz. The frequency of 433 MHz is used extensively in many European countries.
In practice, clinical diathermy is guided by patient report of pain and warmth sensation. Since local
elevation of tissue temperature is apparently the most significant factor in physiological response to diathermy,
objective measures of subcutaneous tissue temperature in real time would enhance both its efficacy and safety.
While accurate and reliable noninvasive sensing of subcutaneous temperature must await further technological
advance, the combination of multiple invasive sensors and computational estimates can provide some useful
information.
Therapeutic indications are based on local elevations of tissue temperature brought about by volume
heating (1,2). In particular, diathermic heating to 41 C to 45 C produces hyperemia-enhanced blood perfusion
to the body part under treatment. The augmentation in blood flow is accompanied by elevations in capillary
1
pressure, in membrane permeability, and in the rate of metabolism. These increases can facilitate tissue
healing and can also facilitate clearance of metabolites, debris, and toxic substances from diseased tissue under
treatment. Diathermic heating of deep tissues promotes relaxation in muscles, reduces pain, and provides relief
from muscle spasms (2,3). Heating can also produce greater extensibility in fibrous collagen tissues, which is
significant in the management of joint contractures due to tightness of the capsule, fibrosis of muscle, and
scarring.
Hyperthermia Treatment for Cancer

Hyperthermia cancer therapy is a treatment procedure in which tumor temperatures are elevated to the range
of 40 C to 45 C. The rationale for hyperthermia therapy is related to the ability of elevated temperature to
selectively destroy malignant cells. While tumor cells exhibit inherent hyperthermic sensitivity, their response
is characterized by the acidic, hypoxic, and nutritionally deprived environment often found in the interior of
various tumors (4,5,6). Poor blood perfusion in the interior of a tumor also facilitates heat buildup. Moreover,
the cytotoxic effects of some autitumor drugs are enhanced and the cell-killing ability of ionizing radiation is
potentiated by hyperthermia serving as a sensitizing agent. Hyperthermia also increases bloodbrain barrier
permeability (7). The synergism of hyperthermia and ionizing radiation is particularly poignant since it is
accomplished by thermal killing of hypoxic cells and cells in S phase (DNA synthesis), which are resistant to
ionizing radiation.
Clinical and laboratory results from various countries have indicated a promising future for hyperthermia
(8). Its efficacy depends on the induction of sufficient temperature rise throughout the tumor volume. A recent
assessment of superficial breast cancer has indicated that local complete response with hyperthermia and
ionizing radiation is about 60% compared to 40% with irradiation alone (9). Currently, hyperthermia is still an
experimental treatment in the United States for late-stage patients with advance tumors, but it has gained
some acceptance in Europe and Japan (10,11,12).
While beyond the scope of this article, whole-body hyperthermia has been employed in some cases to
enhance the effectiveness of chemotherapy for patients with systemic metastatic cancer. A variety of conductive
and convective heating techniques such as warm air, water, and wax are used (13,14,15). Mild whole-body
hyperthermia at 40 C for as long as 10 hours in rats has shown promising therapeutic potentials on primary
tumor. Temperature up to 41.8 C is found to be safe and is well-tolerated by human patients for up to 60 min
(16,17). While clinical results still remain guarded, they provide a foundation for further exploration.
Monitoring and control of tumor temperature in real time during hyperthermia treatment is essential
for effective therapy. While progress in temperature sensing in vivo has been dramatic, considerable advance
is needed prior to widespread clinically application of hyperthermia for cancer. Among approaches that may
impact this outcome include invasive multipoint sensing (18) and noninvasive magnetic resonance imaging
and diagnostic ultrasound temperature mapping (19,20,21).
A prominent problem in hyperthermia treatment for cancer is the generation of heat and control of temperatures in tumors. Ultrasonic and electromagnetic energies are the commonly used sources for regional and
local hyperthermia. A large number of external and implanted antennas and applicators have been designed
to produce therapeutic heating of localized tumors of different volumes in a variety of anatomical sites. Clearly,
each modality has its own liabilities. Recently, there was a comparison of temperature distributions obtained
in the same tumors, and there was also a comparison of acute and subacute toxicities in patients that were
treated with both external ultrasound and electromagnetic applicators (22). It was concluded that there is no
preferred modality. The type of applicator should be selected on the basis of specific site and type of the tumor.
Ultrasonic Heating. Absorption of ultrasonic pressure wave in biological tissue is determined by ultrasound frequency, velocity, and tissue density. Several frequencies between 0.5 MHz and 3.5 MHz have been
used for hyperthermia cancer treatment. At these frequencies, ultrasound can penetrate deep inside the body
while maintaining the ability to focus energy into the tumor to raise the temperature tumor volume to above the
minimum therapeutic temperature. Its clinical application is constrained by the available ultrasound window
between the transducer and target tumor, as well as by the presence of bone and soft tissue interfaces in the
propagation pathway. Differences in tissue density can give rise to excessive temperature elevation resulting
from accumulation of reflected power at these interfaces. For a given set of anatomic and physiologic parameters, temperature distribution in a tumor is determined by transducer design, scanning pattern, scanning
speed, and output power.
Ultrasonic modalities for noninvasive hyperthermia cancer treatment include spot focus transducers and
phased arrays. By mechanically scanning a focused transducer around a treatment volume, uniform temperature distribution with a sharp falloff outside the treatment volume may be obtained (23,24,25). However, for
large, deep-seated tumors, scanning transducers often produce hot spots proximal to the tumor along the central axis ahead of the focal plane (26). It should be mentioned that frequency sweeping and transducers with a
nonvibrating center can be used to reduce the central hot spot (27). There are two classes of phased arrays that
do not require physical movement of the transducer elements (28,29,30,31,32). The class of phased arrays with
geometric focusing and spot scanning has features and limitations similar to those of mechanically scanned
spot focus transducers (28). In this case, the transducer array is fixed in position and electrical spot scanning is
accomplished through adjustment of array element phases which maximize constructive interference at each
focal plane.
Alternative array element configuration and phase excitation can avoid hot spots that result from constructive interference along the arrays central axis (29,30,31,32). Several phased array configurations have
been proposed. They include the concentric ring, sector vortex, spherical section, and the square arrays. With
proper selection of array element phases, these phased arrays can be operated to directly synthesize, without scanning, ultrasonic power deposition patterns for improved localization of heating within the tumor
volume. Phased arrays offer another advantage over single focused transducer: They enable electronically
programmable treatment planning (33). Pretreatment analysis can provide strategies aimed at satisfying
therapeutic requirements for individual patients and specific tumor sites and spare other sensitive anatomic
structures. It is noted that recently ultrasonic applicators have been tested for interstitial hyperthermia (34).
Electromagnetic Heating. Various frequencies of electromagnetic energy within the range of 0.05 MHz
to 2450 MHz have been used for hyperthermia treatment of cancer. The interaction of electromagnetic fields
and waves in biological tissue is governed by (1) source frequency and intensity, (2) antenna or applicator
design and polarization, (3) tissue structure, and (4) dielectric permittivity (35,36,37). In thermal therapeutic
applications, the final temperature is affected also by tissue blood flow and heat conduction. However, the
time rate of heating and spatial distribution of electromagnetic energy at any given moment in time are direct
functions of specific absorption rate (SAR or power deposition), which are functions of antenna or applicator
design.
At frequencies below a few hundred megahertz [i.e., radio frequencies (RF)], wavelengths in tissue are
100 cm or longer (see Table 1). Power deposition for local tissue heating is characterized by quasistatic displacement or conduction currents, and heating comes about through tissue resistance to current flow. At microwave
frequencies, wavelengths are much shorter, radiated power dominates, and dielectric loss gives rise to heat
production. Power coupling from air into tissue is substantial and can exceed 50%. In addition, the effective
depth of penetration can provide useful insight into the performance of various applicators. For example, because of both the focusing ability and the depth of energy penetration, single-contact applicators operating at
915 MHz or 2450 MHz have been used to heat well-localized superficial tumors extending to a depth up to 3
cm to 6 cm.
RF Heating. For noninvasive subcutaneous tissue heating by RF energy, simple capacitive plates and
inductive coil applicators have been used. Tissues are positioned between the plates and are heated by displacement currents (38,39). A water bolus is often placed between the plate and skin to prevent superficial
burns from large electric field concentration near the edges. A limitation of the capacitive applicator is that the
electric field is predominately normal to the interface between fat and other tissues. Overheating by as much
as 20 times that in muscle can occur in subcutaneous fat greater than 2 cm in thickness. Note that it is possible
to treat tumors of patients with subcutaneous fat as thick as 3 cm by precooling the fat prior to the initiation
of heating (40).
The common inductive applicator consisting of a planar or pancake coil with a small number of turns
when placed parallel to the body surface can avoid the excessive heating problem in fatty tissue. Since the
induced electric fields form eddy currents that flow parallel to the tissue interface, heating is highest in muscle
instead of fat. The heating pattern is toroidal with a null along the axis of the applicator (41). Some recent
designs have SARs that do not include a null in the center and are considerably more uniform than that of the
planar coil (38,42,43). While inductive applicators are used predominately for superficial treatment, some of
the newer applicators can produce effective heating up to a depth of nearly 7 cm.
Several RF applicators have been invented to provide noninvasive heating of deep-seated tumors. These
include the large capacitive applicator mentioned previously (44,45) as well as the ridged waveguide (46), helical
coils (47,48), and multielement arrays (49,50,51,52). The helical coil applicator is simple in construction, and it
provides SAR patterns that vary slowly with radial distance. However, the region to be heated must be located
near the center since the axially directed electric field has a maximum near the center of the coil structure.
Its performance may be improved by judicious selection of the diameter-to-length ratio and incorporation
of external tuning. The multielement array concept has gained considerable utility in the clinic. A primary
advantage of multielement array systems is the ability to steer the heating pattern electronically by varying
the amplitude and phase of each element, thereby allowing phased arrays operating at RF to be used for
selective heating of deep-seated tumors in a variety of anatomic sites.
In particular, the annular phased array system is utilized to heat large anatomical regions such as
the thorax, abdomen, and pelvic area (49,50,53). Regional heating is frequently complicated by systemic
hyperthermia and hemodynamic compensation and by excessive heating or adjacent normal tissue structures, especially the bonetissue interface. However, recent advances using feedback algorithms and adaptive
software modifications to control the amplitude and phase of each element showed that it is possible to maximize the SAR at a target tumor position in a complex anatomy and simultaneously minimize or reduce the
power deposition at locations where undesirable hot spots may occur.
A novel noninvasive or minimally invasive concept using ferro- or paramagnetic compounds for intracellular hyperthermia treatment of both primary and metastatic cancers was first proposed in the late 1950s.
Earlier studies have demonstrated both the preferential accumulation of submicron-sized magnetic particles
(magnetites) in tumors and the feasibility of selective heating using 0.24 MHz to 80 MHz RF magnetic fields.
Current investigations (54,55,56) are directed toward cellular uptake of fluidized magnetic particles, bounding
of magnetite with targeting activity towards cancer cells, and the hyperthermic effects of fine magnetic particles on tumor cells in vitro. It is expected that a magnetite-labelled antibody may soon be available clinically
as a therapeutic agent for hyperthermia treatment of cancer.
A related technique for RF hyperthermia involves implanted ferromagnetic seeds activated by externally
applied 0.05 to 2 MHz magnetic fields. Heating is produced by eddy currents induced on the surface of the
implant and is therefore dependent on the permeability of the thermoseed material (57,58,59,60,61,62). Using
Curie temperatures close to the maximum temperature desired in the tissue, the ferromagnetic seeds can be
designed to provide thermal self-regulation so that a constant tumor temperature can be maintained throughout
the treatment regime. Since volume tissue heating is by passive thermal conduction, these 0.1 mm to 1.0 mm
diameter thermoseed of various length must be implanted closely. Nevertheless, under certain conditions this
invasive seed implant method like the interstitial RF electrodes and microwave antennas to be discussed later
may be preferable for local hyperthermia of deep-seated tumors. It is noteworthy that ferromagnetic seed
hyperthermia in combination with other modalities are used in the control of ocular tumors in animals (63,64).
Recently, multifilament seeds such as the palladiumnickel (PdNi) thermoseeds have gained interest because
of a more effective power deposition than solid seeds (65).
For some deep-seated tumors or tumors of large volume, interstitial techniques have been employed to
generate the desired hyperthermic field. RF electrodes operate in the frequency range of 0.5 MHz to 1 MHz
(66,67,68,69). The advantages of interstitial techniques are safe (without skin burn) and more uniform heat
distribution within the tumor. RF current flowing between pairs of needle-like bare electrodes is dissipated
by the ohmic resistance of tissue and is converted to heat. The temperature distribution produced is strongly
dependent upon blood flow in the tissue and spacing between electrodes. Most clinical applications require an
array of these electrodes spaced at 1.0 cm to 1.5 cm intervals in parallel for optimal temperature uniformity.
Excessive or inadequate heating could be minimized by independent control of RF currents and by varying the
lengths of electrodes.
Microwave Heating. For superficial tumors, single-contact applicators operating at 433 MHz to
2450 MHz have been used. The shorter wavelength at these frequencies allows microwave radiation from
a small applicator some focusing ability in tissues for selective hyperthermia. Because of the limited depth of
energy penetration, these antennas have been applied to heating well-localized tumors extending to depths of
up to 3 cm to 6 cm depending on the particular applicator (39,44,69,70).
The types of external applicators that have been reported for cancer hyperthermia include horns, microstrip applicators, and circular, rectangular, and ridged waveguides (71,72,73,74,75,76,77,78,79). These applicators are used with a high-permittivity, dielectric material to match them to tissue. In the case of a water-like
bolus, it serves also to provide surface cooling of the skin and to avoid the problem of burns and blisters. Microstrip applicators are lightweight and have a low profile. They offer efficient energy coupling and are easier
to use clinically (77,78,79). One limitation of a single applicator is its small area of tissue coverage. Another
is that the SAR distribution cannot be modified during use, making it difficult to improve the nonuniform
temperature distribution that are inevitably produced during patient treatments. One approach to overcome
this problem is to scan the applicator over the tissues (80).
A favorable external system to treat tumors of wide area (tumors that exceed several cm in diameter)
is the phased array consisted of multiple microstrip applicators. The primary advantage of the multielement
array system is the ability to control electronically the SAR distribution by varying the amplitude and relative
phase of each element, independently. Moreover, a planar or quasiplanar phased array operating at microwave
frequencies can be used to improve depth of penetration for selective heating of deep-seated tumors in a variety
of anatomic sites (81,82,83,84,85,86). A further advantage is that the SAR distribution can be adjusted during
treatment, enabling it to enhance the homogeneity of temperature distribution in the target region. The added
sophistication needed for controlling a multitude of array parameters is well within the capability of current
electronic technology. Although a bolus of cooling fluids can be used to prevent undesirable heating of superficial
tissues, tissue layers and curvatures in the near field of the applicator present considerable challenge to quality
control in patient treatments. In practice, the commonly accepted SAR variation is 50% throughout the entire
treatment region.
Intracavitary techniques can be used for certain tumors at hollow viscera and cavity sites such as the
esophagus, cervix, bladder, prostate, and rectum (69,87). Properly designed intracavitary applicators and
antennas can lead to a highly targeted heating of tumors and a reduced risk of unwanted heating of normal
tissues. There are several reports of devices designed for various tumor sites (88,89,90,91). Clinical applications
may require the antennas to be equipped with an integrated cooling system.
The technical difficulty in heating deep-seated tumors without overheating adjacent normal tissue confronted by external applicators has enabled interstitial array techniques to become a viable treatment modality
(67,69,92,93,94,95). The technique has the capacity to adapt its SAR distribution to an irregularly shaped tumor
volume and to provide uniform temperature in deep-seated tumors. Also in combination with brachytherapy,
interstitial hyperthermia renders a treatment modality for malignancies with little additional risk to the
patient (69,95,96).
The efficacy of interstitial microwave heat treatment of soft-tissue tumors is predicated on a sufficient
temperature distribution throughout the tumor. A major determinant is the catheter antenna. Recent designs
have provided microwave interstitial array systems capable of inducing uniform temperature distribution
throughout the entire tumor volume without the need for insertion of the tip of the antenna well beyond the
tumor boundary (97,98,99,100). That requirement was a major drawback of many older catheter antennas
which had the tendency to produce a cold spot or low-heating zone near the distal tip of the antenna (67,101,
102,103), which creates an unnecessary situation for damage to normal tissue. A desirable feature of some of
the newer catheter antenna designs, especially those with integral sleeves or coaxial chokes, is that the SAR
distribution is independent of insertion depth (90,104,105). These antennas have also managed to alleviate the
common problem of excessive heating of the skin from current accumulation at the insertion point. In the clinic,
interstitial microwave antennas are inserted into plastic catheters implanted into the tumor. Computational
and experimental studies have shown that SAR distributions vary with antenna design, catheter size and
material, and air space between the antenna and the catheter (100,105,106).
Array configurations (i.e., geometry and antenna spacing) would also dictate the performance of the
interstitial array treatment modality. Current microwave interstitial array systems rely mostly on equilateral
triangle and square arrays of catheter antennas operating at 433 MHz to 2450 MHz and use element spacings
of 10 mm to 20 mm. Theoretical and experimental results have shown that uniform power deposition and
temperature distribution can be attained from both triangular and square arrays. However, power deposition
and temperature elevation are higher for the triangular configuration at a given level of delivered microwave
power. Moreover, for coherent phase excitations, constructive interference can provide SARs at the array centers
an order of magnitude higher than those corresponding to a single interstitial microwave antenna. A flexibility
afforded by an array of interstitial antennas is that the point of maximum SAR may be shifted from location
to location by changing the amplitude and phase of each antenna. This would avoid low SAR spots during
treatment and would ensure uniform tumor temperature over the entire treatment session. Nevertheless, it
should be noted that ideal operating conditions are difficult to assure in the clinical setting.
Coagulative Ablation Therapy

The development of coagulative ablation therapy over the past decade has revolutionalized the practice of cardiology, gynecology, and urology. For many of the diseases, surgical intervention has been the principal method
of treatment, although alternatives to surgery have been sought in an effort to reduce the cost and morbidity
of treatment (107,108,109,110,111,112,113,114). Minimally invasive catheter ablation offers several potential
benefits: Long incisions are replaced with a puncture wound, major cardiac and pulmonary complications from
general anesthesia are side-stepped, and the need for postoperative intensive care is significantly reduced and,
in many cases, offers a complete or lasting cure. It also has important advantages over drugs that are merely
palliative. It avoids the side effects, expense, and inconvenience of chronic drug therapy, often with only partial
success.
Energy sources that can be used for ablation therapy include RF, microwave, laser, and ultrasound. Before
describing the RF and microwave technology, a brief discussion of laser interactions with tissue is given since
it is used for the thermal therapy.
The effect of lasers on tissue is related to fluence (the product of power density and duration of irradiation). Thermal ablation involves the delivery of laser power for periods lasting tens to hundreds of seconds.
The temperature can easily reach 100 C, depending on the delivery system. The desire to limit excessive tissue
damage (myocardium, uterine wall, etc.) has led to several innovative technologies for laser thermal intervention. However, the complication of perforation or dissection is a cause for special attention in laser sources
(115,116,117,118,119,120,121,122,123). Nevertheless, laser catheter irradiation (e.g., Nd:YAG, 1064 nm, 15 W
to 50 W, spot diameter 2.0 mm to 2.5 mm) has been shown to produce lesions selectively in the targeted segment
of the right ventricular conduction system in dogs, and the method can be performed in a controllable manner
(120). Endometrial ablation techniques using laser coagulation under direct hysteroscopic control have been attempted with varying success (115,116,117,118,119). The variability arises principally from the unpredictable
nature of induced thermal injury and perforation of the uterine wall. Laser coagulation prostatectomy is used
to improve urinary flow rates (121,122,123). While results are comparable to standard electrocautery resection,
the procedure can be enhanced by modifying the laser regimen and the spatial distribution of lesions. Since
medical applications of laser are discussed elsewhere, it will not be addressed further in this section.
RF and Microwave Ablation. In RF thermal ablation therapy, the current flows between a small
electrode inside the body to a large grounded electrode on the surface. The current rapidly diverges from the
small electrode, so that current density is the highest at the electrodetissue interface. The tissues resistance to
current flow results in termal lesions: desiccation and coagulation of tissue in direct contact with the electrode.
The desiccated and coagulated tissue would raise the resistance to current flow, impede effective tissue heating,
and limit the size of RF-induced lesions. Lesion beyond the immediate vicinity of the electrodetissue interface
occurs as a result of passive heat transfer from the thin high-temperature region. Investigations have shown
that RF-induced lesions increase rapidly in size during the initial period of power application; then the rate of
increase diminishes rapidly as the resistance at the electrodetissue interface rises and the current flow falls
(124,125,126,127). For this and reasons described below, studies comparing the power deposition patterns of
RF and microwave catheters have shown that the absorbed microwave energy could be 10 times higher than
RF at the same tissue depth (128,129).
The frequencies of most interest to microwave ablation are 915 MHz and 2450 MHz. Typical values
of microwave dielectric permittivity and conductivity at 37 C are given in Table 2. The biological tissues of
interest to ablation therapy can be classified into three major groups according to their water content. The
group with very high water content includes blood, uterine lining, or physiological fluids. The second group is
of moderately high water content and includes muscle or cardiac wall. The third group is made up of tissues
with low water content such as bone, fat, or desiccated tissue. It can be seen that there is a modest change
in dielectric constant and conductivity as a function of frequency. However, differences among the tissues are
quite large. The higher 2450 MHz frequency is chosen because at this frequency the dielectric constant for
blood is 20% higher than that for muscle, and the dielectric constant muscle is about 800% higher than that
for fat. While conductivities of blood and muscle are approximately the same, they are about 300% higher than
that of fat. As the microwave radiates into the tissue medium, energy is absorbed and converted to heat by
dielectric loss. This absorption will result in a progressive reduction of the microwave power intensity as it
advances in the tissue. The time rate of heating and spatial distribution of radiated microwave energy at any
given moment in time are direct functions of SAR and antenna radiation pattern, respectively.
The reduction is quantified by the depth of penetration; a measure of the distance through which the
intensity of a plane wave field is reduced to 13.5% of its initial level in a medium. At 2450 MHz, the depths
of plane wave penetration for blood, muscle, and fat are 19 mm, 17 mm, and 81 mm, respectively (35,36,37).
For microwave catheter antennas which do not have plane wavefronts, the penetration depth is reduced
according to the specific antenna design. Nevertheless, these values clearly suggest that microwaves can
deposit energy directly into distant tissues. Furthermore, the difference in the dielectric permittivity yields a
depth of penetration for tissues with low water content about four times deeper for muscle or higher water
content tissue at 2450 MHz. This means that a microwave field can propagate more readily through and be
absorbed less by low water content tissues than that of high water content. It also implies that microwaves
can propagate through intervening desiccated tissue or fat to deposit energy directly into more deeply-seated
tissue.
Cardiac Ablation for Tachyarrhythmia. For a significant portion of patients suffering from tachyarrhythmias, available drug therapy has been found unsatisfactory because of a lack of meaningful response
or unacceptable side effects (107,108,109). In some cases, these patients can be meanaged by open-heart
surgery. Percutaneous catheter ablation of arrhythmogenic foci inside the heart is a potentially curative mode
of treatment. Indeed, RF ablation has emerged as an effective therapy for many supraventricular tachycardias
and has become accepted as the standard treatment for arrhythmias associated with the WolfParkinson
White syndrome (107,108,130). Typically, the catheter is inserted percutaneously into the femoral vein and,
under the guidance of a fluoroscope, is then advanced to inside the heart chamber. The cardiac conducting
tissue responsible for the tachycardia is identified with the aid of endocardiac electrograms. A burst of RF
energy is delivered through the electrodes to thermally ablate the cardiac conducting tissue responsible for the
tachycardia and restores the heart to its normal rhythm.
Rapid and reliable mapping of the endocardiac electrogram for identification remains a technical challenge. Also, the lesions induced by RF current is quite small and shallow (125,127,131). Increasing the output
power to heat tissue at a distance often results in excessive temperatures at the electrodetissue interface
without the desired enlargement of lesion size (126,132). Note that temperature-guided RF catheter ablation
with very large distal electrodes can be used to improve lesion size (133). The impedance of the ablating electrode would rise due to poor coupling between the electrodes and adjacent tissue; desiccated and coagulated
tissue raises the resistance to current flow, thwarts effective tissue heating, and limits the size of RF-induced
lesions.
There is a need for energy sources that can produce larger and deeper lesions than RF currents. Large
lesions are required for certain types of cardiac ablation to cure ventricular tachycardias secondary to coronary
artery disease, for example, and arrhythmias due to reentry located deep in the myocardium in particular.
The radiating and dielectric heating features of microwave energy theoretically may be useful for ventricular ablation. The interaction of microwaves as mentioned earlier can result in a greater volume distribution
of energy and deeper penetration. The feasibility of ablating the atrioventricular (AV) junction in dogs with
microwave catheter antennas has been shown both in vitro (134,135,136) and in vivo (137,138,139,140). Furthermore, using fresh bovine hearts and closed-chest dogs, the feasibility of a larger (4 mm long) split-tip
catheter antenna has been demonstrated for ablation treatment of ventricular tachycardia (141,142). The results suggest that if the lesion size is sufficiently large, it would be possible to ablate a ventricular tachycardia
focus using this split-tip microwave catheter antenna system. In addition to the split-tip catheter antenna,
microwave antennas reported for cardiac ablation include monopole, helical coil, cap-slot, and cap-choke designs (134,135,136,137,138,139,140,141,142,143,144,145,146). A drawback of some catheter antennas is that
a considerable amount of microwave energy is reflected by the antennas to the skin surface and is deposited at
the point of antenna insertion into the blood vessel. The problem has been addressed by integrating a sleeve
or choke in the antenna design (134,135,136,139,141,142,143,144).
It is noted that the feasibility of using ultrasound for cardiac ablation was investigated and a catheter
mounted transducer has been reported (147,148).
Endometrial Ablation. Hysterectomy is performed to surgically remove the uterus in order to stop
intractable bleeding or menorrhagia (149,150). Endometrial ablation is a relatively new treatment for menorrhagia and is a reliable alternative treatment for patients with dysfunctional uterine bleeding (151,152,153).
It is superior to hysterectomy in terms of operative complication and postoperative recovery. While still in the
beginning stages, RF and microwave thermal ablation of the endometrium have been reported as efficacious
procedures for treatment of abnormal uterine bleeding (154,155,156,157). The technique is easier and quicker
to perform than current alternatives. Quantitative measures and patients subjective responses suggest that a
meaningful fraction of patients treated with RF and microwave ablation experience significant flow reduction.
Investigations with microwave energy indicate that a treatment temperature of 55 C is related to significant reduction or complete elimination of menstrual flow (156,157). Besides the difference between microwave
and RF approaches mentioned already, the use of high-intensity RF power (500 W) could produce burns at
points where electrocardiographic (ECG) electrodes come in contact with the body (158). Considerably more
investigation is needed before microwave or RF ablation can become a safe and efficacious clinical modality.
Treatment of Benign Prostate Hyperplasia. Benign prostatic hyperplasia or hypertrophy is a major
cause of morbidity in the adult male. At present, open surgery and transurethral resection of the prostate are the
gold standards for treatment of benign prostatic hypertrophy. They can provide immediate relief of obstructive
symptoms that remain fairly to extremely durable (159). A new, less invasive procedure uses thermal energy
delivered by microwaves (160165). An early report of a thermal microwave technique from 1985 employed a
transurethral microwave applicator. It showed coagulation of the prostate in mongrel dogs and some salutary
effects in an initial six patients treated with this device (160). An ensuing study used 2450 MHz microwave
energy to treat 35 patients and compared transurethral resection alone to preliminary microwave coagulation
followed by transurethral resection of the gland (161). Significant reduction in blood loss by initial treatment
with microwave thermal therapy was observed. Numerous reports have appeared since that time on various
10
aspects of both transrectal and transurethral microwave therapy of the prostate using 915 MHz and 2450 MHz
energy (162,163,164,165,166,167).
Most of the research in human subjects to date has focused on methods of delivery. Initial attempts to
deliver the energy transrectally have not been effective, and injury to the rectal mucosa has occurred due to
the difficulty of interface cooling of this organ (166,167). Recent investigations have focused on transurethral
delivery of the energy with cooling systems within the catheter to ensure urethral preservation (143,144,
162,163,164,165,168,169). Sensors placed in the microwave antenna maintain temperature on the urethral
surface between 43 C and 45 C. It is noted that while the number of treatment sessions and the temperature
attained are extremely important predictors of response, sufficient hyperthermia volume is crucial for enhanced
efficacy. Virtually no data clearly demonstrating reduction in prostate volume in human subjects have been
reported, although most investigators have shown improvement in measured urinary flow rates compared
to preoperative studies. Randomized studies comparing microwave thermotherapy to transurethral resection
conclude that microwave hyperthermia treatment had a definite therapeutic effect on symptomatic prostatic
hypertrophy (169,170,171,172). Thus, microwave thermal ablation of prostatic tissue and enlargement of the
urethra with minimal clinical complications offers a therapeutic alternative to surgery in select patients with
benign prostatic hyperplasia.
BIBLIOGRAPHY
1. J. F. Lehmann (ed.), Therapeutic Heat and Cold, Baltimore, MD: Williams & Wilkins, 1990.
2. J. F. Lehmann, Diathermy, in F. H. Krusen, F. J. Kottke, and P. M. Elwood (eds.), Handbook of the Physical Medicine
and Rehabilitation, Philadelphia: Saunders, 1971, Chap. 11, pp. 273345.
3. E. Fisher, S. Solomon, Physiological responses to heat and cold, in S. Licht (ed.), Therapeutic Heat and Cold, New
Haven, CT: Licht, 1965, pp. 126169.
4. G. ter Haar, Effects of increased temperature on cells, on membranes and on tissues, in D. J. Watmough and W. M.
Ross (eds.), Hyperthermia, Glasgow and London: Blackie, 1986, pp. 1441.
5. R. J. Griffin et al., Mild temperature hyperthermia combined with carbogen breathing increases tumor partial
pressure of oxygen (pO2 ) and radiosensitivity, Cancer Res., 56: 55905593, 1996.
6. D. M. Brizel et al., Radiation therapy and hyperthermia improve the oxygenation of human soft tissue sarcomas,
Cancer Res., 56: 53475350, 1996.
7. J. C. Lin, M. F. Lin, Microwave hyperthermia-induced bloodbrain barrier alterations, Radiat. Res., 89: 7787, 1982.
8. E. W. Gerner, T. C. Cetas (eds.), Hyperthermia Oncology 1992, Tucson: Arizona Board of Regents, 1993.
9. C. C. Vernon et al., Radiotherapy with or without hyperthermia in the treatment of superficial localized breast
cancerresults from five randomized controlled trials, Int. J. Radiat. Oncol. Biol. Phys., 35: 731744, 1996.
10. J. Overgaard et al., Randomised trial of hyperthermia as adjuvant to radiotherapy for recurrent or metastatic
malignant melanoma, Lancet, 345: 540543, 1995; Hyperthermia as an adjuvant to radiation therapy of recurrent
or metastatic malignant melanomaa multicentre randomized trial by the European Society for Hyperthermic
Oncology, Int. J. Hyperthermia, 12: 320, 1996.
11. B. Emami et al., Phase III study of interstitial thermoradiotherapy compared with interstitial radiotherapy alone
in the treatment of recurrent or persistent human tumorsa prospectively controlled randomized study by the
Radiation Therapy Oncology Group, Int. J. Radiat. Oncol. Biol. Phys., 34: 10971104, 1996.
12. H. Kuwano et al., Preoperative hyperthermia combined with chemotherapy and irradiation for the treatment of
patients with esophageal carcinoma, Tumori, 81: 1822, 1995.
13. J. M. C. Bull, A review of systemic hyperthermia, Hyperthermia Radiat. Ther./Chemother. Treat. Cancer, 18: 171176,
1984.
14. J. van der Zee et al., Whole body hyperthermia as a treatment modality, in S. B. Field and C. Franconi (eds.), Physics
and Technology of Hyperthermia, Dordrecht, The Netherlands: Martinus Nijhoff, 1987, pp. 420440.
15. S. B. Field, J. W. Hand (eds.), An Introduction to the Practical Aspects of Clinical Hyperthermia, London: Taylor &
Francis, 1990.
11
16. H. Matsuda et al., Long duration mild whole body hyperthermia of up to 12 hours in rats: Feasibility and efficacy
on primary tumour and axillary lymph node metastases of a mammary adenocarcinomaimplications for adjuvant
therapy, Int. J. Hyperthermia, 13: 8998, 1997.
17. H. I. Robins et al., Phase I clinical trial of melphalan and 41.8-degrees-C whole-body hyperthermia in cancer patients,
J. Clin. Oncol., 15: 158164, 1997.
18. P. Vanbaren, E. S. Ebbini, Multipoint temperature control during hyperthermia treatmentstheory and simulation,
19. C. J. Lewa, J. D. Decertaines, Body temperature mapping by magnetic resonance imaging. Spectrosc. Lett., 27:
13691419, 1994.
20. J. R. Macfall et al., H-1 MRI phase thermometry in vivo in canine brain, muscle, and tumor tissue, Med. Phys., 23:
17751782, 1996.
21. R. Seip et al., Noninvasive real-time multipoint temperature control for ultrasound phased array, IEEE Trans.
Ultrason. Ferroelectr. Freq. Control, 43: 10631073, 1996.
22. R. Benyosef, D. S. Kapp, Direct clinical comparison of ultrasound and radiative electromagnetic hyperthermia
applicators in the same tumours, Int. J. Hyperthermia, 11: 110, 1995.
23. P. Lele, Physical aspects and clinical study with ultrasound hyperthermia, in F. K. Storm (ed.), Hyperthermia in
Cancer Therapy, Boston: Hall Medical, 1983, pp. 333367.
24. K. Hynynen et al., A scanned focused multiple transducer ultrasound system for localized hyperthermia treatment,
Int. J. Hyperthermia, 3: 2125, 1987.
25. E. Moros, R. Roemer, K. Hynynen, Pre-focal plane high temperature regions induced by scanning focused ultrasound
beams, Int. J. Hyperthermia, 6: 351366, 1990.
26. W. L. Straube et al., An ultrasound system for simultaneous ultrasound hyperthermia and photon beam irradiation,
Int. J. Radiat. Oncol. Biol. Phys., 36: 11891200, 1996.
27. M. Mitsumori et al., A phase I and III clinical trial of a newly developed ultrasound hyperthermia system with an
improved planar transducer, Int. J. Radiat. Oncol. Biol. Phys., 36: 11691175, 1996.
28. K. B. Ochetree et al., An ultrasonic phased array applicator for hyperthermia, IEEE Trans. Sonics Ultrason., SU-31:
526531, 1984.
29. C. A. Cain, S. Umemura, Concentric ring and sector vortex phased arrays for tumor treatment, IEEE Trans. Microw.
Theory Tech., MTT-34: 542551, 1986.
30. E. S. Ebbini, C. A. Cain, A spherical sector ultrasound phased array applicator for deep localized hyperthermia,
IEEE Trans. Biomed. Eng., BME-38: 634643, 1991.
31. S. Umemura, C. A. Cain, Acoustical evaluation of a prototype sector-vortex phased-array applicator, IEEE Trans.
Ultrason. Ferroelectr. Freq. Control, 39: 3238, 1992.
32. R. J. McGough et al., Mode scanning: Heating pattern synthesis with ultrasound phased arrays, Int. J. Hyperthermia,
10: 433442, 1994.
33. R. J. McGough et al., Treatment planning for hyperthermia with ultrasound phased arrays, IEEE Trans. Ultrason.
Ferroelectr. Freq. Control, 43: 10741084, 1996.
34. C. J. Diederich, Ultrasound applicators with integrated catheter-cooling for interstitial hyperthermiatheory and
preliminary experiments, Int. J. Hyperthermia, 12: 279297, 1996.
35. J. C. Lin, Engineering and biophysical aspects of microwave and radio-frequency radiation, in D. J. Watmough and
W. M. Ross (eds.), Hyperthermia, Glasgow: Blackie, 1986, pp. 4275.
36. S. M. Michaelson, J. C. Lin, Biological Effects and Health Implications of Radiofrequency Radiation, New York:
Plenum, 1987.
37. J. C. Lin, O. P. Gandhi, Computer methods for predicting field intensity, in C. Polk and E. Postow (eds.), Handbook
of Biological Effects of Electromagnetic Fields, Boca Raton, FL: CRC Press, 1996, pp. 337402.
38. J. Hand, Technical and clinical advances in hyperthermia treatment of cancer, in J. C. Lin (ed.), Electromagnetic
Interaction with Biological Systems, New York: Plenum, 1989, pp. 5980.
39. M. Gautherie (ed.), Methods of External Hyperthermia Heating, Berlin: Springer-Verlag, 1990.
40. C. K. Lee et al., Clinical experience using 8 MHz radiofrequency capacitive hyperthermia in combination with
radiotherapyresults of a phase I/II study, Int. J. Radiat. Oncol. Biol. Phys., 32: 733745, 1995.
41. A. W. Guy, J. F. Lehmann, J. B. Stonebridge, Therapeutic applications of electromagnetic power, Proc. IEEE, 62:
5575, 1974.
12
42. C. Franconi et al., Low-frequency RF dipole applicator for intermediate depth hyperthermia, IEEE Trans. Microw.
Theory Tech., MTT-34: 612619, 1986.
43. Y. Fujita, H. Kato, T. Ishida, An RF concentrating method using inductive aperture-type applicators, IEEE Trans.
Biomed. Eng., 40: 110113, 1993.
44. S. B. Field, C. Franconi (eds.), Physics and Technology of Hyperthermia, Dordrecht, The Netherlands: Martinus
Nijhoff, 1987.
45. C. W. Song et al., Capacitive heating of phantom and human tumors with an 8 MHz radiofrequency applicator, Int.
J. Radiat. Oncol. Biol. Phys., 12: 365372, 1986.
46. R. Paglione et al., 27 MHz ridged waveguide applicators for localized hyperthermia treatment of deep-seated malignant tumors, Microw. J., 24: 7180, 1981.
47. P. S. Ruggera, G. Kantor, Development of a family of helical coil applicators which produce transversely uniform
axially distributed heating in cylindrical fatmuscle phantoms, IEEE Trans. Biomed. Eng., BME-31: 98106, 1984.
48. M. J. Hagmann, R. L. Levin, Analysis of the helix as an RF applicator for hyperthermia, Electron. Lett., 20: 337338,
1984.
49. P. Turner, Regional hyperthermia with an annular phased array, IEEE Trans. Biomed. Eng., BME-31: 106114,
1984.
50. P. Turner, Mini-annular phased array for limb hyperthermia, IEEE Trans. Microw. Theory Tech., MTT-34: 508513,
1986.
51. P. Wust et al., Simulation studies promote technological development of radio frequency phased array hyperthermia,
52. A. J. Fenn, G. A. King, Experimental investigation of an adaptive feedback algorithm for hot spot reduction in
radio-frequency phased-array hyperthermia. IEEE Trans. Biomed. Eng., 43: 273280, 1996.
53. K. Urata et al., Radiofrequency hyperthermia for malignant liver tumorsthe clinical results of seven patients,
Hepatogastroenterology, 42: 492496, 1995.
54. A. Jordan et al., Inductive heating of ferromagnetic particles and magnetic fluid: Physical evaluation of their potential
for hyperthermia, Int. J. Hyperthermia, 9: 5168, 1993.
55. M. Suzuki et al., Preparation and characteristics of magnetite-labelled antibody with the use of poly(ethylene glycol)
derivatives. Biotechnol. Appl. Biochem., 21: 335345, 1995.
56. A. Jordan et al., Cellular uptake of magnetic fluid particles and their effects on human adenocarcinoma cells exposed
to AC magnetic fields in vitro, Int. J. Hyperthermia, 12: 705722, 1996.
57. J. C. Lin, Induction thermocoagulation of the brainquantitation of absorbed power, IEEE Trans. Biomed. Eng.,
BME-22: 542546, 1975.
58. W. J. Atkinson, I. A. Brezovich, D. P. Chakraborty, Usable frequencies in hyperthermia with thermal seeds, IEEE
Trans. Biomed. Eng., BME-31: 7075, 1984.
59. P. R. Stauffer, T. C. Cetas, R. C. Jones, Magnetic induction heating of ferromagnetic implants for inducing localized
heating in deep-seated tumors, IEEE Trans. Biomed. Eng., BME-31: 235251, 1984.
60. R. F. Meredith et al., Ferromagnetic thermoseeds suitable for an afterloading interstitial implant, Int. J. Radiat.
Oncol. Biol. Phys., 17: 13411346, 1989.
61. S. K. Jones et al., Evaluation of ferromagnetic materials for low frequency hysteresis heating of tumors, Phys. Med.
Biol., 37: 293299, 1992.
62. J. A. Paulus et al., Evaluation of inductively heated ferromagnetic alloy implants for therapeutic interstitial hyperthermia, IEEE Trans. Biomed. Eng., 43: 406413, 1996.
63. R. A. Steeves et al., Thermoradiotherapy of intraocular tumors in an animal modelconcurrent vs. sequential
brachytherapy and ferromagnetic hyperthermia, Int. J. Radiat. Oncol. Biol. Phys., 33: 659662, 1995.
64. T. G. Murray et al., Radiation therapy and ferromagnetic hyperthermia in the treatment of murine transgenic
retinoblastoma, Arch. Ophthalmol., 114: 13761381, 1996.
65. N. Vanwieringen et al., Power absorption and temperature control of multi-filament palladiumnickel thermoseeds
for interstitial hyperthermia, Phys. Med. Biol., 41: 23672380, 1996.
66. J. M. Cosset, Interstitial techniques, in J. Overgaard (ed.), Hyperthermia Oncology 1984, London: Taylor & Francis,
1985, pp. 309316.
67. J. M. Strohbehn, T. A. Mechling, Interstitial techniques for clinical hyperthermia, in J. W. Hand and J. R. James
(eds.), Handbook of Techniques for Clinical Hyperthermia, London: Research Studies Press, 1986, pp. 210219.
13
68. B. Gao, S. Langer, P. M. Corry, Application of the time-dependent greens function and Fourier transforms to the
solution of the bioheat equation, Int. J. Hyperthermia, 11: 267285, 1995.
69. M. H. Seegenschmiedt, P. Fessenden, C. C. Vernon (eds.), Medical Radiology Thermoradiotherapy and Thermochemotherapy, Berlin: Springer-Verlag, 1996, Vol. 2.
70. J. C. Lin (ed.), Special issue on phased arrays for hyperthermia treatment of cancer, IEEE Trans. Microw. Theory
Tech., MTT-34: 481648, 1986.
71. A. W. Guy et al., Development of a 915-MHz direct contact applicator for therapeutic heating of tissues, IEEE Trans.
Microw. Theory Tech., MTT-26: 550556, 1978.
72. J. C. Lin, G. Kantor, A. Grods, A class of new microwave therapeutic applicators, Radio Sci., 17: 119s123s, 1982.
73. Y. Nikawa et al., A direct-contact microwave lens applicator with a microcomputer-controlled heating system for
local hyperthermia, IEEE Trans. Microw. Theory Tech., MTT-34: 481648, 1986.
74. M. Hiraoka et al., Clinical evaluation of 430 MHz microwave hyperthermia system with lens applicator for cancer
therapy, Med. Biol. Eng. Comput., 33: 4447, 1995.
75. M. D. Sherar et al., Beam shaping for microwave waveguide hyperthermia applicators, Int. J. Radiat. Oncol. Biol.
Phys., 25: 849857, 1993.
76. E. G. Moros et al., Clinical system for simultaneous external superficial microwave hyperthermia and Cobalt-60
radiation, Int. J. Hyperthermia, 11: 1126, 1995.
77. T. V. Samulski et al., Spiral microstrip hyperthermia applicator: technical design and clinical performance, Int. J.
Radiat. Oncol. Biol. Phys., 18: 233242, 1990.
78. Y. Nikawa, M. Yamamoto, A multielement flexible microstrip patch applicator for microwave hyperthermia, IEICE
Trans. Commun., 78B: 145151, 1995.
79. C. Michel et al., Design and modeling of microstripmicroslot applicators with several patches and apertures for
microwave hyperthermia, Microw. Opt. Tech. Lett., 14: 121126, 1997.
80. F. Sterzer et al., A robot-operated microwave hyperthermia system for treating large malignant surface lesions,
Microw. J., 29: 147152, 1986.
81. J. W. Hand, J. L. Cheeham, A. J. Hind, Absorbed power distributions from coherent microwave arrays for localized
hyperthermia, IEEE Trans. Microw. Theory Tech., MTT-34: 484489, 1986.
82. J. T. Loane et al., Experimental investigation of a retrofocusing microwave hyperthermia applicator: conjugate-field
matching scheme, IEEE Trans. Microw. Theory Tech., MTT-34: 490494, 1986.
83. J. T. Loane, S. W. Lee, Gain optimization of a near field focusing array for hyperthermia applications, IEEE Trans.
84. E. J. Gross et al., Experimental assessment of phased array heating of neck tumors, Int. J. Hyperthermia, 6: 453474,
1990.
85. T. P. Ryan, V. I. Backus, C. T. Coughlin, Large stationary microstrip arrays for superficial microwave hyperthermia,
86. R. M. Najafabadi, A. F. Peterson, Focusing and impedance properties of conformable phased array antennas for
microwave hyperthermia, IEEE Trans. Microw. Theory Tech., 44: 17991802, 1996.
87. A. Yerushalmi, Fifteen years of experience with intracavitary hyperthermia in cancer therapy, Exp. Oncol., 17:
325332, 1995.
88. R. L. Liu et al., Heating patterns of helical microwave intracavitary oesophageal applicator, Int. J. Hyperthermia, 7:
577586, 1991.
89. D. J. Li et al., Design of intracavitary microwave applicators for the treatment of uterine cervix carcinoma, Int. J.
Hyperthermia, 7: 693701, 1991.
90. J. C. Lin, Y. J. Wang, The cap-choke catheter antenna for microwave ablation treatment, IEEE Trans. Biomed. Eng.,
43: 657660, 1996.
91. D. Roos et al., A new microwave applicator with integrated cooling system for intracavitary hyperthermia of vaginal
carcinoma, Int. J. Hyperthermia, 12: 743756, 1996.
92. J. W. Strohbehn, E. B. Douple, Hyperthermia and cancer therapy: a review of biomedical engineering contributions
and challenges, IEEE Trans. Biomed. Eng., BME-32: 779787, 1984.
93. Y. Zhang, W. T. Joines, J. R. Oleson, Microwave hyperthermia induced by a phased interstitial antenna array, IEEE
Trans. Microw. Theory Tech., MTT-38: 217221, 1990.
14
94. J. W. Hand, R. Cardossi, Therapeutic applications of electromagnetic fields, in W. R. Stone (ed.), Review of Radio
Science 19901992, London: Oxford Univ. Press, 1993, pp. 779796.
95. D. J. Lee, R. Mayer, L. Hallinan, Outpatient interstitial thermoradiotherapy, Cancer, 77: 23632370, 1996.
96. T. Nakajima et al., Pattern of response to interstitial hyperthermia and brachytherapy for malignant intracranial
tumour: A CT analysis, Int. J. Hyperthermia, 9: 491502, 1993.
97. Y. Wang, J. C. Lin, A comparison of microwave interstitial antennas for hyperthermia, Proc. IEEE Eng. Med. Biol.
Conf., 1986, pp. 14631466.
98. J. C. Lin, Y. J. Wang, Interstitial microwave antennas for thermal therapy, Int. J. Hyperthermia, 3: 3747, 1987.
99. V. Sathiaseelan et al., Performance characteristics of improved microwave interstitial antennas for local hyperthermia, Int. J. Radiat. Oncol. Biol. Phys., 20: 531539, 1991.
100. G. Schaller, J. Erb, R. Engelbrecht, Field simulation of dipole antennas for interstitial microwave hyperthermia,
IEEE Trans. Microw. Theory Tech., 44: 887895, 1996.
101. J. W. Strohbehn et al., Evaluation of an invasive microwave antenna system for heating deep-seated tumor, J. Natl.
Cancer Inst. Monogr., 61: 489491, 1982.
102. T. Z. Wong et al., SAR patterns from an interstitial microwave antenna-array hyperthermia system, IEEE Trans.
103. P. F. Turner, Interstitial equal-phased arrays for EM hyperthermia, IEEE Trans. Microw. Theory Tech., MTT-34:
572578, 1986.
104. T. P. Ryan, J. A. Mechling, J. W. Strohbehn, Absorbed power deposition for various insertion depth for 915 MHz
interstitial dipole antenna arrays: Experiment versus theory, Int. J. Radiat. Oncol. Biol. Phys. 19: 377387, 1990.
105. J. C. Lin, Y. J. Wang, Power deposition patterns for minature 2450 MHz interstitial antenna and arrays, in T.
Sugahara and M. Saito (eds.), Hyperthermia Oncology 1988, London: Taylor & Francis, 1988, Vol. 1, pp. 891893.
106. M. S. Wu et al., Effect of a catheter on SAR distribution around interstitial antenna for microwave hyperthermia,
IEICE Trans. Commun., 78B: 845850, 1995.
107. S. K. S. Huang (ed.), Radiofrequency Catheter Ablation of Cardiac Arrhythmias: Basic Concepts and Clinical Applications, Armonk, NY: Futura Publ. Co., 1995.
108. A. B. Wagshal, S. K. S. Huang, Application of radiofrequency energy as an energy source for ablation of cardiac
arrhythmias, in J. C. Lin (ed.), Advances in Electromagnetic Fields in Living Systems, New York: Plenum, 1997, Vol.
2, pp. 205254.
109. G. Breithardt, M. Borggrefe, D. P. Zipes, Nonpharmacological Therapy of Tachyarrhythmias, Mount Kisco, NY:
Futura Publ. Co., 1988.
110. A. W. Shaw, A risk benefit analysis of drugs used in the treatment of endometriosis, Drug Saf., 11: 104113, 1994.
111. A. Lalonde, Evaluation of surgical options in menorrhagia, Br. J. Obstet. Gynecol., 101: 814, 1994.
112. H. Lepor et al., A randomized, placebo-controlled multicenter study of the efficacy and safety of terazosin in the
treatment of benign prostatic hyperplasia, J. Urol., 148: 14671474, 1992.
113. G. J. Gormley et al., The effect of finasteride in man with benign prostatic hyperplasia, New Engl. J. Med., 327:
11851191, 1992.
114. M. J. Barry, Medical outcome research and benign prostatic hyperplasia. Prostate (Suppl.), 3: 6174, 1990.
115. J. Davis, The principles and the use of the Nd-YAG laser in gynecological surgery, Clin. Obstet. Gynecol., 1: 331350,
1987.
116. D. C. Marlin, R. Vander Zwaag, Excisional techniques for endometriosis with the CO2 laser laparoscope, J. Reprod.
Med., 32: 753758, 1987.
117. P. C. Reid et al., Nd-YAG laser endometrial ablationhistological aspects of uterine healing, Int. J. Gynecol. Pathol.,
11: 174179, 1992.
118. D. R. Phillips, A comparison of endometrial ablation using the Nd-YAG laser or electrosurgical techniques, J. Amer.
Assoc. Gynecol. Laparosc., 1: 235239, 1994.
119. M. Kelly, H. M. L. Mathews, P. Weir, Carbon dioxide embolism during laser endometrial ablation, Anaesthesia, 52:
6567, 1997.
120. H. P. Weber et al., Mapping guided laser catheter ablation of the atrioventricular conduction in dogs, Pace-Pacing
Clin. Electrophysiol., 19: 176187, 1996.
121. J. N. Kabalin et al., Comparative study of laser versus electrocautery prostatic resection18-month follow-up with
complex urodynamic assessment, J. Urol., 153: 9497, 1995.
15
122. E. Rihuela et al., Histopathological evaluation of laser thermocoagulation in the human prostateoptimization of
laser irradiation for benign prostatic hyperplasia, J. Urol., 153: 15311535, 1995.
123. P. Narayan et al., A randomized study comparing visual laser ablation and transurethral evaporation of prostate in
the management of benign prostatic hyperplasia, J. Urol., 154: 20832088, 1995.
124. R. H. Hoyt et al., Factors influencing transcatheter radiofrequency ablation of the myocardium, J. Appl. Cardiol., 1:
469486, 1986.
125. L. T. Blouin, F. I. Marcus, The effect of electrode design on the efficiency of delivery of RF energy to cardiac tissue in
vitro, PACE, 12: 136143, 1989.
126. F. H. M. Wittkampf, R. N. W. Hauer, E. O. Roblesde Medina, Control of RF lesions size by power regulation,
Circulation, 80: 962968, 1989.
127. J. J. Langberg et al., Radiofrequency catheter ablation: The effect of electrode size on lesion volume in vivo, PACE,
13: 12421248, 1990.
128. J. C. Lin, Y. J. Wang, R. J. Hariman, Comparison of power deposition patterns produced by microwave and radio
frequency cardiac ablation catheters, Electron. Lett., 30: 922923, 1994.
129. T. L. Wonnell, P. R. Stauffer, J. J. Langberg, Evaluation of microwave and radio frequency catheter ablation in a
myocardium-equivalent phantom model, IEEE Trans. Biomed. Eng., 39: 10861095, 1992.
130. M. M. Scheinman, Pattern of catheter ablation practice in the United States: results of the 1992 NASPE survey.
PACE, 17: 873875, 1994.
131. S. Nathan, J. P. Dimarco, D. E. Haines, Basic aspects of radiofrequency ablation, J. Cardiovasc. Electrophysiol., 5:
863876, 1994.
132. I. D. McRury, D. E. Haines, Ablation for the treatment of arrhythmias, Proc. IEEE, 84: 404416, 1996.
133. J. J. Langberg et al., Temperature-guided radiofrequency catheter ablation with very large distal electrode, Circulation, 88: 245249, 1993.
134. J. C. Lin, Transcatheter microwave technology for treatment of cardiovascular diseases, in M. E. OConnor, R. H.
C. Bentall, and J. C. Monahan (eds.), Emerging Electromagnetic Medicine, New York: Springer-Verlag, 1990, pp.
125134.
135. K. J. Beckman et al., Production of reversible and irreversible atrioventricular block by microwave energy [abstract],
Circulation, 16: 1612, 1987.
136. J. C. Lin et al., Microwave ablation of the atrioventricular junction in open heart dogs, Bioelectromagnetics, 16:
97105, 1995.
137. J. J. Langberg et al., Catheter ablation of the atrioventricular junction using a helical microwave antenna: A novel
means of coupling energy to the endocardium, PACE, 14: 21052113, 1991.
138. J. C. Lin et al., Microwave catheter ablation of the canine atrio-ventricular junction [abstract], J. Amer. Coll. Cardiol.,
21: 357a, 1993.
139. J. C. Lin et al., Microwave catheter ablation of the atrioventricular junction in closed-chest dogs, Med. Biol. Eng.
Comput., 34: 295298, 1996.
140. L. B. Liem et al., In vitro and in vivo results of transcatheter microwave ablation using forward-firing tip antenna
design, Pacing Clin. Electrophysiol., 19: 20042008, 1996.
141. S. K. S. Huang et al., Percutaneous microwave ablation of the ventricular myocardium using a 4-mm split tip antenna
electrode: A novel method for potential ablation of ventricular tachycardia [abstract], J. Amer. Coll. Cardiol., 25: 285a,
1994.
142. F. S. Mazzola et al., Determinants of lesions size using a 4-mm split-tip antenna electrode for microwave catheter
ablation, in North American Society for Pacing Electrophysiology (NASPE), Memphis, TN, Newton Upper Fall, MA:
NASPE, 1994, p. 814.
143. J. C. Lin, Y. J. Wang, A catheter antenna for percutaneous microwave therapy, Microw. Opt. Technol. Lett., 8: 7072,
1995.
144. J. C. Lin, Y. J. Wang, The cap-choke catheter antenna for microwave ablation treatment, IEEE Trans. Biomed. Eng.,
43: 657660, 1996.
145. S. Shetty et al., Microwave applicator design for cardiac tissue ablations, J. Microw. Power Electromagn. Energy, 31:
5966, 1996.
146. S. Labonte et al., Monopole antennas for microwave catheter ablation, IEEE Trans. Microw. Theory Tech., 44: 1832
1840, 1996.
16
147. J. E. Zimmer et al., The feasibility of using ultrasound for cardiac ablation, IEEE Trans. Biomed. Eng., 42: P891897,
1995.
148. K. Hynynen et al., Cylindrical ultrasonic transducers for cardiac catheter ablation, IEEE Trans. Biomed. Eng., 44:
144151, 1997.
149. A. H. DeCherny, M. L. Polan, Hysteroscopic management of intrauterine lesions and intractable uterine bleeding,
Obstet. Gynecol., 61: 392397, 1983.
150. A. Lalonde, Evaluation of surgical options in menorrhagia, Br. J. Obstet. Gynecol., 101: 814, 1994.
151. I. S. Fraser et al., Short and medium term outcomes after rollerball endometrial ablation for menorrhagia, Med. J.
Aust., 158: 454457, 1993.
152. M. S. Baggish, E. H. M. Sze, Endometrial ablationa series of 568 patients treated over an 11-year period, Amer. J.
Obstet. Gynecol., 174: 908913, 1996.
153. R. Garry et al., Six hundred endometrial laser ablations, Obstet. Gynecol., 85: 2429, 1995.
154. J. H. Phipps et al., Treatment of functional menorrhagia by radiofrequency-induced thermal endometrial ablation,
Lancet, 335: 374376, 1990.
155. J. H. Phipps et al., Validation of a method of treating menorrhagia by endometrial ablation, Clin. Phys. Physiol.
Meas., 13: 273280, 1992.
156. J. C. Lin, Microwave technology for minimally invasive interventional procedures, Chin. J. Med. Biol. Eng., 13:
293304, 1993.
157. N. C. Sharp et al., Microwave for menorrhagiaa new fast technique for endometrial ablation, Lancet, 346: 1003
1004, 1995.
158. V. J. Page, Anaesthesia and radiofrequency endometrial ablation, Eur. J. Anaesthesiol., 10: 2526, 1993.
159. J. Aagaard et al., Total transurethral resection vs minimal transurethral resection of the prostatea 10-year followup
study of urinary symptoms, uroflowmetry and residual volume, Br. J. Urol., 74: 333336, 1994.
160. T. Harada et al., Microwave surgical treatment of diseases of the prostate, Urology, 26: 572576, 1985.
161. T. Harada et al., Microwave surgical treatment of diseases of the prostate: clinical application of microwave surgery
as a tool for improved prostatic electroresection, Urol. Int., 42: 127131, 1987.
162. M. A. Astrahan et al., Microwave applicator of transurethral hyperthermia of benign prostatic hyperplasia, Int. J.
Hyperthermia, 5: 383396, 1989.
163. W. L. Strohmaier et al., Local microwave hyperthermia of benign prostatic hyperplasia, J. Urol., 144: 913917, 1990.
164. A. Lindner et al., Local hyperthermia of the prostatic gland for the treatment of benign prostate hypertrophy and
urinary retention, Br. J. Urol., 65: 201203, 1990.
165. S. St C. Carter et al., Single session transurethral microwave thermotherapy for the treatment of benign prostate
obstruction, J. Endourol., 5: 137143, 1991.
166. F. Montorsi et al., Transrectal microwave hyperthermia for benign prostatic hyperplasialong term clinical, pathological and ultrastructural patterns, J. Urol., 148: 321325, 1992.
167. P. Debicki et al., Temperature steering in prostate by simultaneous transurethral and transrectal hyperthermia,
Urology, 40: 300307, 1992.
168. L. Baert et al., Transurethral microwave hyperthermia: An alternative treatment for prostdynia, Prostate, 19: 113
119, 1991.
169. D. G. Bostwick, T. R. Larson, Transurethral microwave thermal therapypathologic findings in the canine prostate,
Prostate, 26: 116122, 1995.
170. M. Zerbib et al., Localized hyperthermia vs. the sham procedure in obstructive benign hyperplasia of the prostatea
prospective randomized study, J. Urol., 147: 10481052, 1992.
171. C. Dahlstrand et al., Transurethral microwave thermotherapy vs. transurethral resection for benign prostatic hyperplasia: Preliminary results of a randomized study, Eur. Urol., 23: 292298, 1993.
172. H. Matzkin, Hyperthermia as a treatment modality in benign prostatic hyperplasia, Urology, 43: 1720, 1994.
JAMES C. LIN
University of Illinois at Chicago
451
homes, clinics, medical and dental offices, and ambulatory

care centers.
Clinical engineers are trained in engineering principles,
basic sciences, the life sciences and patient-care principles.
They are knowledgeable of regulatory agency requirements,
health-care codes, and medical equipment standards. This diverse body of knowledge and experience enables clinical engineers to understand how technology and patient care interact
within the clinical setting (3). This knowledge is vital when
medical equipment is integrated into the environment of patient care. It allows clinical engineers to interface medical
systems to the patient, and to other medical systems including those used for data collection (4).
Employment
OVERVIEW
Definition
Clinical engineering is a relatively new profession. The term
clinical engineering was coined by Cesar A. Caceres, M.D. in
1967 to describe a George Washington University Medical
School program in which he envisioned engineers and physicians working together to provide better patient care. This
program, which was not medicine, engineering, nor statistics,
contained elements of all of these disciplines. As the program
focus was to be patient oriented, he chose to couple the term
clinical with engineering to describe it, so as to distinguish it
from the research-oriented activities of biomedical engineering (1).
In 1992 the American College of Clinical Engineers
(ACCE) defined a clinical engineer as a professional who supports and advances patient care by applying engineering and
management skills to healthcare technology (2). The Clinical
Engineering Board of Examiners of the International Certification Commission for Clinical Engineering and Biomedical
Technology endorses this definition.
Environment of Patient Care
At the heart of clinical engineering is the concept of providing
engineering expertise to ensure that the environment of patient care (EC) is safe for both patient and clinician. Medical
equipment used for patient care comprises a large part of this
environment. It must be safe, efficacious (performing the
function for which it was intended), and cost effective. Clinical engineers are the professionals who provide technical support services to ensure this. Their knowledge is invaluable to
health-care provider institutions, such as hospitals, nursing
Primarily employed by health-care provider institutions (as

part of in-house or shared-service clinical engineering departments), clinical engineers are also employed by original
equipment manufacturers (OEM), third-party independent
service organizations (ISO), independent testing laboratories,
clinical engineering consulting firms, regulatory agencies,
technical publishing houses, law firms, and academic institutions.
Clinical engineers hold positions in management, engineering, medical equipment sales, equipment test and evaluation, field service, health-care regulation, technical publishing, and education. They serve as expert witness of patient
incidents and provide input to legislative bodies. They also
serve on curriculum advisory committees on which clinical engineers from both industry- and hospital-based clinical engineering programs sit, allowing curriculum to keep pace with
the most recent industry trends.
Biomedical Equipment Technicians
Closely associated with clinical engineers are biomedical
equipment technicians (BMET). BMETs are skilled technicians who are specially trained to work with medical instrumentation. Although BMETs focus their activities on the repair and maintenance of medical equipment, they are called
on to inspect, install, and modify medical devices, as well as
to provide guidance in proper equipment usage and safety.
Some take on managerial responsibility, supervising other
BMETs.
Education
Clinical Engineering. As a minimum, new practitioners in
clinical engineering require a bachelor of science degree in
engineering. This degree should be obtained from an institution that is accredited by the Accreditation Board for Engineering Technology (ABET).
Formal clinical engineering curricula are offered by colleges and universities. Bachelors, masters, and doctoral degree programs are available in biomedical and clinical engineering. The important difference between the curriculum for
these programs and the curriculum for other engineering disciplines is the mix of engineering and life sciences that it offers. Included are traditional engineering courses (electrical,
mechanical, chemical, computer engineering), as well as
courses in the physical sciences, life sciences (biochemistry,
452
biology, physiology, and anatomy), mathematics, humanities,

and management (5).
It is also possible to enter the clinical engineering field
with a traditional bachelor of engineering degree (such as
electrical engineering) and then acquire the necessary life sciences knowledge by taking supplementary courses, on-the-job
training, and self-study.
Clinical engineers are qualified to pursue advanced degrees in such diverse fields as medicine, law, business administration, health-care management, and technology assessment.
It should be noted that prior to formal degree programs
becoming available in the 1970s, early leaders in the field entered with backgrounds in the physical sciences or life sciences and are considered to be grandfathered into the profession.
The following sources of information are useful (6) :
Service Training. As new medical equipment is acquired,

employee technical knowledge must be updated with regard
to its operation, preventive maintenance, and servicing.
Training is available from the manufacturer or from independent schools. Training can sometimes be included in purchase
requisitions and request for quotations (RFQs) for new equipment. Service training not only benefits the clinical engineer
and BMET involved in maintaining this equipment, it also
benefits the equipment user each time clinical engineering is
called on to assist them with equipment-related questions.
Formal service training can be expensive. In addition to tuition there are travel and lodging expenses. To supplement,
but not replace service training, clinical engineering staff can
attend equipment operator training provided by vendors for
clinical users of medical equipment, within their own institution. Training can also be obtained using the expertise available within the clinical engineering department (Fig. 1).
Directory of Engineering and Engineering Technology: Undergraduate Programs from the American Society for
Engineering Education
Self-Study. Formal training can be supplemented with selfstudy of technical journals, periodicals, and trade publications, as well as equipment operator and service manuals,
VCR training tapes and computer-based training programs.
A clinical engineering library provides an invaluable tool for
the clinical engineering staff and for other health-care workers (physicians, nurses, laboratory technicians) to whom clinical engineering services are provided. Libraries could include
technical video, equipment operator and service manuals, and
technical publications (books and magazines). Such material
also allows staff to keep pace with changes in regulatory requirements and biomedical standards.
Petersons Guide to Undergraduate Programs in Engineering and Applied Sciences

Petersons Guide to Graduate Programs in Engineering
and Applied Sciences
Biomedical Equipment Technician. BMET education leading
to an Associate in Applied Science Degree (AAS) in Biomedical Engineering Technology is offered via seven accredited
programs in the country. One such two-year program is offered by the State University of New York, College at Farmingdale. This program provides balanced course work in electricity and electronics, chemistry, physics, physiology, and
biomedical engineering technology. Students have the option
of continuing their education an additional two years earning
a Bachelor Degree in Electrical Engineering Technology (7).
Continuing Education
Expositions. The environment in which clinical engineering
functions changes daily as new technologies such as telemedicine, robotics, and wireless local area networks (LAN) are introduced into the clinical setting. In this dynamic field continuing education is the rule. One method of obtaining this
education is by attending technical expositions and professional organization meetings. The Association for the Advancement of Medical Instrumentation (AAMI) and the American Society of Healthcare Engineering (ASHE) hold meetings
and expositions that expose attendees to the latest medical
instrumentation being introduced into the marketplace. The
pulse of the health-care industry can be sampled in a relatively short time by attending roundtable discussions and
member and industry presentations. Courses are provided in
regulatory requirements, medical devices, instrumentation
repair and maintenance, clinical engineering and BMET professional certification preparation, and clinical engineering
management. Technical information is also presented at
monthly meetings of the Institute of Electrical and Electronics Engineers (IEEE) Engineering in Medicine and Biology
Society, as well as local clinical engineering and BMET organization meetings.
Safety Training. Employee right-to-know and safety training that discusses the hazards encountered in the workplace
is also necessary. This includes subject matter related to
blood-borne pathogens, hazardous materials, proper protection when entering patient-care areas (gloves, masks, etc.),
environmental hazards, fire hazards, patients bill of rights,
and other items. The latest trend makes use of interactive
computer program modules. This allows training at a time
convenient to the employee and no longer requires attendance
at lengthy seminars.
Certification
Certification is provided for clinical engineers [Certified Clinical Engineer (CCE)] and biomedical equipment technicians
[Certified Biomedical Equipment Technician (CBET)] by examining boards guided by the International Certification
Commission for Clinical Engineering and Biomedical Technology. The Commission is composed of health-care community members including engineering, medical, industrial, and
governmental groups and agencies. Certification provides formal recognition that an individual has mastered a body of
knowledge that is useful in job performance. This knowledge,
which is both theoretical and practical, includes theory of operation of medical equipment, physiological principles, and
safety issues related to medical equipment (8).
Clinical engineering certification requires passing a written exam (multiple-choice and essay questions), and an oral
interview, aimed at determining the candidates depth and
breadth of experience. BMET certification requires passing a
written multiple choice examination. The Association for the
453
Figure 1. In-service education, ventilator

tester. Inservice education can be provided by
manufacturers and vendors, as well as by
clinical engineering staff. Here a clinical engineering supervisor is providing training to
other clinical engineers in the use of an automated ventilator tester. Such devices are used
during preventive maintenance and repair.
They reduce the number of individual test instruments required as they integrate several
test functions into one device. This reduces
service time leading to more rapid equipment
turnaround. Such education also helps to satisfy JCAHO training requirements for clinical
engineering staff.
Advancement of Medical Instrumentation (AAMI) assists candidates by providing certification training courses and study
materials.
Certification renewal requires demonstration of continued
training. Points are assigned and accumulated for various activities that contribute to ones ability to do his job.
Ethics
Confidentiality. Working in a health-care environment,
clinical engineers and BMETs have access to information that
must be kept confidential. If confidentiality is not adhered to
credibility is soon lost.
For example, the following applies
Patient data must not be indiscriminately discussed.
Some service manuals are proprietary.
During the bid process in which new equipment is being
purchased, bidder quotes and bid evaluations must not
be shared with competitors.
Research activities must not be discussed until data are
published.
Work toward the containment of costs by better utilization of technology.

Promote the profession of clinical engineering.
Professional Organizations
Participation in professional organizations exposes a clinical
engineer and BMET to the latest industry trends. These organizations provide up-to-date information, the sharing of ideas,
and networking. National, regional, state, and local organizations exist.
National organizations include:
Association for the Advancement of Medical Instrumentation (AAMI)
American Society of Healthcare Engineering (ASHE)
American College of Clinical Engineers (ACCE)
Institute of Electrical and Electronics Engineers (IEEE),
Engineering in Medicine and Biology Society
Instrument Society of America (ISA)
HISTORY
Code of Ethics. The ACCE addresses these and other issues

in their code of ethics (2), which states that a clinical engineer
will act as follows:
Strive to prevent a person from being at risk of injury
due to dangerous or defective devices or procedures.
Accurately represent my level of responsibility, authority, experience, knowledge and education.
Reveal any conflict of interest that may effect information provided or received.
Protect the confidentiality of information from any
source.
Work toward improving the delivery of health care to all
who need it.
Mid 1960s1970s
Clinical engineerings great impetus for growth occurred in
the 1970s. This came about as follows.
Equipment Problems
During the mid-1960s the medical device industry as a whole
did not yet have adequate performance or safety standards.
Equipment designers were not fully familiar with the requirements of the hospital environment. Equipment design defects
included inadequate energy from defibrillators, ungrounded
equipment chassis, and alarms that could be falsely triggered.
Quality control was also poor as evidenced by physiological
monitors that were grossly out of calibration and equipment
454
that was cracked, broken, or missing components. New medical equipment that was purchased and delivered in supposedly ready-to-use condition was found to have incidence of defects ranging from 25% to 50% (9).
At this time the dangers of microshock and leakage current were starting to be recognized and discussed. Of special
concern was the medical equipment used for coronary care
and the procedures used to maintain this equipment.
During the 1970s professional organizations such as AAMI

and ASHE became more prominent. They sought to promote
safer use of medical equipment. AAMI and the National Fire
Protection Association (NFPA) issued standards for leakage
current and grounding, which further promoted electrical
safety testing of medical equipment (14,15).
1980s
Ralph Nader
Ralph Nader raised national consciousness about accidents
that could occur in hospitals as a result of poorly designed or
faulty medical equipment. His article in the March 1971 Ladies Home Journal claimed that too many hospitals are hazardous electrical horror chambers. To eliminate these dangers, Nader suggested that hospitals hire engineers to provide
advice on electrical equipment and its installation, as well as
on electrical wiring (10). As a result, the clinical engineering
profession was spurred forward as hospitals hired additional
staff to test their equipment and verify electrical safety. This
was also the beginning of independent service organizations
(ISO), which provided an alternative to original equipment
manufacturer (OEM) service. Naders claims have since been
refuted (11).
Kellogg Foundation. The W. K. Kellogg Foundation (established in 1930 to help people improve their quality of life by
providing grants to solve identifiable problems) addressed the
need for improved equipment maintenance prior to the Nader
article when it funded the nations first experimental preventive maintenance (PM) program for hospital equipment. A
three-year grant starting May 1, 1970 was awarded to the
biomedical/clinical engineering department of the State University of New Yorks Downstate Medical Center. The Downstate Medical Center has since changed its name to the
Health Science Center at Brooklyn, University Hospital of
Brooklyn. The department, the Scientific and Medical Instrumentation Center (SMIC) established in 1963, one of the first
biomedical/clinical engineering programs in the nation, is
still active today. The Kellogg Foundation also funded the nations first shared clinical engineering program in 1972, the
Northwest Ohio Clinical Engineering Center. The center provided equipment maintenance, consultation, and educational
services to hospitals in that local (12,13).
Response of the Joint Commission on the Accreditation of
Healthcare Organizations. The Joint Commission on the Accreditation of Healthcare Organizations (JCAHO) also responded to the apparent need for additional safety testing.
Their 1974 standard required quarterly leakage testing for
electrically powered equipment. Their April 1976 Accreditation Manual for Hospitals required hospitals to establish comprehensive instrumentation programs that included preventive maintenance programs with written records of inspection
testing and corrective action taken. It also required all new
patient-related equipment to be evaluated for proper performance before being used clinically (9). The JCAHO requirements further hastened the establishment of in-house clinical
engineering departments that strove to satisfy these requirements as well as to improve on manufacturer-provided maintenance.
Medical Equipment Advances. During the 1980s medical

equipment became more sophisticated. Designers utilized devices, techniques, and technologies that had been developed
and then filtered down from NASA and military contract
work and also focused more on safety. Included were ultrasound imaging techniques, charge-coupled devices (CCD) first
used in spy satellites now used in endoscopic cameras, solidstate electronics, integrated circuits, denser multilayer board
packaging, microprocessors, and dedicated computers. Health
care became more equipment-dependent. The number of medical devices purchased increased, and sophisticated medical
programs and procedures grew in number. The services that
the clinical engineer was requested to provide expanded from
just maintenance, repair, and safety testing of medical equipment to include activities now considered to be part of an
equipment management program, namely patient incident investigation, medical device regulation reporting, equipment
planning, and new equipment acquisition. This further increased the demand for clinical engineers.
Modification of Inspection Requirements
As the economics of health care started to change and funding
became more scarce, the JCAHO modified its PM inspection
frequency requirement, reducing it from quarterly to semiannually. This was done to reduce the cost of compliance to
health-care institutions. In 1988 JCAHO introduced riskbased management. This allowed institutions to create more
realistic PM programs. Gone was the requirement that all
electrically powered equipment had to be tested regularly.
AAMI and NFPA electrical leakage test parameter specifications also became less stringent.
1990s
Running Health Care as a Business. During the 1990s the
effects of managed care, reductions in Medicaid and Medicare
reimbursements, state budgetary problems, university budgetary limitations, and cutbacks in research funding all came
to a head, putting tremendous financial pressure on healthcare institutions. Institutions seeking ways to cut costs and
remain competitive formed alliances, merged (16), and investigated the out-sourcing of support services including clinical
engineering. These economic pressures often resulted in reduction of full-time employees (FTE) throughout the healthcare institution, including within clinical engineering departments, necessitating reexamination of clinical engineering
core services. Outsourcing provided further opportunity for
third-party ISO to expand. The importance of running health
care, including clinical engineering as a business, became the
new paradigm (17).
CLINICAL ENGINEERING IN HOSPITALS

Resources
Health-care facilities of varying size tailor their clinical engineering programs to satisfy the needs of their own specific
environments, while striving to meet regulatory requirements
such as that of the JCAHO. For example, community hospitals with less sophisticated health-care systems tend to require less clinical engineering services than tertiary care
teaching hospital centers.
The depth and breadth of services that clinical engineering
can provide are directly related to the personnel and financial
resources allocated to it. Smaller clinical engineering departments must select services that are feasible for them to provide, that is, concentrating their efforts mainly on the very
core elements of equipment management such as PM and repair. As many clinical engineering services are geared toward
risk reduction, each institution must realize that by choosing
to limit these services, they are at the same time increasing
their risk exposure and the possibility of lawsuits.
Institutions that cannot afford in-house programs turn to
shared services in which several neighboring institutions
share their specialists, or to independent service organizations.
Report Structure
An institutions structure determines to whom clinical engineering reports. Some clinical engineering departments report to facilities engineering and are grouped with other engineering services. Some report to hospital administration.
Others, viewed as university departments, report to a university vice president. Generally, the higher up in the reporting
structure, the more resources are made available to clinical
455
engineering. The department status is also more credible.

These resources include staff allocation, other-than-personnel
service funding (OTPS), and physical plant space.
Department Structure
Full service clinical engineering departments could include
clinical engineers, BMETs, machinists, equipment designers
and prototype builders, and possibly an optics specialist (for
lasers, microscopes, and other optical devices). The technical
staff is supplemented by administrative, secretarial, and clerical staff. Such departments have the capability to be involved
in sophisticated equipment maintenance and repair, equipment modification, and research activity.
Staff Duties. The director responsible for all managerial aspects of the department interfaces with hospital administration and other departmental managers both local and national, and sits on institutional committees. The director
negotiates on behalf of the institution with vendors, manufacturers, and other service providers. He or she sets the course
for the department, adopts policies that provide cost-effective
quality service, and tracks industry trends and standard
practice to benefit the institution. The director also ensures
compliance with regulatory and investigatory agency requirements.
The administrator, a key position, handles personnel issues, as well as matters related to financial analysis, budgets,
billing, tracking of capital equipment purchases, and supervision of the secretarial and clerical staff (Fig. 2).
The clinical engineering managers are responsible for ensuring smoothness of day-to-day operation, assigning jobs,
managing on-call and recall, providing engineering consulta-
Figure 2. Administrative support, purchase

order processing. The clinical engineering administrative function including secretarial
support is critical to a successful department.
The administrator serves a key function and
is heavily involved in tracking the institutions clinical capital equipment purchases
and clinical engineerings own purchases. The
administrator is also involved in budget preparation, monthly financial report analysis,
and preparation of clinical engineering annual reports. Shown here, purchase requisition information is being reviewed to assure
timely processing so as not to delay equipment service. Such delay loses income for
the hospital.
456
tion, and in-service education. They also assist in setting

standards, policies, and procedures.
The clinical engineers are involved in acceptance testing,
PM, repair, on-call and recall, providing emergency assistance, design and prototype construction, and in-service education.
The BMETs are primarily responsible for performing PM,
repair, and calibration of equipment, per procedures set by
clinical engineers.
The clerical staff assists with data entry and documentation filing.
Job Titles. The Journal of Clinical Engineering conducts an
annual nationwide survey of salaries and responsibilities for
hospital biomedical and clinical engineering and technology
personnel (18). This survey includes a set of generic titles and
generalized job descriptions that provide a convenient industry overview. These titles follow:
Jr. BMET
BMET
Sr. BMET
Equipment Specialist [Laboratory (RES) or Radiology
(RES)]
BMET Supervisor
Clinical Engineer (CE)
Clinical Engineer Supervisor
Director or Department Manager
Goal and Responsibilities
Goal. A hospital-based clinical engineering departments
goal is to support its institution in its mission (typically patient care, education, and research), and while so doing en-
Figure 3. Research design and development,

optical tomographic system. Research activities enhance a clinical engineering departments image and keep the staff current with
the latest technological developments. Specialpurpose devices that are not available commercially or are cost prohibitive are constructed.
A team with diverse expertise in electronics,
mechanics, electromechanics, and physiology
is required. A precision machinist plays a
prominent role. Shown, a positioning device is
being modified for incorporation into an optical
tomographic laser system, which one day may
prove as clinically beneficial as MRI.
sure the safety and efficacy of the hospitals medical instrumentation. This goal is achieved by providing appropriate
technical services. These services can range from the basic
maintenance, calibration, and repair of medical equipment, to
the more sophisticated research activities of design and development of medical equipment and devices usually associated
with biomedical engineering, thus resulting in some overlap
between these two disciplines (Fig. 3).
Equipment Responsibilities. Clinical engineers apply engineering and management principles to issues that relate to
medical equipments entire life cycle. They help determine
what equipment to purchase and how long it is cost-effective
to keep this equipment in service, and when to turn to newer
technologies. Such guidance saves hospitals money and reduces liability.
Clinical engineers manage a diverse group of medical devices located throughout their institutions. This includes instrumentation used in cardiology, intensive care, clinical laboratory, respiratory therapy, anesthesiology, neurology,
physical therapy, ultrasound, and the operating rooms. Some
clinical engineering departments also provide service for xray or ionizing radiation devices used in radiology, radiation
therapy, or nuclear medicine that are typically managed by
radiation physics staff. Others may service purely mechanical
devices such as stretchers, hospital beds, and wheelchairs,
but these are usually managed by facilities engineering.
Clinical engineers provide emergency instrumentation
troubleshooting expertise. This can take place within an operating room during cardiothoracic surgery, a patient-care
area (Fig. 4), or in a researchers laboratory during animal
experimentation. Typically clinical engineers do not operate
the medical equipment or select levels of treatment (i.e., balloon pump inflate or deflate timing), or give fluids to or take
457
to investigate the feasibility of providing a broader range of

services. They are becoming areas of excellence for items that
were previously outside their domain, including x-ray devices,
computers, telecommunications, and nurse-call systems.
Some also take on risk management and many clinical engineering departments have technology assessment responsibilities.
Clients
As a service department, clinical engineers interface daily
with staff from most other departments and entities within a
health-care institution, all of which are considered to be clinical engineering clients. A partial list includes facilities engineering and planning, hospital administration, cardiothoracic
surgery, central sterile, expenditures, contracts, risk management, OPD administration, clinics, clinical laboratories, ambulatory surgery, pharmacy, surgery, anesthesiology, off-site
satellite clinics, clinical areas, nursing units, the personnel
department, purchasing, and cardiology. Interaction between
researchers and educators also occurs.
Committees
Figure 4. Emergency support, balloon pump. Clinical engineering

plays a key role in providing emergency support to troubleshoot and
answer questions relating to equipment operation and capability.
Here a clinical engineer is running a test to assure that a balloon
pump located outside a Cath-Lab is functioning properly. This device
provides support to critically ill patients by reducing their hearts
workload until it strengthens. Balloon inflate/deflate timing and
mode of triggering assure optimal patient assistance.
fluids from patients (i.e., cell saver). This is left to clinical

specialists, perfusionists, and licensed technicians specially
trained for these purposes. However, clinical engineers do
provide instrumentation troubleshooting expertise during
these procedures and provide guidance on the operation and
performance of the equipment.
Typically clinical engineers do not maintain the physical
plant. They deal with medical equipment external to the
walls. This equipment may require connection to utilities including electricity, gases, and water, as well as to other electrical systems that may not fall within clinical engineerings
domain [e.g., patient line isolation monitors (LIM) within
headboards or nurse call systems which interface with bedside monitor alarms]. Knowledge of the physical plant and
such systems is, however, helpful especially during equipment selection and installation, when analyzing the cause of
equipment failure, and when setting policies for equipment
use during utility failures. A useful reference guide is the National Fire Protection Association (NFPA) Health Care Facilities Handbook (15).
Changing Responsibilities. As health care changes, some
hospital-based clinical engineering departments are starting
Clinical engineering participation in committees is important

for a successful clinical engineering program. It provides clinical engineering with greater exposure to other hospital departments and administrators and vice-versa and provides information about how the department is doing which
supplements the formal survey process. Equipment-related
problems and questions voiced allow clinical engineering to
provide immediate feedback to the clinical user. This keeps
open and improves channels of communication between clinical engineering and their clients. It allows clinical engineering to become essential members of the multidisciplinary
team of health care delivery and have an input on decisions
relating to that delivery. It also allows trends to be spotted
that clinical engineering staff may be unaware of, for example, equipment that is down but was not yet formally reported
to clinical engineering.
Committees include:
Safety committees such as safety and laser safety.
Clinical committees such as neonatal interdisciplinary,
special care units, adult critical care, cardiopulmonary
resuscitation (CPR), infection control, and the Institutional Review Board.
Equipment-related committees such as capital acquisition,
product standardization, and sole source (some of which
may be chaired by clinical engineering).
Quality assurance and investigatory planning committees
such as clinical departments, quality assurance, and
those related to JCAHO such as Environment of Care.
Ad hoc special committees such as those for efficient lighting studies, research activities, year 2000 (Y2K) compliance and so on.
Physical Plant Requirements
To ensure efficient services, clinical engineering must be allocated adequate facilities to allow all clinical engineering functions to be performed. This includes sufficient space to store
new equipment delivered, equipment awaiting servicing, and
458
equipment awaiting delivery back to the user, as well as

equipment sequestered because of its involvement in a patient incident. Space must also be adequate to house all of the
tools, test equipment, computers, office equipment, parts and
if possible a machine shop. As clinical engineering is the central repository of all regulatory-related medical instrumentation history files both active and inactive, adequate accessible
storage for them is needed as well as for the equipment operator and service manuals and technical library.
Test space must be such that it allows performance of acceptance testing, PM, and repair. It must contain appropriate
electrical power, suction, compressed air, secure gas tank
storage, proper lighting, ventilation, sinks, fume hoods, workbenches, and storage cabinets.
Test Equipment, Tools, and Test Fixtures
Required test equipment includes electrical safety analyzers,
physiological simulators, oscilloscopes, power supplies, multimeters, ventilator testers, electrosurgery analyzers, waveform generators, photometer or radiometer lightmeters, laser
power meters, etc.
Tools required include screwdrivers, pliers, wrenches,
drills, soldering stations, etc. A machine shop equipped with
a drill, lathe, grinder, and milling machine is useful.
It is beneficial for the clinical engineering staff to construct
setups of equipment, which are readily available for equipment testing purposes during acceptance testing, PM, and repair. These devices can sometimes be purchased, but most
times they can be put together in-house. All such devices
should be inventoried in a test-fixtures manual for ease of
access.
Full-Service In-House Department
Service Overview. A full-service clinical engineering department provides a multitude of services. As example, the
biomedical/clinical engineering department (Scientific and
Medical Instrumentation Center) of the State University of
New York, Health Science Center at Brooklyn, University
Hospital of Brooklyn has a broad-based clinical engineering
program that includes biomedical engineering components
such as support for research and education (19).
Its program includes the following:
Clinical engineering consultation
Design, construction, and modification of clinical instrumentation and devices including electronic, electromechanical, and mechanical
Education
Research and Development assistance
Patent and grant assistance
Institutional Review Board (IRB) approval assistance
Equipment planning for clinical areas and new programs
Generating reports to administration including clinical
capital equipment purchase tracking
Computerized clinical instrumentation inventory
Centralized patient-care instrumentation history files
Centralized instrumentation technical manuals library
Equipment evaluation library
Instrumentation pre-purchase evaluation
RFQ generation
Purchase requisition review
Bid evaluation
Vendor and manufacturer interface
Coordination of outside services
Acceptance testing (initial checkout, incoming inspection,
incoming test) of new equipment safety, operation, and
technical specifications
Clinical equipment installation coordination and supervision
Defect resolution and documentation
User in-service training
Preventive maintenance test procedure generation and update
Testing of rental, loaner, demonstration, patient-owned,
and physician-owned equipment for hospital use
Scheduled PM
Equipment repairs
Equipment upgrades
Oversight and evaluation of equipment service contracts
Emergency clinical engineering support to all patient-care
areas
On-call and recall for critical care areas
Specialized clinical engineering support dedicated to cardiothoracic surgery
Quality assurance and risk management
Regulatory agency survey support
Equipment related patient incident investigation
Hazard and recall alert notification
Clinical engineering participation on hospital and health
center committees
Represent hospital on the University Healthcare Consortium (UHC) Clinical Engineering Council
Represent hospital in the New York City Metropolitan
Area Clinical Engineering Directors group
BMET internship programs
NYC Board of Education Substitute Vocational Assistance
(SVA) internship programs
Volunteer training
Clinical engineering staff and departmental development.
Equipment Modification. At times clinical engineering is
called on to modify instrumentation. Modification must not be
done indiscriminately. Care must be taken so as not to violate
the integrity of the equipment. It is best to limit modifications
to external operations. Nonmanufacturer approved internal
modifications must be approached with extreme caution and
are best not done as they may void warranties and violate
FDA guidelines. This includes securing devices to a cart so
they will not fall off in transit or be stolen, assembly of devices into working systems, and modification of equipment to
allow easier PM. For example, a monitor used in an endoscopic video system may have to be secured to a cart, or, a
medication cart may have to be modified to allow its use as a
crash cart. Crash carts typically are medication carts that
house a defibrillator, suction device, O2 tank, and supplies.
Purchased as separate entities, integration is required. Elec-
459
Figure 5. Equipment modification, crash

cart. Equipment modification takes many
forms and runs the gamut from modifying a
specific device to assembling separately purchased components into functioning systems
which may not be available commercially, or
which must have specific characteristics to
satisfy institution requirements. Shown here,
a medication cart was modified by clinical engineering to accept an electrical outlet strip
and retractable reel (with 25-ft line cord), O2
cylinder, a defibrillator, and an aspirator so
that it can be used as a cardiac arrest crash
cart, which is brought to the patient bedside
for resuscitation during patient cardiac or respiratory emergencies.
tric reel extension cords, auxiliary power outlets, as well as

on/off switches and tie-down straps to prevent defibrillator
removal must be added (Fig. 5). Another example is the construction and assembly of an operating room (OR) transport
cart connected to the patient bed upon transport from the OR
to the Cardiothoracic Intensive Care Unit (CTICU), which
houses a ventilator, monitor, defibrillator, and dc to ac supply.
An example of a modification that assists in PM is the external provision for a jumper (with manufacturer approval),
which when removed allows the backup thermostat of a hyperthermia blanket to be tested, without the need for dismantling the unit. Equipment controls may also be physically constricted so as to prevent inadvertent operator-induced error.
Examples include covering of stylus heat adjustment controls
(after optimization), limiting rotation of a ventilator alarm
control to prevent complete alarm shut-off, and plugging an
unneeded ECG sync-pulse phone jack output to prevent its
use with a defibrillator as it induces too much signal delay.
Research and Education. Most health-oriented universities
with teaching hospitals have education, research, and health
care as their mission. Clinical engineering is capable of providing support for all of these goals. Clinical engineering assistance with health care has already been discussed. With
regard to education, in addition to providing in-service education, clinical engineers are called on to teach courses in the
health-related professions. Clinical engineering support for
research activities includes assistance in the selection of experimental equipment, setting up a laboratory, measurement
technique, grant writing, patent applications, and IRB approval. Also included are the design, prototype development,
and final construction of special-purpose devices that are not
readily available commercially or that can be more cost effectively built in-house.
Involvement in research and educational activities are
beneficial to a clinical engineering department. These challenging opportunities help to keep the staff technically competent and involved at the forefront of technology. Such activities also bring prestige to the department, enhancing its
professionalism and reputation, helping it to gain additional

resources.
REGULATORY REQUIREMENTS
It is necessary that clinical engineers become familiar with
mandated standards, voluntary standards, accreditation body
requirements, and licensing agency requirements that apply
to their particular health-care institution and to the medical
equipment for which they are responsible. Typical examples
are shown below (20). An in-depth list of Biomedical Standards is available in The Guide to Biomedical Standards (21).
Voluntary Standards Organizations
American Association of Blood Banks (AABB)
American Dental Association (ADA)
American National Standards Institute (ANSI)
American Society of Histocompatibility and Immunogenetics (ASHI)
Association for the Advancement of Medical Instrumentation (AAMI)
College of American Pathologists (CAP)
National Fire Protection Association (NFPA)
Underwriters Laboratories (UL)
Governmental Agency Standards
Federal Communications Commission (FCC)
Food and Drug Administration (FDA)
Standards for the Operation of Hospitals
Local and state requirements, such as the Department of
Health
American Hospital Association
Joint Commission on Accreditation of Healthcare Organizations (JCAHO)
460
Joint Commission
The Joint Commission on Accreditation of Healthcare Organizations (JCAHO) runs voluntary three-year accreditation programs for health-care facilities aimed at improving the quality of patient care. Information gathered about a hospital can
be released to the public. Accredited health-care facilities are
eligible to receive federal Medicare reimbursement. Many
state governments recognize accreditation as a requirement
for licensure and Medicaid reimbursement (22).
The environment of care (EC) in which todays health care
is provided is complex. It includes plant facilities, medical
equipment, drugs, information, finance, staff, third-party services, and diverse technologies (23). JCAHO is concerned that
this environment be managed so as to provide a hazard-free
environment that reduces the risk of human injury.
JCAHO requires management programs to be set up that
deal with safety, security, hazardous waste, emergency preparedness, life safety, medical equipment, and utility systems. Clinical engineers tend to be most involved with activities that constitute a medical equipment management
program, the purpose of which is to promote the safe and effective use of the institutions medical equipment. The medical equipment management program encompasses equipment
acquisition, technical management, and education for both
equipment operators and maintainers. As part of this program, JCAHO requires clinical engineering to submit periodic
reports to the institutions safety committee. Performance
standards (quantifying factors relevant to program effectiveness) are developed and selected indicators (activities) such
as those dealing with timely PM and repair performance are
tracked. Changes observed in the indicators are used to spot
and correct deficiencies in the clinical engineering program.
The aim of this activity is to improve the quality and costeffectiveness of the clinical engineering services provided (24).
Safe Medical Devices Act
The Safe Medical Devices Act (SMDA) of 1990 and its 1992
amendment requires health-care institutions to report equipment incidents resulting in serious injury or death to a patient or employee. An institutions risk manager determines
whether the incident is reportable, using a documented decision-making process. Medical-device-related deaths must be
reported within 10 days to the FDA and to the manufacturer.
Medical-device-related serious injuries or illnesses must be
reported within 10 days to the manufacturer, or if the manufacturer is unknown, to the FDA. Periodic summary reports
are also required. The SMDA also requires that specific medical devices be tracked, and that medical equipment be properly disposed of when it is taken out of service (25,26). SMDA
compliance is also a requirement of the JCAHO.
However, as a result of a reform bill, the FDA Modernization Act of 1997, in a few years all hospitals may no longer be
required to submit reports to the FDA when patient deaths
or serious injuries involving medical devices occurs. Instead
the FDA will rely on a small sample of representative hospitals and nursing homes called sentinels to collect the data.
Patient Incident Investigation
When a patient incident occurs, incident reports from nursing, physicians, and others are submitted to the risk manage-
ment department. A clinical engineering technical evaluation

is also prepared and submitted. Risk management staff uses
this material to determine what caused the incident (e.g., was
the equipment at fault or was operator error indicated). The
intent of the investigation is to prevent a recurrence and to
determine if the incident must be reported under the SMDA
of 1990.
Clinical engineering should be notified about an incident
as quickly as possible to allow a thorough technical evaluation to be made. Doing so may allow investigation and on-site
testing to be done with the equipment setup still intact (making note of how the unit was used, dial settings, etc.) and with
peripheral equipment still in place. Subsequently the instrument, accessories, and disposables are taken out of service
and sequestered. Further inspection and testing may be required within the clinical engineering laboratories (Fig. 6), or
by a third party. Such determination is made by clinical engineering working with the risk manager. The manufacturer
should only be contacted with the risk managers approval.
Sometimes during investigation, minor equipment problems
are detected, that could not have caused the incident. Prior
to repairing these, the risk manager should be consulted to
determine the legal ramifications. Determination must also
be made as to whether the equipment needs modification to
prevent future recurrences. The clinical engineer report will
be important should a lawsuit ensue. For this reason imprecise language must be avoided, so as not to jeopardize the
institutions legal position. For legal reasons, the equipment
must be stored in a secure location and not put back into service until the risk manager concurs.
User Error; Equipment Abuse; No Fault Found
It is important that clinical engineering workers track service
requests whose resolution indicates no fault found (NFF),
equipment abuse, or user error. This information should be
submitted to risk management for analysis even if patient injury did not result. User errors are typically more common
than true equipment malfunctions. JCAHO requires that user
errors that have a potential for harm receive the same type
of review that hazardous equipment failures receive. Risk
management analysis may indicate the need for additional
user in-service education to alleviate future problems.
Informal one-on-one training is provided by clinical engineering staff to the user when returning such equipment back
to service, by demonstrating the proper equipment operating
technique. Through this educational activity clinical engineering helps improve patient care and reduces the possibility of lawsuits.
Hazard Alerts and Recalls
The clinical engineering department acts as the hospitals
hazard and recall coordinator. Typically the manufacturer notifies clinical engineering and risk management about recalls
and alerts. At other times the clinical engineering department, upon review of commercially available listings, notifies
appropriate hospital departments including risk management. Hazard alerts and recalls are available from the FDA,
as well as from private publishers such as the Emergency
Care Research Institute (ECRI) and Quest.
The clinical engineering department queries the equipment inventory list to locate equipment affected, and when
461
Figure 6. Incident investigationfiber optic

light. Equipment-related incident investigation is conducted by clinical engineering
whenever there is the possibility that a medical device may have caused injury to a patient
or clinician. This requires investigation at the
scene, as well as additional testing within the
clinical engineering laboratories. Picture taking (a digital camera is most useful) documents observations. Clinical engineering staff
may also anticipate and resolve equipment
problems before they result in an incident.
Shown here, an examination lamp bracket
was found to not meet the lamp OEM specification, which could result in the lamp being
easily dislodged and falling. The bracket
manufacturer in coordination with the lamp
OEM worked with clinical engineering to resolve the issue and supplied newly designed
brackets.
required removes it from service and sequesters it until remedial action is taken. Should equipment retrofiting be required,
the manufacturer may choose to provide an upgrade kit with
instructions, opt to send a field-service engineer on-site, or
require that the equipment be picked up or sent to the factory. Appropriate paperwork must be provided to the institution for inclusion in the instruments history folders, and entries made into the computerized equipment records. Updated
operators manuals and additional user in-service training
may also be required.
The equipment that a typical university hospital clinical

engineering department such as the Scientific and Medical
Instrumentation Center is responsible for (excluding x-ray or
ionizing radiation devices) runs to 10,000 active items. These
include capital and noncapital devices. Capital devices are
classified as costing greater than or equal to $500 per item.
Noncapital devices cost less than $500. To describe this equipment approximately 500 different equipment nomenclatures
are used. This alone shows the diversity of knowledge that a
clinical engineering staff must have.
MEDICAL EQUIPMENT
Patient Monitoring
Patient-Care Equipment
Equipment used on patients or for patient care in healthcare facilities is both varied and numerous. It runs the gamut
from simple thermometers to sophisticated MRI machines.
Equipment used on the patient, such as an electrocardiogram (ECG) monitor, is readily visible in the patients immediate physical vicinity. Equipment used for patient care,
such as a clinical chemistry analyzer, may be housed in a
laboratory at a location remote from the patient. Both types
are important when considering the environment of patient
care.
Medical equipment falls mainly into three different categories. These categories are diagnostic, therapeutic, or assistive.
Diagnostic equipment such as a monitor acquires data and
uses transducers to enhance and supplement human senses.
Therapeutic instruments such as high-voltage X rays, pacemakers, and defibrillators arrest or control physiological processes affected by disease or trauma. Assistive devices supplement diminished or lost functions, and include life-support
(ventilator) and life-sustaining (dialysis unit) devices (27).
Medical instrumentation used for patient monitoring has become quite sophisticated. This microprocessor-controlled
equipment provides multiphysiological parameter monitoring
with alarm generation and recording capability. It incorporates telemetry, ST segment analysis, and full physiological
parameter disclosure capability (which stores selected waveforms for recall), allowing clinical study of abnormalities. It
also includes automatic arrhythmia detection at the bedside,
which until a few years ago required a large stand-alone computer housed in a specially cooled room. Using individual personal computers, patient data can also be collected and archived for additional statistical studies.
The patients physiological parameters are viewed on bedside monitors as well as on remote slave displays. Parameters
monitored include ECG, heart rate, respiration rate, cardiac
output, noninvasive blood pressure, invasive blood pressures
(arterial, pulmonary artery, central venous, etc.), oxygen saturation (SAO2), pulse rate, end-tidal carbon dioxide (ET CO2),
and temperature.
In critical care areas, the bedside monitors are hard-wire
connected to central nursing stations allowing centralized
462
Figure 7. Central nursing station. Physiological monitoring has grown quite sophisticated.
Patient information gathered at the bedside is
routed to a central nursing station providing
clinical staff with a comprehensive viewing
area. Each central station monitor typically
shows waveforms and parameters for four different patients and has the ability to zoom in
on a specific patient to show all monitored parameters. Recorders provide documented
printouts of alarm conditions including detected arrythmias. Closed circuit TVs visually
monitor patient isolation rooms as well. Clinical engineering is involved with the entire life
cycle of such equipment from prepurchase selection through acceptance testing, PM, repair,
and eventual obsolescence retirement.
viewing at one location (Fig. 7). Nursing stations may be connected together via local area ethernet-type networks,
allowing remote patient viewing between nursing stations
and sharing of full disclosure equipment. Telemetry information is likewise routed to a nursing station for centralized
viewing of ambulatory patients. In this case, the telemetry
transmitter takes the place of the bedside monitor, transmitting a signal to an antenna system that routes it to a receiver
and display unit.
PM. As much information as possible should be included for

each piece of equipment in the list (30), such as the following:
Unique identification number, which could be a property
control asset number, but is usually assigned by clinical
engineering and is generally not the serial number
Equipment manufacturer, model, serial number, and description (nomenclature)
Equipment location
Equipment Classification; Nomenclatures
Purchase order number
Equipment classification nomenclature systems bring order to

the vast array of medical equipment presently in use. These
systems simplify the gathering and distribution of data relating to medical devices. Complete nomenclature listings are
found in ECRIs Health Devices Sourcebook (28), and the Medical Device Register (29). Another nomenclature system was
developed by the U.S. Food and Drug Administration (FDA)
as part of its regulatory responsibilities for medical devices.
ECRI and the FDA are presently attempting to standardize
their two systems.
In the sample nomenclature listing below, note the two
ways of listing an ECG monitor.
Departmental owner
Service organization responsible for the equipment (inhouse, contract, etc.)
Acceptance date, when approved for clinical usage
Warranty expiration date
Equipment acquisition cost
PM frequency and PM procedure number to be used
Additional information the organization believes useful for
proper equipment management
Equipment Records
Cart, resuscitation
Pacemaker, cardiac
Heart rate monitor, ECG
ECG monitor
Diathermy unit
Equipment Inventory List
To be effective an equipment management program requires
maintenance of an up-to-date, complete inventory of medical
equipment used in the health-care institution. This equipment inventory list helps identify equipment for product recall and hazard alerts, as well as to locate equipment due for
Equipment history files are maintained to provide information for equipment management and technology assessment
purposes, as well as to satisfy regulatory requirements. When
equipment is taken out of service and is disposed of, its history file should be maintained for a minimum of three additional years (31), or longer if an institutions legal council or
risk manager deems it necessary. This will offer the institution some protection in the event that a patient incident lawsuit is initiated at the time of equipment disposal, of which
clinical engineering or risk management is unaware. Records
for equipment involved in patient incidents are usually sequestered by the risk manager so as to avoid possible tampering.
ELECTRICAL SAFETY
Ongoing Testing
Medical equipment is tested for electrical safety throughout
its lifetime. Baseline tests are run during acceptance testing.
Tests are also run during PM, following equipment repair,
upgrade, or patient incident. The measurements are recorded
and compared to previous readings. Changes indicate possible
electrical degredation that must be investigated to eliminate
electrical hazards before an incident can occur. Training of
equipment users in electrical safety concepts is also important.
Electrical Safety Analyzers. Electrical safety analyzers are
used to determine that electrical devices, ac receptacles, and
conductive surfaces meet required safety standards and are
safe for use. These solid-state instruments incorporate true
rms measurement capability. They allow testing of portable
medical equipment and fixed (hard-wired) installations. Internal circuitry [AAMI test load (14)] simulates the human
bodys impedance to current flow. The measurements made
are representative of the leakage currents (if present), which
could flow through the body. Normal and reverse polarity
tests, as well as current source tests, are run (32).
Micro- and Macroshock
Electrical safety as related to medical instrumentation concerns itself with limiting the amount of electric current allowed to pass through the body to a few microamps. This limits the current density (current per unit area) to values below
a threshold that could affect or damage tissue and vital organs such as the heart and brain (33).
In a health-care setting patients are compromised when
their skin is punctured and catheters are inserted, or when
their skin is prepped (rubbed and cleaned with alcohol) prior
to the placement of electrodes, and where moist environments
exist. The electrical resistance of patients bodies to current
flow is reduced from its normal range of 10,000 to 100,000
, to a range of 1000 to 10,000 . Under normal conditions
110 V ac applied to the skin results in currents of 1 mA to 10
mA. Under these compromised conditions larger currents of
10 mA to 100 mA result.
Macroshock (current above 1 mA) can be hazardous when
delivered at the bodys surface. For example, 100 mA applied
at the skin could cause ventricular fibrillation. Microshock
(current below 1 mA) can be hazardous when delivered directly or close to heart tissue. For example, current in the
order of 0.1 mA may cause ventricular fibrillation. Currents
such as these that can injure the patient are usually too low
to affect the uncompromised equipment operator.
Ac Leakage Current
Ac leakage currents are found in electrical instruments other
than battery-operated direct-current (dc) devices. Leakage
currents are produced as a result of the ac signal coupling to
the chassis of the instrument due to capacitance effects. Such
currents flow from chassis to ground when a low-resistance
path is made available.
The ground wire within the equipments three-wire line
cord provides a safe low-resistance path for the leakage cur-
463
rent. It is for this very reason that two-wire line cords are
prohibited for hospital use. The ground wire is connected to
the chassis of the instrument on one end and to the ground
pin of the ac plug on the other. While this connection is intact
the leakage current is safely conducted away from the patient, as it flows from the chassis through the ground wire to
ground via the ac wall outlet. Should this path open or present a high resistance from chassis to ground due to a loose
wire connection in the plug, or an improperly grounded ac
outlet, the leakage current seeking other pathways could flow
through the compromised patient. Leakage currents can also
flow between patient leads and ground due to poor lead isolation. The large number of medical devices that surround and
could route electrical current to the patient compounds the
problem.
Manufacturers limit leakage current by (34):
Incorporating patient isolation circuitry utilizing isolation
amplifiers, optical coupling, and infrared transmission
techniques
Doubly insulating some devices with an outer nonconductive plastic housing so that even if touched, they cannot
conduct electricity
Using specially constructed low-leakage ac line cords
Incorporating isolation transformers into systems, which
have components whose total leakage current exceeds
safety standards
Hospital Grade Plugs and Outlets
Safety is also provided by use of heavy-duty hospital grade ac
plugs (with a green dot). These plugs are mechanically keyed
to prevent polarity reversal. Explosion-proof plugs previously
used due to the explosive nature of some anesthetic gases are
no longer prevalent. Prior to opening new clinical areas, in
addition to having the clinical gases certified, all ac outlets
should be tested with a tension tester to verify that the ac
outlets will tightly grip equipment plugs when inserted and
with an ac polarity checker to ensure that the wiring has been
properly done.
PROCUREMENT OF MEDICAL DEVICES
Reasons for Equipment Acquisition
Equipment is acquired by a health-care facility for a multitude of reasons, including the following:
Replacement of obsolete equipment that cannot be repaired as parts are no longer available or that is not
cost-effective to repair as a new unit would be comparable in price to the repair cost. Included is equipment
that breaks down frequently, resulting in lost patient
revenue to the institution. Such equipment replacement
increases the hospitals cost-effectiveness and reduces
its risk exposure.
Replacement of technologically obsolete equipment that is
not as precise as newer microprocessor equipment, to
improve diagnostic and therapeutic efficiency.
Introduction of new types of technologies, such as magnetic
resonance imaging (MRI) and Catscan to provide enhanced services.
464
Requirement of additional units of a type already being

used in the facility to reduce equipment downtime and
patient waiting.
Attracting highly qualified physicians including new department chairmen.
Provided free of charge to the institution as part of a disposable contract.
Brought into the facility by clinicians for specific practice
purposes.
Loaned to, or rented by, the institution.
Clinical Capital Equipment Committee
Equipment acquisition usually starts with a perceived need
expressed by a clinician, a hospital administrator, or clinical
engineer and a request is forwarded to the institutions Clinical Capital Equipment Committee. However, equipment is
sometimes purchased on an emergency basis based upon medical contingencies or for political reasons without committee
input.
The Clinical Capital Equipment Committee is made up of
clinical department chairpersons, physicians, hospital administrators, as well as representatives from nursing, clinical engineering, finance, and purchasing. The committee reviews
the equipment requests. Clinical engineering staff provides
equipment inventory lists, instrumentation repair trends, and
other equipment management information requested to expedite the decision-making process. Priorities are determined, a
purchase list is generated, and requesting departments are
notified. They prepare appropriate purchase requisitions, and
necessary hospital administration signatures are obtained.
The purchase requisitions are then submitted to clinical engineering for technical review.
Turn-Key Installations
Large turn-key installations require a request for quotation
(RFQ) to be prepared for a bid process. Turn-key installations
require the vendor to provide all equipment, materials (cables, mounting devices, etc.), and labor to install the system
completely, and, when ready, to turn it over to the institution for acceptance testing. The RFQ document includes
equipment specifications, environmental specifications, and
legal requirements that address issues of noncompliance and
penalties. During installation, such systems may require extensive vendorclinical engineering interaction and problemsolving, as fully detailed documentation is not always
possible. They also require extensive acceptance testing. Subsequent to the bid award, a detailed purchase requisition is
generated.
Purchase Requisition Review
Assists the clinician in obtaining needed equipment, ensuring
that everything required (peripheral items, supplies, etc.) is
being ordered and that all items are compatible with each
other and with existing equipment. It also ensures that the
physical plant is ready (e.g., water, gas, special electrical requirements) so that equipment installation and use will not
be delayed.
Requisitions are first reviewed to determine the following:

If equipment falls within clinical engineering jurisdiction
(i.e., items used in the health-care facility for which
clinical engineering is responsible).
If the FDA has approved the equipment for clinical use. If
approved only for investigational purposes (i.e., a research phase requiring clinical trial to prove its efficacy), clinical engineering staff could assist the clinician
in obtaining IRB clearance for clinical trials.
If the equipment utilizes a new technology requiring an
engineering evaluation and clinical trial period prior to
purchase. Evaluation may also be required if several
vendors have viable products that should be compared.
Visits to other health-care institutions are sometimes
required to view the equipment in use. Larger systems
such as replacement of all of an institutions obsolete
physiological monitoring equipment necessitates input
from future users including physicians and nurses.
If the equipment requires special physical plant utilities or
has physical attributes (size, weight) that the facilities
engineering department must be made aware of. If so,
the facilities engineering managers purchase approval
is required.
If equipment or accessories require special treatment to
not pose an infection threat to patient or user (i.e., sputum chamber certification). If so, the infection control
department should be notified so that appropriate hospital policies will be generated.
If equipment is year 2000 (Y2K) compliant. If not, the impact of this equipment on patient care must be determined.
If sole-source justification (exemption from advertisement)
is required. Sole-source acquisition is justifiable if the
unit must be compatible with an existing item, a vendor
holds a service contract and must supply parts, the unit
has unique features needed by the requester, or no competition by manufacturer or vendor exists.
Purchase requisitions are next checked to ensure the following (35):
All needed accessories have been specified and are compatible.
Vendor or manufacturer will uncrate, assemble, or calibrate the unit (if required).
Vendor will assist with acceptance testing (if required).
Vendor will install the equipment (if required, i.e., mount
to walls, etc.).
Vendor will provide (or loan) test kits or fixtures (i.e.,
phantom for diagnostic imaging) or simulators specifically geared to the unit.
Vendor will provide user in-service training.
Vendor will provide sufficient number of operator and service manuals.
Vendor will provide VCR training tapes.
Vendor will provide acceptance testing and PM protocols.
Vendor will provide clinical engineering service training.
465
Vendor will provide an equipment loaner in the event of

delayed delivery.
Vendor will provide system isolation transformers (if required).
Specification of the correct delivery location (clinical engineering). This is true even for large items so that the
clinical engineering department will be aware of delivery, at which time the receiving department could be
notified to route the unit to the intended user site.
Contact person has been specified should the vendor have
to make arrangements with the clinical engineering department or for training purposes.
Sufficient start-up materials are specified both for acceptance testing and for start of clinical use.
Warranty period is specified and service contract specified
(if required).
Although clinical engineers must concentrate on the technical issues, they might also verify quotations, specify discounts if appropriate, ensure that buying service pricing is
adhered to, determine if special promotions are offered, see if
trade-in of obsolete equipment is feasible, and check on availability and delivery dates.
The requisition is next submitted to the purchasing department. A copy of the entire paperwork package, including
all technical information gathered, is stored in clinical engineerings open purchase requisition file awaiting equipment
delivery. After delivery and acceptance, it will be stored in
the equipments history file.
Bid Review
This review assists the clinician and purchasing department
in determining if a low bidder offering an equivalent unit to
what has been specified meets clinical requirements.
Depending upon the institution, equipment cost, and if a
sole source is not justifiable a bid process may be required.
Following bid opening a purchasing department bid analysis
is sent to the clinical engineering staff and to the clinical requester. Working together, they determine if the equivalent
device proposed by the low bidder is a viable alternative that
meets clinical needs and the important specifications of the
desired unit. If not, the more expensive unit may be justifiable. This process requires comparison of the low bidders
equipment to the unit originally specified and bid comparison
to ensure that items have not been excluded that could artificially lower the price. The low bidder may have to supply a
loaner unit for engineering test and clinical trial. Subsequently, a letter of justification is written, the award is made,
and the equipment delivered.
Equipment Acceptance Testing
Equipment acceptance testing, also know as initial checkout
or incoming inspection, ensures that all items ordered have
been received and are undamaged, the equipment functions
as per the manufacturers performance specifications, and the
equipment is safe for both the clinical user and the patient.
Acceptance testing uncovers equipment defects including
those not readily apparent, prior to the equipment being used
on or for patients so as to reduce liability to the institution.
Such testing is usually more in-depth than PM testing and is
Figure 8. Acceptance testing, endoscopic system. Acceptance testing

assures that medical equipment functions properly, meets manufacturers specifications, and is safe for use. It also verifies that all items
ordered have been received, and appropriate in-service education is
provided to the clinical and engineering staff. Shown here, an endoscopic video system is undergoing an acceptance test. Such devices
allow intra-body images to be displayed on monitors for ease of viewing and allow their documentation via video recording or printout.
much more than just electrical safety testing (leakage current

and grounding resistance). All medical equipment (whether
purchased, leased, rented, loaned, physician owned, used as a
demonstration model, or donated) should undergo acceptance
testing. Short-term items are given a loaner tag, while longterm items are assigned an inventory number.
Acceptance testing should be thorough (36). A visual inspection externally, as well as internally (when justified), ensures that the instrument was not damaged in transit and
has no loose or extraneous components. The visual inspection
also verifies that the device is new, of the latest model, and
has not been previously used (occasionally factory refurbished
demonstration units may be purchased). Devices are checked
for electrical safety, mechanical safety, and functionality and
to assure that they meet the manufacturers own performance
specifications (an important concept). Built-in diagnostics are
run to provide future confidence in them (Fig. 8).
A report is generated documenting test results, conversations with the manufacturer, and information learned about
the device, such as electromagnetic compatibility. These data
serve as a baseline for future repair and PM testing and to
help answer questions posed by the clinical staff. Acceptance
testing also provides a practical training ground for the clinical engineering staff, keeping them current should emergency
clinical situations develop or a patient incident occur involving this equipment. All defects uncovered and the steps taken
to resolve them should be documented in a defect log, which
466
becomes part of the equipment history file. Until the defects

are resolved, the equipment should not be released for clinical
use. There is typically a 60-day period starting with equipment delivery in which acceptance testing is expected to be
concluded. Should there be a defect (or should only a partial
shipment of equipment be received) expenditures processing
must be notified immediately so that they can inform the vendor that the payment clock has been stopped to avoid the institution paying a penalty. Also, equipment warranty must
not start until acceptance testing is successfully concluded.
The documentation showing that the hospital did thorough
testing is invaluable to the institution during a lawsuit.
EQUIPMENT MAINTENANCE AND REPAIR
Service Options
An institution must manage its medical equipment well in
order to provide high-quality medical care at competitive
prices. For this reason, the clinical engineering department
should avail itself of the many methods of providing service
to its institution. A clinical engineering director must be flexible and fully aware of the skills of the clinical engineering
staff and the budgetary constraints of the department in order to select the proper cost-effective mix of services. The intent is always to provide quality service while striving to reduce overall costs. When outside services are selected to
supplement in-house capabilities, they must be monitored to
ensure quality, correctness of charges, and receipt of appropriate documentation for entry into the equipment history
files and computerized system (30). Several service options
are discussed below.
In-House Service
Advantages. In-house service is cost-effective, provides
very short response times (measured in minutes rather than
in hours (20), and allows for single-point (one phone call) service. Informal (not always chargeable) service requests can
sometimes be accommodated. Specialized service such as support for a cardiothoracic surgery program is also feasible (Fig.
9). Providing in-house service for complex state-of-the-art
equipment requires an adequate number of well-trained staff.
If staffing levels permit, in-house service should be substituted for service contracts whenever possible.
Considerations. Maintaining equipment in-house requires
consideration of the critical importance and downtime that
can be tolerated for each device. Consideration must also be
given to the availability of backup equipment, tools, spare
parts, test equipment, diagnostic software, and manuals, as
well as how the equipment will be repaired should it fail offhours (30). Some equipment such as ECG machines lend
themselves to in-house repair as they are not one of a kind,
parts are easily obtained, and backup units are readily
available.
Original Equipment Manufacturer (OEM) Service
Advantages. Manufacturers service has the advantage of
parts availability, servicer familiarity with the equipment,
possibility of equipment upgrade as part of the service, and
possibly remote diagnostic capability (37).
Figure 9. Cardiothoracic support, bedside monitor setup. Clinical

engineering support for cardiothoracic surgery assures a specially
trained engineer is available in the operating room throughout the
surgical procedure to check the physiological monitoring equipment
prior to patient connection as well as to troubleshoot equipment problems should they develop. The engineer is shown checking the patient
bedside monitoring setup in the Cardiothoracic Surgical Intensive
Care Unit (CTICU) to assure its functionality. He also verifies that
all patient cables are available for quick connection upon patient arrival to the CTICU following surgery.
Service Contracts and Clinical Engineering Screening. Service

contracts are available that include parts, materials, and labor. Yearly cost can be roughly estimated by taking 10% of
the equipment acquisition cost (the closer to 5% the better the
deal). Original equipment manufacturers often bundle PM
and upgrades with repair service as an enticement to select
them as a service provider. Decisions must be made as to
whether these items should be unbundled, and their value
and need for determined separately (17). Clinical engineering
screening lowers service contract cost. Screening requires
that the clinical engineering department verifies that the
equipment malfunctioned and the problem was not due to
user error, prior to a service call being requested. For easily
rectified problems the service provider may opt to supply the
parts for clinical engineering to install. Screening keeps the
clinical engineering staff familiar with a wider variety of instrumentation, allowing them to better assist during emergency situations. Screening may not be feasible for equipment
that must be up continuously and requires the service contractor to be called in immediately to reduce downtime and
minimize revenue lost to the institution. Equipment lends itself to a service contract if it is relied on heavily, only limited
downtime is acceptable, and backup equipment is not readily
available. Intra-aortic balloon pumps could fall into this category.
467
Fee-for-Service. OEM service is also available on an asneeded basis (fee-for-service). Fees include travel time (to or
from the institution), labor, parts and materials, or, using
printers as an example, a flat fee may be specified. Repair
and/or PM service can be provided. Service may be provided
either on-site (infusion pump) or at a remote depot or facility
(glucometer). A vendor-supplied repair estimate assists in determining if the repair is cost-effective. Fee-for-service may be
chosen for sophisticated repairs of equipment or when clinical
engineering staff cannot find the cause of a problem after a
reasonable troubleshooting time period has elapsed.
Third-Party Service Providers
Independent service organizations (ISO) tend to be less expensive than OEM. Service vendors should be selected based
on the quality and timeliness of past service. Service contracts and fee-for-service are available. Repair and/or PM service can be provided. It should be determined if parts other
than OEM will be used and whether the manufacturer might
void the warranty or negate product liability if a nonfactory
authorized service provider is used. The equipment has less
chance of getting factory upgrades and product recall retrofits.
Shared-Service Providers
Services can also be obtained from shared-service providers,
which can be for-profit or nonprofit. These organizations are
formed by health-care institutions usually located close together that do not have the resources necessary to maintain
an equipment management program on their own. Instead
they pool their resources and have a common entity provide
service to all of them. They share in the capital cost of setting
up such an entity and then pay for services in proportion to
their use (20). The logistical problems of providing such services must be overcome.
Some clinical engineering programs after becoming successful within their own institution expand and provide
shared services to neighboring institutions as well. As an example, Thomas Jefferson University Hospital in Philadelphia,
Pennsylvania, has a full-service in-house program as well as
a shared-service component.
Maintenance Insurance
This insurance protects against catastrophic failures by
smoothing out service cost. Service is done on an as-needed,
fee-for-service basis. The insurance company either pays the
vendor directly or reimburses the institution for the service.
Some programs pay clinical engineering personnel to handle
those repairs it wishes to in-house. Proper clinical engineering screening of service calls and good equipment management decisions can result in year-end rebates. However,
the paperwork in managing an insurance program often requires dedicating at least one full-time employee (FTE) to this
task. Maintenance insurance backup provides a reasonable
way for clinical engineering to start assuming equipment
maintenance duties in areas in which they may not as yet be
involved, such as radiology and clinical laboratories.
Preventive Maintenance (Scheduled Maintenance)
Purpose and Methodology. Scheduled maintenance ensures
that equipment previously acquired continues to function
Figure 10. Preventive maintenance cell saver. Depending on the

medical instrument, preventive maintenance can be more involved
than just a functional and safety test. Problems that are uncovered
must be rectified, and some instrumentation requires that components be replaced due to wear or number of hours of use. This, as
well as observed spills, may necessitate opening the unit. The cell
saver shown is used to salvage blood shed during an operation
allowing its return to the same patient.
properly, has not deteriorated (due to usage and aging), and

is safe for use. An attempt is made to uncover and correct
problems that have not been reported or of which the user is
unaware. Problem correction at an early stage can prevent
incidents from occurring.
Testing done during PM tends to be more functionally oriented and is not as inclusive as that done during acceptance
testing. Equipment that is mechanical in nature is tested to
ensure that moving parts are structurally sound. If electrically operated, ac safety tests are performed (Fig. 10).
Some equipment requires replacement of parts normally
expected to deteriorate with use, such as O-rings, gaskets,
and brushes. In fact, some PM (i.e., for dialysis machines and
some ventilators) is scheduled not by period (yearly, etc.), but
by number of hours of equipment operation. PM kits are obtainable from the manufacturer. Other sensitive medical
equipment (audiometer) requires extensive calibration during PM.
PM procedures specifically geared toward each instrument
are used, except when the instrument is simple enough that
a generic PM procedure can be used. Procedures can either be
written in-house or purchased commercially. A PM worksheet
keyed to the instruments unique identification number is
filled out and filed in the equipments history folder.
Most equipment is maintained on-site in the user facility
or clinical area, while others must be done in the clinical engineering laboratories. Precautions should be taken to ensure
468
that the equipment has been properly cleaned and/or sterilized before work on it is attempted. The infection control department has guidelines on cleaning prior to repair. Notation
must be made in the computerized maintenance management
system (CMMS) of equipment that is temporarily taken out
of service, its storage location, and whether PM must be done
while it is stored. The unit should be tagged indicating that
clinical engineering staff must inspect it prior to its being put
back into service. Clinical engineerings test equipment used
to maintain and calibrate the medical instrumentation must
also be periodically checked and calibrated. Certification
against standards traceable to the National Bureau of Standards may be required.
A recent trend is to use laptop computers to collect test
data on-site which are then imported into the CMMS. Computer-compatible test equipment can also be used to somewhat automate the test process. Such systems are available
from Bio-Tek and DNI Nevada Inc.
PM Risk Management. Risk factors are used to determine if
equipment requires scheduled PM, and if so, how often. This
allows health-care organizations to concentrate their resources on equipment presenting the greatest risk. All patient-care equipment is evaluated, independent of the manner
in which the institution acquired it.
During risk analysis, consideration is given to equipment
function, physical risk associated with clinical application,
and equipment maintenance requirements. A weighted numbering system is used and an appropriate threshold is set.
Clinical engineering experience (incident history and frequency of use) is used to modify the initial assessment as required (22). Some low-risk devices with no PM requirements
only require acceptance testing when first acquired, and a
zero PM frequency assigned. The following is one example of
assigning risk levels.
Equipment Function. This assessment considers how a device and its data are used and the possible consequences of
its failure. It is important whether a device is used for life
support, routine treatment, diagnosis, monitoring, or for minor functions.
Equipment function is weighted as follows (38):
Therapeutic
Life support
Surgical and intensive care
Physical therapy and treatment
10
9
8
Diagnostic
Surgical and intensive care monitoring
Additional physiological monitoring and diagnostic
7
6
Analytical
Analytical laboratory
Laboratory accessories
Computer and related
5
4
3
Miscellaneous
Patient related and other
Physical Risk. This assessment considers the possible consequences to the patient and/or operator in the event of an
equipment failure or malfunction.
Physical risk is weighted as follows:

A device malfunction could result in
Patient death
Patient or operator injury
Inappropriate therapy or misdiagnosis
Patients discomfort
No significant risk
5
4
3
2
1
Maintenance Requirements. This assessment considers

whether the device requires periodic parts replacement, recalibration, lubrication, and clinical engineering tasks necessary
to supplement user maintenance.
Maintenance is weighted as follows:
Extensive
Above average
Average
Below average
Minimal
5
4
3
2
1
Cannot Locate. Equipment that cannot be located (CNL) is

an ongoing problem that most clinical engineering departments face when attempting to do PM. Movable equipment
often winds up in locations other than those indicated in the
inventory records. The equipment may even have left the institution with the patient upon transport. This requires extensive search time, entails hospital sweeps, and if unsuccessful, notification to the user and the Safety Committee. Should
the device not turn up in a reasonable time period set by the
institution (i.e., within three PM periods, two years), the clinical owner and property control personnel should be notified
and the device removed from the active equipment inventory
list.
Reduction of PM Requirements. From 1974, when the
JCAHO first required that all electrically powered equipment
be tested four times a year, through 1988, when the JCAHO
introduced risk-based equipment management, which encouraged health-care facilities to develop more realistic equipment
management programs, there has been a decline in the requirement to do PM (39).
Today (1998), as competition between health-care institutions intensifies and resources dwindle, further PM reduction
is being discussed. At issue is the extent that PM contributes
to patient care, patient and user safety or quality. PM is labor-intensive and uses personnel resources that are in short
supply. Advances in the manufacture of medical devices (including the use of solid-state integrated circuits) produce instruments with longer mean times to failure rates and better
electrical isolation. It is questioned if PM further improves
these failure rates and whether clinical engineering resources
would be better spent in providing additional training to
equipment operators to improve their skills, thus reducing
patient incidents and enhancing patient care (39). It remains
to be seen if this approach will be adopted. For now, time
spent on PM should be limited to what is required by law or
as determined by prudent practice (17).
Repair (Unscheduled Maintenance)
Purpose and Methodology. The purpose of repair is to restore equipment so it meets original equipment manufacturer
469
Figure 11. Repair, infant warmer. Repair of

medical instrumentation requires electrical
troubleshooting skills. Here, an infant
warmer is being evaluated using a multimeter to assure that the thermostat controls
will function properly so as to prevent harm
to the infants.
specifications. This is accomplished by determining the malfunction and fixing it so as to retain the efficacy and safety of
the device (Fig. 11).
Prior to doing repair a determination should be made as to
whether the device is under warranty or service contract. If
so, the appropriate service provider should be contacted. If
under contract, clinical engineering screening may be required to verify that a problem does exist and warrants a vendor service call. Determination should be made as to whether
it is cost-effective to repair the device or if it should be retired
from service (due to lack of parts availability or expense) and
a replacement purchased.
Depending upon severity repairs can be done either in the
clinical engineering laboratories or on-site in the user facility.
In general, unless one has a good reason not to, original OEM
parts should be used. During repair, built-in diagnostics are
helpful, and the instrument operator and service manuals
from the clinical engineering technical library, as well as the
devices history file, prove invaluable.
On-site emergency support during the day allows the engineer to witness the problem first-hand. On-call/recall for
emergencies during off-hours allows instrumentation problem
troubleshooting by phone. This coupled with substitution of
spare equipment often eliminates the need for return to the
institution.
Work reports are filled out in a similar manner as for acceptance test and PM. Included should be the problem, steps
taken to resolve the problem, parts used, and pertinent test
data. These reports keyed to the instruments unique identification number are filed in the instruments history folder,
and suitable computerized data entry is made.
Parts and Service Manuals. A problem faced when repairing
medical devices is that manufacturers are sometimes unwill-
ing to provide necessary replacement parts, insisting that

more costly field replaceable units be purchased instead. They
argue that proper repair requires automatic test equipment
verification that is only available at the factory, and that they
will be liable should the device malfunction after such repair
(40). Also, although required under the FDA Federal Medical
Device Amendments of 1976 to provide service literature containing installation, operation, and maintenance information,
some manufacturers are unwilling to provide service manuals
with proper schematics, arguing that their technology is proprietary, or they require a nondisclosure document be signed.
Prior to signing any such agreement it is best to check with
the institutions legal counsel to determine the ramifications
of doing so. It must be stressed that the time to resolve the
manuals issue is during pre-purchase requisition review. Not
only is repair impacted by not having operator and service
manuals, but equipment acceptance testing is as well, as
equipment specifications and detailed test information is to
be found there. Some clinical engineering departments do not
even schedule acceptance test unless these manuals are first
received.
Equipment Retirement
Equipment is retired from service when it:
Is obsolete and can no longer be cost-effectively repaired
due to lack of parts or expense.
Becomes unreliable and prone to constant failure.
Poses a hazard to patient or user.
Is replaced by newer technologies.
Is no longer the choice of the clinical staff (3).
470
Depending upon the institution and the equipment condition

it could be:
Cannibalized for parts.
Traded in for a newer device.
Sold.
Offered to a sister institution.
Donated to a school.
In any case, appropriate disposal procedures must be followed, including notifying property control and making entries into the CMMS and equipment history files.
MANAGEMENT
Health-Care Technology Management
Health-care technology includes all of the components needed
to diagnose and treat human disease (illness). This includes
medical equipment (devices, systems, and software), supplies,
pharmaceuticals, biotechnologies, and medical and surgical
procedures as well as the health-care facilities (hospitals, etc.)
that are used to house the patients and medical equipment
(23). Health-care technology management deals with all of
the health-care technology components. Included in healthcare technology management is equipment management,
technology planning, and technology assessment.
Equipment Management
Definition and Purpose. Equipment management deals specifically with the medical equipment and is the cornerstone of
an effective clinical engineering program. An equipment management program that encompasses the more traditional clinical engineering duties (23) controls the risks associated with
using medical equipment for patient care by detecting and
correcting hazards before injuries can occur. Proper equipment maintenance (a core component of equipment management) maximizes the useful life of medical equipment and
minimizes its lifetime cost.
Although equipment management includes the core services of equipment maintenance (PM and repair), it goes beyond these services to include most phases of equipment lifespan. It is involved with equipment acquisition, which
includes equipment assessment, equipment specification,
RFQ generation, and vendor selection. It encompasses installation planning, acceptance testing, user in-service education,
selection of service provider, product recalls and alerts, and
incident investigation, as well as PM and repair. It also includes equipment replacement analysis, removal, and salvage (30).
Better equipment management decisions save money.
They reduce equipment downtime, eliminate the need to pay
for emergency equipment rentals, and allow better equipment
replacement decisions to be made that consider past equipment failures and expenses in addition to equipment capabilities. Equipment and consumable standardization is also encouraged (3).
An equipment management program must be compliant
with JCAHO guidelines and regulatory requirements such as
those of the Department of Health (DOH), College of Ameri-
can Pathologists (CAP), and American Association of Blood

Banks (AABB).
Centralization. Ideally, a centralized equipment management program capable of handling all patient-care equipment
(including anesthesiology, clinical laboratories, and respiratory care) is desired. More sophisticated institutions are placing all medical equipment maintenance funding in the clinical
engineering budget. Consolidation makes economic sense as it
allows an institution to more easily track its true equipment
maintenance costs. Cost-effective service contracts can be negotiated and unnecessary ones eliminated. Interface between
the institution and equipment vendors and manufacturers becomes simpler (30). Record-keeping is standardized, demonstrating to JCAHO that a uniform level of service for all medical devices exists as well as a centralized data repository.
Centralization also makes it clearer to an institutions staff
that the clinical engineering department should be consulted
for all equipment-related services.
AAMI Recommended Practices. AAMI is presently finalizing
the recommended practice Required characteristics for a medical equipment management program (41), that reflects existing good practice, addresses program structure, required
documentation generation, staffing, and resource allocation.
It acknowledges that many clinical engineering programs already exceed these minimum requirements. AAMIs goal is
that others strive to exceed them as well. This document
should help newly formed, smaller clinical engineering departments (and health-care facility managers) to define what
is expected as a starting point better.
Technology Planning
Technology planning supplements an equipment management program. It helps to further ensure that appropriate
equipment that is cost-effective, efficacious, and safe is available, allowing the institution to meet quality patient-care demands. Technology planning includes greater clinical engineering involvement in the determination of equipment
replacement needs, equipment acquisition, facility planning
and design (to ensure equipment needs are accommodated),
as well as continuous technology assessment (23).
Technology planning allows an institution to remain competitive by choosing new technologies that complement existing hospital services and present strategic advantage to the
institution against competitors. It also reduces liability risk
for the hospital and clinical staff by helping to identify legal
standards of care requirements requiring purchase of new or
additional equipment. As example, the requirement that
pulse oximetry (SAO2) monitoring, end-tidal CO2 monitoring,
or both be provided during general anesthesia (3).
Technology Assessment
Technology assessment (a component of technology planning)
analyzes all of the consequences of introducing a new technology [i.e., bone marrow transplant or picture archiving and
communications systems (PACS)] into a health-care institution, prior to any equipment being purchased. Analysis includes consideration of the equipment required, techniques to
be used, FTE personnel requirements, size and makeup of patient base, community impact, and financial considerations.
471
Figure 12. Computerized Maintenance Management System. Computerized Maintenance

Management Systems (CMMS) are used by
clinical engineers to track their workload.
This includes repairs, preventive maintenance scheduling, and other requests for engineering assistance. CMMS also maintain an
equipment inventory and maintenance history information. The two clinical engineering
supervisors are querying the system to obtain
lists of equipment of specific clinical areas so
that they can plan and assign workloads for
the staff.
The goal of the analysis is to reduce the possibility of purchasing expensive and inappropriate equipment that cannot generate income for the institution. By assisting in this task, the
clinical engineering staff s awareness of new and emerging
technologies aids in more wisely allocating capital resources
(3).
Computerized Maintenance Management Systems
Computerized maintenance management systems (CMMS)
software is a powerful technology management tool used to
collect, store, and analyze data (Fig. 12). To utilize such programs, some clinical engineering departments rely on their
institutions mainframe computer or a file server and local
area network maintained by the information services department, while others maintain their own file server and local
area network taking on the responsibility for data backup and
integrity. Some use internally developed CMMS software,
while others purchase commercially available packages.
CMMS systems generate reports relating to all aspects of the
operation of a clinical engineering department, including
equipment management. Technology assessment software is
also available.
Functions of CMMS are as follows (30):
Maintain an equipment inventory and nomenclature
system
Select and schedule PM (based on risk factors)
Track work including repair, user in-service training, construction and research projects
Prioritize work load
Track equipment and vendor services provided under warranty or service contract
Track loaner and leased equipment
Identify medical equipment needing replacement due to

constant breakdown and large downtime
Select new equipment for purchase based on past performance and cost effectiveness of similar equipment
Detect trends that pinpoint the need for additional user
training due to operator error, NFF, equipment damage,
or abuse
Detect trends that pinpoint the need for additional service
training such as repeat repairs
Generate work-completed reports for customers, including
lists of equipment that cannot be located (CNL)
Bill customers
Maintain parts inventories
Generate reports for hospital administration
Analyze clinical engineering performance (financial, quality, and productivity)
Assist in long-range forecasting
Benchmarking
Purpose and Methodology. A management tool for continuous quality improvement, benchmarking allows comparison of
an organization to other similar organizations so that better
techniques can be ascertained and adopted to improve performance and customer service.
Benchmarking can be informal or formal. Informal benchmarking consists of gathering information about similar institutions by contacting colleagues and comparing clinical engineering parameters and functions. Formal benchmarking
requires filling in a detailed questionnaire and submitting it
to an outside organization for analysis. Such analysis is provided as a service to university hospitals throughout the country that are members of the University Healthcare Consor-
472
tium (42). This benchmarking process allows member

institutions to compare their operations. Institutions that
submit data have agreed that all other member institutions
may gain access to it for purposes of improvement.
Caution in Use. Benchmarking can be helpful (43) and the
possibility of improvement exists. However, the results must
be properly used. It is important when comparing institutions
not only to consider the number of FTE, which may not be an
absolute number as it may be based on overtime or be normalized (i.e., to a 40 h week), but also to compare fully the
services each clinical engineering department provides. This
is critical if such data are used to make operational decisions
that relate to staff size. Analysis may well show that those
chosen as better performers do not provide services that are
vital to another institution. It is also important to consider
the acquisition cost of equipment being maintained. Some departments are required to handle more sophisticated equipment that is both time consuming and costly to repair, thus
increasing FTE requirements.
Finally, some point out that comparison of in-house clinical
engineering departments to each other, using the ratio of
number of engineers per bed, is a wasted effort as their real
competitor is not each other, but outside service providers,
who could replace them both (17).
Clinical Engineering Oversight
Oversight Function. Clinical engineering provides an important oversight function whose primary purpose is to reduce risk of injury to patients and staff as well as liability to
the institution for which they work. While carrying out their
mandate duties of the clinical engineering department can
sometimes cause client irritation. As an example, review of
equipment purchase requisitions, questioning clinical users
about preferred equipment choices, coupled with the bid process, and need to provide justification to purchase other than
the low bidders equipment lengthen the equipment ordering
process. On occasion, clinical engineering management may
even disallow the purchase or use of devices found to be unsafe due to extremely poor workmanship, inherent design defects, or dangerous failure modes. Acceptance testing, which
disallows immediate clinical usage of equipment, tends to
frustrate clinicians anxious to use the equipment and causes
them to question the need for acceptance testing (only wanting electrical safety tests to be run). They sometimes prematurely schedule patients, in-service education, and removal of
old equipment upon equipment arrival into the institution,
leaving little if any time for testing. These issues make it critical that clinical engineers understand that as they strive to
satisfy regulatory requirements, safety requirements, and
ethical considerations, they must also strive to streamline
their operation to provide services as quickly as possible, so
as not to alienate their clients. It is important to demonstrate
to clients through timely feedback and education that clinical
engineering is a vital resource and not an impediment.
Dangers in Bypassing Oversight Function. There are times
even with hospital policies in place, that clinical engineering
is not consulted for what is felt to be expediency. The result
is often increased cost, delays, and increased risk to the institution.
Not involving clinical engineering during the prepurchase

evaluation process is counterproductive. This is particularly
true while planning for new clinical areas or renovating existing ones. Having to open newly constructed walls to add
electrical outlets or to provide additional reinforcement to
allow required medical equipment to be mounted, all of which
should have been planned for prior to construction, is both
costly and time-consuming.
Problems also arise when equipment is ordered, and physical plant conditions are not suitable for its operation. For example, a type of endoscopic sterilizer requires clean water,
and a specified minimum water pressure (often not found in
certain urban areas especially during summer months). Without these environmental conditions being met, the unit will
not function and additional external filtration (which quickly
clogs) does not remedy the situation. Studying specifications
and having clinical engineering personnel interact with facilities engineering staff prior to such equipment being ordered
reduces the chance of such expensive nonfunctioning installations.
Clinical engineering involvement in purchase requisition
review reduces the possibility that incorrect equipment is ordered. Ordering nonstandard equipment requires additional
user training and increases the chance of patient injury due
to user equipment unfamiliarity. This also compounds maintenance requirements and the types of accessory supplies required.
Clinical engineering acceptance testing prevents unsafe
equipment with defects being put into service and eliminates
premature payment authorization that results in loss of leverage in defect problem resolution. Also, this assures the equipment is entered into the equipment inventory, PM scheduling
is done, eliminating problems during regulatory inspections,
such as untested devices turning up just prior to inspection
when limited time is available to process them. To minimize
problems it is important that clinical engineering work closely
with the purchasing and expenditures processing departments to assure that medical equipment falling within the
mandate of the clinical engineering department is not ordered
or paid for without clinical engineering approval.
CLINICAL ENGINEERING CHALLENGES
Competition
Clinical engineering departments face the challenge of improving productivity and quality even though their resources
are shrinking. The threat of downsizing, out-sourcing, and insourcing is very real. Competition with ISOs of the field service industry has intensified over the last 10 years, requiring
in-house clinical engineering programs to organize themselves as businesses, so as to reduce costs and to be competitive (17).
Justification
Clinical engineering departments must continually demonstrate their value to their institution. This is especially true
as hospital staff changes. New people, both administrative
and clinical, come on board who may be unfamiliar with the
benefits of clinical engineering services. Continuous advertising (brochures, newsletters, Web sites) and attendance at
meetings is required to educate them as to clinical engineerings vital role within the institution.
Remote Service
As health-care institutions merge and collaborate, and as additional satellite clinics are established to provide hospitals
with clients, clinical engineering is faced with the logistics of
providing services to remote locations.
Regulation
The FDA is considering extending good manufacturing practices (GMP) rules and regulations to medical equipment refurbishers, reconditioners, and servicers. This would require
them to meet requirements similar to those of original medical device manufacturers and remanufacturers. Clinical engineering departments as equipment servicers may be impacted (44).
Home Safety
Clinical engineers may have to become more involved in
safety issues related to the increased use of medical equipment for home care and how to provide such services. For
example, home dialysis requires preliminary inspection of the
patients home site to ensure adequate electricity and water
and then periodic visits for PM and repair.
Year 2000 (Y2K) Compliance Healthcare devices and systems (information systems, medical equipment, and general
hospital systems) that use software or contain microprocessors may be prone to the Year 2000 problem. If so, as the date
changes from Dec. 31, 1999 to Jan. 1, 2000 they may incorrectly represent the year 2000 as 1900 (or some other date).
Some equipment might operate erroneously, others not at all.
Such failure could affect patient safety, produce incorrect
printouts and archiving, and increase risk to the institution.
Clinical engineering involvement and allocation of resources
are required to ensure equipment compliance (45).
Seeking New Opportunities
Clinical engineering departments must be flexible, adapting
to the times and conditions of the ever-changing health-care
institutions they serve. The feasibility of providing additional
services for X-ray and ionizing radiation equipment, computers, computer networks, patient information systems, telecommunications, and nurse call systems should be investigated. Although with proper training, clinical engineering
staff should be able to repair these items just as they repair
other sophisticated equipment falling within their domain, a
realistic approach must be taken with consideration given to
available resources (i.e., funds for training and FTE allocation), as well as the political realities of turf within their
particular institutions. Clinical engineers should also strive
to become more involved in technology assessment issues for
new technologies including telemedicine, robotics, PACS, and
wireless LAN, helping to determine the value of introducing
them into their institution.
Such flexibility will ensure that the relatively new profession of clinical engineering will mature and continue to provide value to the institutions it serves, as it moves forward
into the 21st century.
473
ACKNOWLEDGMENTS
Photographs were taken by Ernest Cuni, Biomedical Communication, SUNY HSCB, 1998.
I wish to thank all members of SMICs staff, both past and
present, with whom I have worked over the past 20 years. It
is through our daily interaction that the concepts presented
have been better defined. In particular, thanks to John Czap,
Luis Cornejo, Leonard Klebanov, and Marcia Wilkow. Also,
M. K. Venugopal who understood and valued SMICs services.
Thanks also to Barbara Donohue and Kelly Galanopoulos who
offered specific suggestions to enrich the content.
BIBLIOGRAPHY
1. Phone conversation between Dr. Caceres and the author, November 1997.
2. Anonymous, Whats a Clinical Engineer?, Pamphlet, Houston,
TX: ACCE.
3. Anonymous, Special report on technology management: preparing
your hospital for the 1990s, Health Technol. 3 (1): Plymouth Meeting, PA: ECRI, Spring 1989.
4. M. J. Shaffer, The reengineering of clinical engineering, Biomed.
Instrum. Technol., 31 (2): 177178, 1997.
5. T. J. Bauld, The definition of a clinical engineer, J. Clin. Eng., 16
(5): 403405, 1991.
6. Anonymous, Frequently asked questions: biomedical engineering,
Department of Biomedical Engineering, Tulane University,
http://www.bmen.tulane.edu/BMEFAQ/Welcome.html, 5/23/96.
7. S. Thanasas and Y. Dathatri, Course description of the Biomedical Engineering Technology Associates in Applied Science Degree, Farmingdale, NY: SUNY Farmingdale, January 1997.
8. Anonymous, Certification Program for Clinical Engineers, Virginia: International Certification Commission for Clinical Engineering and Biomedical Technology, February 1996.
9. S. Ben-Zvi and W. Gottlieb, Inspection and maintenance of medical instrumentation. In B. Feinberg (ed.), Handbook of Clinical
Engineering, Boca Raton, FL: CRC Press, 1980, Vol. 1, pp.5778.
10. W. J. Curran, P. E. Stanley, and D. F. Phillips, Electrical Safety
and Hazards in Hospitals, New York: MSS Information Corporation, 1974.
11. J. M. R. Bruner and P. F. Leonard, Electricity, Safety and the
Patient. Chicago: Year Book Medical Publishers, 1989.
12. D. Roth, The History of Medical Equipment Service, Part I, 24x7,
2 (4): 24-27, 34, April 1997.
13. D. Roth, The History of Medical Equipment Service, Part II, 24x7,
2 (5): 32, 3435, May 1997.
14. American National Standard, Safe current limits for electromedical apparatus, ANSI/AAMI1993. Arlington, VA: AAMI, December 2, 1993.
15. B. Klein (ed.), Health Care Facilities Handbook. 4th ed., Quincy,
MA: NFPA, 1993.
16. E. B. Fein, N.Y.U. Center and Mt. Sinai resume talks, The New
York Times, B1, B6, September 24, 1997.
17. G. Gordon, Breakthrough Management, A New Model for Hospital
Technical Services. Arlington, VA: AAMI, 1995.
18. G. F. Nighswonger, 1997 survey of salaries and responsibilities
for hospital biomedical/clinical engineering and technology personnel, J. Clin. Eng., 22 (4): 214232, 1997.
19. M. Wilkow and I. Soller, The Scientific and Medical Instrumentation Center Brochure. Brooklyn, NY: SUNY Health Science Center at Brooklyn, University Hospital of Brooklyn, 1995.
474
CLOCK DISTRIBUTION IN SYNCHRONOUS SYSTEMS
20. Guideline for establishing and administering medical instrumentation maintenance programs, AAMI MIM3/84, Vol. 1: Biomedical Equipment, AAMI Standards and Recommended Practices.
Arlington, VA: AAMI, 1989, pp. 342.
21. L. C. Brush et al., The Guide to Biomedical Standards. 20th ed.,
Brea, CA: Quest Publishing Company, 1995/1996.
22. Inspections and enforcement chapter 1510 JCAHO accreditation,
BNAs Health Care Facilities Guide, No. 36, 1500:1001
1500:1106, The Bureau of National Affairs, 1997.
23. Y. David and T. M. Judd, Medical Technology Management, Redmond, WA: Space Labs Medical, 1993.
24. O. R. Keil, Accreditation and clinical enginering, J. Clin. Eng., 21
(6): 410, 412, 440, November/December, 1996.
25. M. Shepherd, SMDA 90: User facility requirements of the final
medical device reporting regulation, J. Clin. Eng., 21 (2): 114
118, March/April, 1996.
26. Safe Medical Devices Act: Medical Device Tracking Regulations.
Plymouth Meeting, PA: ECRI Advisory, July 1997.
27. L. A. Geddes and L. E. Baker, Principles of Applied Biomedical
Instrumentation. 3rd ed., New York: Wiley, 1989.
28. Health Devices Sourcebook Medical Product Purchasing Directory
with Official Universal Medical Device Nomenclature System.
Plymouth Meeting, PA: ECRI, 1997.
29. Medical Device Register, The Official Directory of Medical Suppliers. Montvale, NJ: Medical Economics Data Production Company, 1994.
30. T. Cohen, Computerized Maintenance Management Systems for
Clinical Engineering. Arlington, VA: AAMI, 1994.
31. Conversation between C. Yeaton (SUNY University Hospital of
Brooklyn, Risk Manager) and the author, October 1997.
32. Digital Safety Analyzer Model 501 Operators Manual, Rev. C.
Burlington, VT: BIO-TEK Instruments, 1993.
33. S. A. Rubin, The Principles of Biomedical Instrumentation A Beginners Guide. Chicago: Year Book Medical Publishers, 1987.
34. R. Aston, Principles of Biomedical Instrumentation and Measurement. Columbus, OH: Merrill Publishing Company, 1990.
35. I. Soller and L. Klebanov, SMIC checklist for the purchase of scientific and medical instrumentation. Brooklyn, NY: SUNY Health
Science Center at Brooklyn, University Hospital of Brooklyn,
September, 1992.
36. S. Polaniecki, SMIC guide for the performance of initial checkout.
Brooklyn, NY: SUNY Health Science Center at Brooklyn, University Hospital of Brooklyn, August, 1989.
37. Siemens-BJC multivendor service breaks new ground with remote
diagnostics, 24x7, 2 (9): September 7, 1997.
38. L. Fennigkoh and B. Smith, Clinical Equipment Management.
Plant Technology and Safety Management Series, implementing
the 1989 PTSM standards: case studies, Number 2, 1989, The
Joint Commission.
39. O. R. Keil, Is preventive maintenance still a core element of clinical engineering?, Biomed. Instrum. Technol., 31 (4): 408409,
1997.
40. K. Beckley, SpaceLabs explains service stance, 24x7, 2 (11): November 7, 1997.
41. Recommended practices for a medical equipment management
program, final draft standard, AAMI EQ56-1997, June 1997, Arlington, VA: AAMI.
42. Clinical Benchmarking, Oak Brook, IL: University Healthcare
Consortium, 1993.
43. M. S. Gordon, The long and short of benchmarking, Biomed.
Technol., 6 (3): 2026, May/June, 1995.
44. FDA eyes GMP regs for refurbishers and servicers, 24x7, 2 (9):
September 7, 1977.
45. Medical devices and the year 2000 problem, Health Devices Alerts
Number 1998-F2, Plymouth Meeting, PA: ECRI, February 27,
1998.
IRA SOLLER
State University of New York
Health Science Center at
Brooklyn
CLINICAL MONITORING. See PATIENT MONITORING.
366
NEUROTECHNOLOGY
THE MOTOR SYSTEM: NERVE REGENERATION

AND NEURAL PROSTHETICS
Lesions in the peripheral nervous system in humans can lead
to several disabling effects in sensory and motor functions because the primary information carrier, the propagating action
potential, can no longer travel from sensory organs to the
brain (afferent information, sensory nerve fibers) or from the
brain to muscles (efferent information, motoneurons). In
many cases, peripheral nerves may repair themselves (regeneration), provided that the source of the lesion (for example, pressure on the nerve) is removed soon enough or that
adequate surgical measures are taken in due time in order to
bring nerve stumps together or to transplant nerve sections
to bridge a large gap. During the healing process, nerve fibers
will first degenerate and then regenerate all the way, from
the spinal cord toward the periphery, reusing the old channels of myelin sheaths and connective tissue. The nerve regenerates with a typical speed of 1 mm per day.
However, this ability to regenerate more or less autonomously is a property of peripheral nerves only. The central
nerve fibers of the spinal cord cannot be induced to regenerate, although extensive research tries to bring this about by
manipulating the biochemical environment of the fibers, offering proteins such as neural growth factors or semaphor proteins and other agents that may stimulate nerve growth.
If a person has a central neural lesion but no harm to the
peripheral nervesfor example, in paraplegic individuals
(with neural interruptions in the spinal cord)the peripheral
nerves may be stimulated artificially by short electric pulses,
which evoke propagating action potentials toward the paralyzed muscles and restore force.
Crude restoration of basic motor function has been
achieved in laboratory settings using surface electrodes or implanted wires, to control on the order of ten muscles, in a
NEUROTECHNOLOGY
more or less onoff way of operation, which causes fast fatiguing of the muscle. More complicated everyday functions will
require independent control of a large number of nerve fibers/
fascicles/muscle units, which allows finely tuned motion and
does not cause fatigue. Besides highly developed, multisite
contacting technology, sophisticated closed-loop control is necessary for those functions, as well as the help of mechanical
and other nonelectrical prosthetic aids. Research on all aspects is in full swing but will take many years to reach the
clinical application level.
z
x
y
( , ,z)
o
r , z
Nonmotor Systems
Artificial electric stimulation is used to stimulate the auditory

nerve in cases of profound hair cell damage in the cochlea.
This application is widespread clinically. Other applications
are bladder stimulation of the nerves of the urinary system,
diaphragm pacing, cardiac pacing. In these cases, the number
of electrodes is only one or relatively modest.
367
Current source
Figure 2. The volume conduction model of the nerve and its surroundings. Longitudinal and radial conductivity inside the fascicle
are z and r, respectively. Perineural sheath conductivity is s, epineural conductivity o, and extraneural conductivity e.
MODELING OF ELECTRICAL STIMULATION OF FIBERS

IN PERIPHERAL NERVE
Peripheral nerve consists of (up to thousands of) nerve fibers,
or axons, with diameters ranging from a few to tens of micrometers. Nerves may contain subbundles, called fascicles,
with a typical diameter of 0.5 mm. Motor fibers have a myelin
sheath wrapped around them, to speed propagation of the action potential. At regular intervals the myelin sheath is interrupted over a few micrometers, at the so-called nodes of
Ranvier. These are the sites where membrane channels exchange ions into and out of the membrane, to keep the action
potential traveling. The ratio of internode distance to fiber
diameter is approximately 100 : 1.
A negative-going extracellular current pulse close to a node
may trigger the action potential artificially. This is the basis
of artificial electrical stimulation.
Modeling is usually done in two stages, with a nerve fiber
excitation model and a volume conductor model.
The Nerve Fiber
First, the response of a nerve fiber to an electrical field is
modeled (1,2). For this, the approximate activating function
may be used, in which a fiber is considered over a length of
three nodes only, modeled by two sections of a passive RC
network (Fig. 1). The nerve becomes active when the second-
order difference f of external node potentials Ve of a central

node and its two neighbors exceeds a threshold (about 20
mV). As the exact node positions are unknown and f for a
given diameter class of fibers only depends on the internode
distance , activating functions are calculated for each position x,y,z and x,y,z in the fascicle, for each electrode. Thus
f = Ve,n1 2Ve,n + Ve,n+1
= Ve (x, y, z ) 2Ve (x, y, z) + Ve (x, y, z + )
(1)
If an electrode is sufficiently close to a node of Ranvier, compared to , the two terms Ve,n1 and Ve,n1 may be set to zero.
This is the local approach.
The activating function sets the external potential condition but does not take into account ionic currents through the
membrane ion channels, which can be modeled by the famous
HodgkinHuxley equations and their refined forms. Because
of this, the activating function approach is only valid for short
rectangular stimulus current pulses, in the range of 10 s
to 100 s duration. Also, the well-known relationship at the
threshold of stimulation between amplitude and duration of
the stimulus (strength-duration threshold curve) is not contained in the activating function.
The effect of pulse duration has been taken into account
recently by Warman et al. (3). Nagarajan and Durand (4),
Grill and Mortimer (5), and others. It was demonstrated that
it may be a tool to influence spatial selectivity of stimulation.
The metal electrode itself, with its interface to the fluid
environment (Helmholtz layer, Warburg impedance, Faradaic
current), is not dealt with here but is an important part of
the stimulation system.
;;;;;;;;;;;;
Cell membrane
Myelin-sheath
Ve,n
Ve,n1
Rm
Cm
Vr
Vi,n1
Node of Ranvier
Axon
Ve,n+1
Rm
Cm
Vr
Ri
Vi,n
Rm
Cm
Vr
Ri
Vi,n+1
Figure 1. The electric network equivalent of a myelinated fiber. Vr

is the membrane rest potential. Ve,n is the extracellular potential at
node n. Vi,n is the intracellular potential at node n. Ri is the intracellular resistance. Cm and Rm are membrane capacitance and resistance.
The Volume Conductor

Second, the potentials Ve,n, generated by currents from stimulating electrode configurations, must be calculated at the node
positions of all fibers and represented as equipotential contours, or equiactivation function contours (6).
Figure 2 shows the volume conductor model of a cylindrical
nerve or fascicle. The fascicle is idealized as an electrically
homogeneous and infinitely long extending cylinder with a ra-
NEUROTECHNOLOGY
dial conductivity r and a longitudinal conductivity z. The

cylinder is surrounded by a layer that represents the thin
perineurium, with a sheath conductivity s. The next layer is
the perineurium, with conductivity o. At the outside of the
fascicle the medium is infinitely homogeneous and isotropic
with conductivity e.
Stimulation electrodes are idealized as point current
sources and may be positioned anywhere in the fascicle. Using the cylinder symmetry, an analytical expression for the
potentials can be derived. The potential Ve for an electrode at
(r,0,0)injecting current Iconsists of the sum of a source
term Vse
Ves (x, y, z) =
4 r z
I
(x r)2 + y2 + z2 r /z
(2)
and a boundary term Veb, which is an expansion of Bessel functions. Similarly, Vse(x,y,z ) follow from (Eq. 2).
Electrode configurations may be monopolar, bipolar, tripolar, and so on. Combinations of anodes and cathodes may
yield some field-steering capability, although at the expense
of higher stimulus currents (6,7).
While the cylindrical idealization of the nerve or fascicle
permits the analytical solution of Laplaces equation, as summarized previously, the more general case of a nerve volume
conductor with many irregular, inhomogeneous, anisotropic
fascicular cross sections inside asks for finite-difference modeling of the tissue (8,9).
SELECTIVITY OF STIMULATION AND EFFICIENCY OF A
STIMULATION DEVICE
At low current, an electrode can stimulate one fiber if its position is close to that fiber, compared to other fibers. Increase
of current will expand the stimulation volume, thus including
more and more fibers.
The ultimate selectivity would be reached if each fiber
would have its own electrode. This would require, however
both a blueprint of positions of fibers in the nerve so that
electrodes could be positioned close to a node of Ranvier, and
enough electrodes. In practice, no blueprint is available, and
microfabrication has technological limits. Therefore, with a
limited number of electrodes, placed optimally (in a statistical
sense), it is important to consider and test how selective stimulation can be.
In this respect one has to measure the extent to which each
electrode controls as few fibers as possible at low current, before potential fields start to overlap with those of other electrodes, with increase of current. Greater overlap means
lower selectivity.
From another point of view, one might define the efficiency
of a multielectrode device: the number of distinct fibers that
can be contacted, divided by the total number of electrodes.
Greater overlap means reduced efficiency.
Fiber selectivity has been addressed in Rutten et al. (10),
among others. It was concluded, on statistical grounds and by
overlap experiments, that an electrode separation of 128 m
was optimal for a rat peroneal nerve fascicle with 350 alpha
motor fibers.
Limited force recruitment experiments with a 2 D 24-electrode array (electrode separation 120 m) (11) yielded that 10
distinct threshold forces could be evoked (efficiency is 10/24

42%).
PERIPHERAL NERVE FIBER RECORDING:
MODELING AND SELECTIVITY
The forward control of muscle by artificial stimulation might
gain importance when this control is supplemented by selective feedback information from nerve fibers attached to sensors such as muscle spindles, tendon organs, and cutaneous
sensors. This asks for insight into selective recording with
multielectrodes.
The same type of calculation previously made for the case
of selective stimulation of nerve fibers in rat peroneal nerve
(isotropic conductor, local approach) (10) could be applied, by
reciprocity, to the case where the device is used to sense natural activity from afferent fibers. These calculations would, for
example, lead to a (statistically optimal) electrode interdistance of 143 m, for the case that there are 250 type I afferent fibers in rat peroneal nerve.
However, while an action potential can be triggered by activation of one node of Ranvier only (stimulation), propagation of an action potential requires about 20 active nodes (recording). So it is not allowed to replace the electrode
(stimulation) by one node of Ranvier (recording).
Another difference is that nerve fibers will almost always
fire as ensembles. Regarding selectivity, when two (not overlapping in time) action potentials (or ap trains) are sensed by
one electrode, the trains can be detected separately when the
selectivity ratio S of their amplitudes V1 and V2 exceeds a
certain threshold (i.e., when S Sth; for example, S 1.1, or
S 2) (compare this to the signal-to-noise ratio; 1.1 means
barely visible, 2 is better).
Quantitative insight in this selectivity ratio S as a function
of spatial and conductivity parameters may be obtained by
the combined use of an electrode lead field model (using the
volume conduction model as outlined previously) and a probability model for the positions of active fibers (12). Figure 3
shows a dramatic decrease in the ability to discriminate two
trains when the nerve is insulated from its surrounding tis-
1.00
0.80
Probability
368
S = 1.1
0.60
0.40
0.20
0.00
0.00
S = 1.5
S=2
0.20
0.40
0.60
0.80
1.00
Conductivity of the extraneural tissue, 1/m
Figure 3. The probability P that the measured action potentials from

the two fibers, which are nearest to a central monopolar electrode,
have an amplitude ratio S Sthreshold for three thresholds 1.1, 1.5, or
2, as a function of the conductivity of the extraneural tissue. The
nerve has 40 active fibers (20 nodes each). (From Ref. 12.)
NEUROTECHNOLOGY
369
sue (i.e., for zero extraneural conductivity), illustrating the

importance of a natural wet surrounding of the nerve.
Glass cap
MICROFABRICATED LINEAR, 2-D,

AND 3-D MULTIELECTRODES
Probes
Spacers
Silicon and Silicon-Glass Arrays

Silicon-based microprobe fabrication has been a major and
outstanding activity of the Center for Integrated Sensors and
Circuits at the University of Michigan and has led to a large
number of single-shaft, multishaft, and 3-D stacked microelectrode arrays, a number of these being supplied with onboard microelectronics (1322). Fabrication was supported by
design studies (23), strength characterization (24), and development of interconnection technology (25,26). Groups in Utah
and Twente tried to fabricate brush or needle-bed 2-D/3-D
multielectrodes in silicon or silicon/glass technology, for cortical and nerve applications, with about 100 electrodes. As anisotropic silicon etching cannot (yet) perform up to the aspect
ratios needed for long, slim needles (a 20 m diameter, 500
m long needle has an aspect ratio of 25); the first step to
obtain a brush structure from a solid piece of silicon is a sawing procedure (12,27,28).
Silicon/glass technology has the advantage of high aspect
ratios, sufficient lengths of needles, and different lengths of
needles in the same device. The disadvantages are the 3-D
nature of many of the process steps, the large number of
steps, and the difficulty of their integration (12).
The 3-D cortical multielectrode array, using microassemblies of 2-D planar probes, of the Michigan group (20) is a
good example of a hybrid fabrication solution: stacking of
multishaft/multisite flat devices, combining many advantages.
Silicon-LIGA Arrays
An alternative, batch-oriented, and larger-scale way to fabricate multielectrode needle-shaped devices is to combine silicon technology with the LIGA technique (Lithographie, Galvano Abformung) (29). Briefly, in the silicon/LIGA process
nickel needles are grown from a combined seed/interconnection layer through narrow channels in 200 m PMMA (polymethylmethacrylate). After removal of PMMA and etching of
the seed layer, the electrode needles stand completely electrically separated and are connected individually to the leads in
the interconnection layer.
In this way, Bielen succeeded at the IMM (Institute fur
Microtechnologie in Mainz, Germany) in fabricating a 2-D
multielectrode of 4 32 needle electrodes, with square as
well as round columns or needles. The electrodes have a
thickness as low as 15 m and an ultimate height of 220 m
(11).
Silicon/LIGA technology reduces the number of steps but
has as a disadvantage the need for synchrotron radiation
facilities. Also, the present limit of the electroplating process
to 220 m long nickel needles has to be extended to a needle
length of about 500 m for useful neuroprosthetic and cortical applications.
A review of electrode technology and its perspectives can
be found in Mortimer et al. (30). An interesting, nonsilicon
approach to contact fibers intrafascicularly is the use of teth-
Digital
Signal
Processing
Silicon platform
Probe array
(a)
Figure 4. (a) Overall diagram of a surface-mounted 3-D recording

array. Several multishank 2-D probes are inserted through the platform and held in place with micromachined spacer bars. (From Ref.
20, their Fig. 1.) (b) Scanning electron microscope (SEM) photographs
of a 3-D 4 4-shank microelectrode array. The shanks on the same
probe are spaced on 150 m centers and are 40 m wide. The probes
are 120 m apart in the platform. (From Ref. 20, their Fig. 2, bottom.)
Figure 5. SEM photograph of silicon-nickel-LIGA array. Array with

150 m tall, 20 m diameter nickel needles, realized with aligned Xray lithography and galvanic growing (LIGA) on silicon substrate
with 8 m Cu interconnection wiring. Interdistance between columns
is 120 m. (From Ref. 11.)
370
NEUROTECHNOLOGY
Distal nerve stump

Guidance channel
Proximal stump
Distal stump
Microelectrode array (die)

Regenerated axons
Proximal nerve stump
Silicone rubber tube

(a)
Silicon chip
(b)
Figure 6. (a) Schematic representation of an intelligent neural interface implanted into an intersected nerve. (From Ref. 43, their Fig. 1.) (b) Schematic drawing of the silicone chamber model
with the inserted silicon chip bridging a 4 mm gap between the proximal and distal stumps of a
transected rat sciatic nerve (From Ref. 42, their Fig. 3.) (c) SEM photograph view of a fabricated
chip with 100 m diameter holes. (From Ref. 42, their Fig. 2.) (d) SEM photograph of nerve
tissue sections distal to a chip with hole diameters of 100 m after 16 weeks of regeneration.
Shown is a minifascicular pattern on the distal surface of the chip. The regenerated nerve structure has a smaller diameter than that of the perforated area of the chip. The circumferential
perineurial-like cell layer is clearly visible. (From Ref. 42, their Fig. 5, top.)
ered Pt microwires (25 m diameter), developed by Horch and

colleagues (3138).
OTHER TYPES OF INTERFACES BETWEEN ELECTRODES AND
NERVE TISSUE
Thus far, insertion of multielectrodes into peripheral nerve
has been considered. As stated, one problem in this approach
is that electrodes may have no target (fiber) close enough to
be exclusive to one electrode (overlap problem). This lowers
the efficiency of a multielectrode. Other ways to interface electrodes and nerve tissue are the regeneration of nerve through
so-called sieves and the culturing of nerve cells on patterned
multielectrode substrates. Both involve growth of nerve fibers
or neurites. If successful, the principal advantage of such devices would be that each electrode has close contact to specific
nerve fibers, reducing the overlap problem and increasing

electrode efficiency.
Especially in neural culturing on planar substrates, a good
understanding of the neuronelectrode interface is of primary
concern and can directly be studied.
Both types of interfaces will be dealt with in subsequent
sections.
REGENERATION SIEVE MICRO ELECTRODE ARRAYS
Another way of interfacing nerves to electrodes is the use of
a 2-D (planar) sieve put in between the two cut end of a nerve.
The silicon sieve permits nerve fibers to regenerate through
metallized hole (or slit) electrodes in the sieve (3943). The
main advantage of this method is that microfabrication of flat
devices is easier than that of 3-D devices. Another advantage
NEUROTECHNOLOGY
371
Figure 7. (a) Low-density neuronal monolayer culture composed of 76 neurons growing over a
matrix of 64 electrodes. The recording craters are spaced 40 m laterally and 200 m between
rows. The transparent indium tin oxide conductors are 10 m wide. Tissue is mouse spinal cord;
culture age is 27 days in vitro; histology is Loots-modified Bodian stain. (From Ref. 60, their Fig.
2, p. 284.) (b) Cultured hippocampal neurons on patterned self-assembled monolayers. A hybrid
substrate pattern of trimethyloxysilyl propyldiethylenetriamine (DETA) and perfluorated alkylsilane (13F) showing selective adhesion and excellent retention of the neurites to the DETA regions
of the pattern. (From Ref. 6, their Fig. 4, p. 18.)
is that, once the nerve has been regenerated, the device is

fixed firmly to the nerve. However, since the flats are typically
only 10 m thick, there is a limited chance that nodes of Ranvier will be close to an electrode (typical internode spacing of
a 10 m fiber is 1 mm), thereby limiting the selectivity of
stimulation/recording. Also, nerve fibers tend to grow through
holes not as single fibers, but as a group (fasciculation),
thereby reducing the possibility of selective stimulation. Zhao
et al. (42) report that only when nerves are regenerated
through 100 m hole diameters do they recover anatomically
more or less normal, after 4 to 16 weeks of regeneration, but
with about 40% loss of force in the corresponding muscle.
Smaller holes yielded morphological and functional failures.
PLANAR MICRO ELECTRODE ARRAYS
FOR CULTURED NEURONS
Planar microelectrode arrays, consisting of transparent leads
(indium tin oxide, or gold) to between 10 and 100 electrode
sites (diameter typically 10 m), spaced at 100 m interdistance on glass plates, were used by Gross et al. (44,45), Novak
and Wheeler (46), and others to study the activity and plasticity of developing cultured neuronal networks or brain slices.
In this way, an attractive alternative was sought for the al-
most impossible job of probing many neurons in a growing

network by micropipettes.
An essential prerequisite for high-quality recordings is to
lower the high impedance of the tiny electrode sites to below
about 1 M by additional electroplating of Pt-black (47) and
to increase the sealing resistance between cell and substrate
by promoting adhesion. The latter can be achieved by coating
of the glass substrate with laminin-, polylysine-, or silanebased (mono)layers (4850).
Yet a number of neurons will adhere too far away from the
electrode sites to produce measurable action potentials. This
led Tatic-Lucic et al. (51) to the design of arrays consisting of
electrode wells, in which single embryonic neural somata
were locked up. Only their neurites could protrude from the
well to form neural networks. In this way, unique contacts
are established, to be used as bidirectional probes into the
network. Alternatively, one can improve the contact efficiency
by patterning the adhesive layer; it is even possible to guide
neural growth (52); for example, around and over electrodes.
On the electrode side, improvements are sought by incorporating an insulated gate field effect transistor (ISFET) in each
electrode (53).
There is a considerable difference regarding whether stimulation or recording concerns an axon in a peripheral nerve
372
NEUROTECHNOLOGY
trunk or a nerve cell body (called soma) lying over an electrode site on a multielectrode substrate. This is studied by
modeling and measurement of electrode impedance as a function of cell coverage and adhesion (5456).
Except for neural network studies, cultured arrays may
once be used as cultured neuron probes. They may be implanted in living nerve tissue to serve as a hybrid interface
between electronics and nerve. The advantage would be that
the electrodecell interface may be established and optimized
in the lab, while the nerve network after implantation may be
a realistic target for ingrowth of nerve (collaterals). Studies of
the feasibility of this approach are currently underway.
CHRONIC IMPLANTATION AND BIOCOMPATIBILITY
For future use in humans, chronic implantation behavior and
biocompatibility studies of microelectrode arrays will become
of crucial importance.
McCreery et al. (57) implanted single Ir microwire electrodes in cat cochlear nucleus and found tissue damage after
long stimulation, highly correlated to the amount of charge
per phase. The safe threshold was 3 nC/phase (while the
stimulus threshold was about 1 nC/phase). Lefurge et al. (32)
implanted intrafascicularly Teflon-coated PtIr wires, diameter 25 m. They appeared to be tolerated well by cat nerve
tissue for six months, causing little damage. The influence of
silicon materials silicon microshaft array rabbit and cat cortical tissue was investigated by Edell et al. (58) and Schmidt et
al. (59). While neuron density around the 40 m shafts decreased, tissue response along the shafts was minimal over
six months (58), except at the sharp tips.
BIBLIOGRAPHY
1. D. McNeal, Analysis of a model for excitation of myelinated
nerve, IEEE Trans. Biomed. Eng., 23: 329337, 1976.
2. F. Rattay, Analysis of models for extracellular fiber stimulation,
3. E. Warman, W. M. Grill, and D. Durand, Modeling the effect of
electric fields on nerve fibers: Determination of excitation threshold, IEEE Trans. Biomed. Eng., 39: 12441254, 1992.
4. S. Nagarajan and D. Durand, Effects of induced electric fields
on finite neuronal structures: A simulation study, IEEE Trans.
Biomed. Eng., 40: 11751188, 1993.
5. W. M. Grill, Jr. and J. T. Mortimer, The effect of stimulus pulse
duration on selectivity of neural stimulation, IEEE Trans. Biomed. Eng., 43: 161166, 1996.
6. J. H. Meier et al., Simulation of multipolar fiber selective neural
stimulation using intrafascicular electrodes, IEEE Trans. Biomed. Eng., 39: 122134, 1992.
7. J. H. Meier, W. L. C. Rutten, and H. B. K. Boom, Force recruitment during electrical nerve stimulation with multipolar intrafascicular electrodes, Med. Biol. Eng. Comput., 33 (3): 409417,
1995.
8. P. H. Veltink et al., A modeling study of nerve fascicle stimulation, IEEE Trans. Biomed. Eng., 36: 683692, 1989.
9. E. V. Goodall et al., Modeling study of size selective activation of
peripheral nerve fibers with a tripolar cuff electrode, IEEE Trans.
Rehabil. Eng., 3: 272282, 1995.
10. W. L. C. Rutten, H. van Wier, and J. M. H. Put, Sensitivity and
selectivity of intraneural stimulation using a silicon electrode
11. W. L. C. Rutten, J. P. A. Smit, and J. A. Bielen, Two-dimensional

neuro-electronic interface devices: Force recruitment, selectivity
and efficiency, Cell. Eng., 2 (4): 132137, 1997.
12. W. L. C. Rutten et al., 3-D Neuro-electronic interface devices for
neuromuscular control: Design studies and realisation steps, Biosensors Bioelectron., 10: 141153, 1995.
13. K. L. Drake et al., Performance of planar multisite microprobes
in recording extracellular single-unit intracortical activity, IEEE
Trans. Biomed. Eng., 35: 719732, 1988.
14. D. A. Evans et al., Multiple-channel stimulation of the cochlear
nucleus, Otolaryngol. Head Neck Surg., 101: 651657, 1989.
15. D. J. Anderson et al., Batch-fabricated thin-film electrodes for
stimulation of the central auditory system, IEEE Trans. Biomed.
Eng., 36: 693704, 1989.
16. S. J. Tanghe, K. Najafi, and K. D. Wise, A planar IrO multichannel stimulating electrode for use in neural prostheses, Sensors
Actuators, B1: 464467, 1990.
17. J. Ji, K. Najafi, and K. D. Wise, A scaled electronically-configurable multichannel recording array, Sensors Actuators, A22: 589
591, 1990.
18. J. Ji and K. D. Wise, An implantable CMOS circuit interface for
multiplexed microelectrode recording arrays, IEEE J. Solid-State
Circuits, 26: 433443, 1992.
19. S. J. Tanghe and K. D. Wise, a 16-channel CMOS neural stimulating array, IEEE J. Solid-State Circuits, 27: 18191825, 1992.
20. A. C. Hoogerwerf and K. D. Wise, A 3D micro-electrode array for
chronic neural recording, IEEE Trans. Biomed. Eng., 41: 1136
1146, 1994.
21. C. Kim and K. D. Wise, A 64-site multishank CMOS low-profile
neural stimulating probe, IEEE J. Solid-State Circuits, 31: 1230
1238, 1996.
22. D. T. Kewley et al., Plasma-etched neural probes, Sensors Actuators, A58: 2735, 1997.
23. K. Najafi, J. Ji, and K. D. Wise, Scaling limitations of silicon
multichannel recording probes, IEEE Trans. Biomed. Eng., 37:
111, 1990.
24. K. Najafi and J. F. Hetke, Strength characterization of silicon
microprobes in neurophysiological tissues, IEEE Trans. Biomed.
Eng., 37: 474481, 1990.
25. J. F. Hetke, K. Najafi, and K. D. Wise, Flexible miniature ribbon
cables for long-term connection to implantable sensors, Sensors
Actuators, A23: 9991002, 1990.
26. J. F. Hetke et al., Silicon ribbon cables for chronically implantable microelectrodes arrays, IEEE Trans. Biomed. Eng., 41:
314321, 1994.
27. P. K. Campbell et al., A silicon-based three-dimensional neural
interface: Manufacturing processes for an intracortical electrode
28. C. T. Nordhausen, P. J. Rousche, and R. A. Normann, Optimizing
recording capabilities of the Utah intracortical electrode array,
Brain Res., 637: 2736, 1994.
29. W. Ehrfeld and D. Munchmeyer, LIGA method, three dimensional microfabrication using synchrotron radiation, Nucl. Instrum. Methods Phys. Res., A303: 523531, 1991.
30. J. T. Mortimer et al., Perspectives on new electrode technology
for stimulating peripheral nerves with implantable motor prostheses, IEEE Trans. Rehabil. Eng., 3: 145154, 1995.
31. M. Malagodi, K. W. Horch, and A. A. Schoenberg, An intrafascicular electrode for recording of action potentials in peripheral
nerves, Ann. Biomed. Eng., 17: 397410, 1989.
32. T. Lefurge et al., Chronically implanted intrafascicular recording
electrodes, Ann. Biomed. Eng., 19: 197207, 1991.
33. N. Nannini and K. Horch, Muscle recruitment with intrafascicular electrodes, IEEE Trans. Biomed. Eng., 38: 769776, 1991.
NOISE AND INTERFERENCE MODELING

34. E. V. Goodall and K. W. Horch, Separation of action potentials in
multi-unit intrafascicular recordings, IEEE Trans. Biomed. Eng.,
39: 289295, 1992.
35. T. McNaughton and K. Horch, Action potential classification with
dual channel intrafascicular electrodes, IEEE Trans. Biomed.
Eng., 41: 609616, 1994.
36. T. G. McNaughton and K. W. Horch, Metallized polymer fibers
as leadwires and intrafascicular microelectrodes, J. Neurosci.
Methods, 70: 103110, 1996.
37. K. Yoshida and K. Horch, Reduced fatigue in electrically stimulated muscle using dual channel intrafascicular electrodes with
interleaved stimulation, Ann. Biomed. Eng., 21: 709714, 1993.
38. K. Yoshida and K. Horch, Closed-loop control of ankle position
using muscle afferent feedback with functional neuromuscular
39. D. J. Edell, A peripheral nerve information transducer for amputees: Long-term multichannel recordings from rabbit peripheral
40. G. T. A. Kovacs, C. W. Storment, and J. M. Rosen, Regeneration
microelectrode array for peripheral nerve recording and stimulation, IEEE Trans. Biomed. Eng., 39: 893902, 1992.
41. G. T. A. Kovacs et al., Silicon-substrate micro-electrode arrays
for parallel recording of neural activity in peripheral and cranial
42. Q. Zhao et al., Rat sciatic nerve regeneration through a micromachined silicon chip, J. Biomater., 18: 7580, 1997.
43. A. Blau et al., Characterization and optimization of microelectrode arrays for in-vivo nerve signal recording and stimulation,
Biosensors Bioelectron., 12: 883892, 1997.
44. G. W. Gross, Simultaneous single unit recording in vitro with a
photoetched laser deinsulated gold multimicroelectrode surface,
45. G. W. Gross, W. Y. Wen, and J. W. Lin, Transparent indium-tin
oxide electrode patterns for extracellular, multisite recording in
neuronal cultures, J. Neurosci. Methods, 15: 243252, 1985.
46. J. L. Novak and B. C. Wheeler, Recording from the Aplysia abdominal ganglion with a planar microelectrode array, IEEE
Trans. Biomed. Eng., 33: 196202, 1986.
47. W. Nisch et al., A thin film microelectrode array for monitoring
extracellular neuronal activity in vitro, Biosensors Bioelectron., 9:
737741, 1994.
48. J. P. Ranieri et al., Spatial control of neuronal cell attachment
and differentiation on covalently patterned laminin oligopeptide
substrates, Int. J. Develop. Neurosci., 12 (8): 725735, 1994.
49. J. F. Clemence et al., Photoimmobilization of a bioactive laminin
fragment and pattern-guided selective neuronal cell attachment,
Bioconjugate Chem., 6 (4): 411417, 1995.
50. J. M. Corey, B. C. Wheeler, and G. J. Brewer, Micrometer resolution silane-based patterning of hippocampal neurons: A comparison of photoresist and laser ablation methods of fabricating substrates, IEEE Trans. Biomed. Eng., 43: 944955, 1996.
51. S. Tatic-Lucic et al., Silicon-micromachined neurochips for in
vitro studies of cultured neural networks, Proc. 7th Int. Conf.
Solid State Sensors Actuators, Yokohama, Japan, 1993, pp.
943946.
52. D. W. Branch et al., Microstamp patterns of biomolecules for
high-resolution neuronal networks, Med Biol. Eng. Comput., 36:
135141, 1998.
53. A. Stett, B. Muller, and P. Fromherz, Two-way silicon-neuron interface by electrical induction, Phys. Rev. E, 55: 17791782, 1997.
54. R. Weis and P. Fromherz, Frequency dependent signal transfer
in neuron transistors, Phys. Rev. E, 55: 877889, 1997.
55. M. Bove, M. Grattarola, and G. Verreschi, In vitro 2-D networks
of neurons characterized by processing the signals recorded with
373
a planar microtransducer array, IEEE Trans. Biomed. Eng., 44

(10): 964978, 1997.
56. J. R. Buitenweg et al., Measurement of sealing resistance of cellelectrode interfaces in neuronal cultures using impedance spectroscopy, Med. Biol. Eng. Comput., 1998.
57. D. B. McCreery et al., Stimulus parameters affecting issue injury
during microstimulation in the cochlear nucleaus of the cat,
Hear. Res., 77: 105115, 1994.
58. D. J. Edell et al., Factors influencing the biocompatibility of insertable silicon microshafts in cerebral cortex, IEEE Trans. Biomed. Eng., 39: 635643, 1992.
59. S. Schmidt, K. Horch, and R. Normann, Biocompatibility of silicon-based electrode arrays implanted in feline cortical tissue, J.
Biomed. Mater. Res., 27: 13931399, 1993.
60. D. A. Stenger and T. M. McKenna (eds.), Enabling Technologies
for Cultured Neural Networks, San Diego, CA: Academic Press,
1994.
Reading List
I. A. Boyd and M. R. Davey, Composition of Peripheral Nerves, Edinburgh: Livingstone, 1968.
W. F. Agnew and D. B. McCreery (eds.), Neural Prostheses Fundamental Studies, Englewood Cliffs, NJ: Prentice-Hall, 1990.
J. Malmivuo and R. Plonsey, Bioelectromagnetism, Oxford, UK: Oxford Univ. Press, 1995.
WIM L. C. RUTTEN
University of Twente
NEUTRON FLUX MEASUREMENT. See FISSION

CHAMBERS.
NIGHT PILOTAGE, HELICOPTER. See HELICOPTER

NIGHT PILOTAGE.
NMR. See MAGNETIC RESONANCE.
ELECTRIC IMPEDANCE IMAGING
conductivity or admittivity j and we have the partial differential equation

The imaging of electrical conductivity and permittivity of the
interior of a body from fixed-frequency electrical measurements at the boundary has come to be called electrical impedance tomography (EIT) although it is quite different from the
true tomographic imaging methods in that slices cannot be
imaged independently. The earliest specific references are
Langer (1) and Slichter (2) in 1933, but little work was published until Henderson and Websters paper entitled An Impedance Camera for Spatially Specific Measurements of the
Thorax (3) in 1979, which proposed EIT as a safe, noninvasive imaging method and stimulated interest in the subject.
By this time it was practical (economic) to implement electronic systems capable of measuring with sufficient accuracy
and of computing with sufficient speed for use in medical imaging applications. The developments in this context led others to pursue the method, notably in process monitoring and
geophysical prospecting, where there are different timescales
and boundary constraints.
The method has potential for imaging because of the impedance differences of naturally occurring substances. For example, in the medical context
Resistivity, in m
Human blood
Lung tissue
Bone
Muscle (longitudinal)
Muscle (transverse)
Brain (gray matter)
Brain (white matter)
Fat
Liver
( ) =
x
x
z
z
=0
(2)
(which is the continuum equivalent of Ohms law and Kirchhoff s law combined) subject to the Neumann boundary condition
n | = J
(3)
For a known admittivity, solving the boundary value problem

given by Eqs. (2) and (3) will be called the forward problem.
It can be solved numerically using, for example, the finite element method (FEM). In EIT one applies a number of independent current patterns J at the surface and makes measurements of the potential also at the surface in an attempt
to determine in the interior. This is called the inverse
problem.
Once the current density on the boundary and the admittivity is specified, the potential is determined up to an additive constant, which we eliminate by setting
dS = 0
As there are no sources of current in the interior, Gausss

law implies that the surface integral of the current density
vanishes

Permittivity is measured when investigating low conductivity substances like air/oil/water mixtures in pipelines by
using capacitively coupled electrode systems. Resistive values
are typically found when low-frequency excitation is used
with directly coupled electrodes on conductive objects. Many
of the resistive substances also have a small reactance that
becomes measurable when high-frequency excitation is used.
PHYSICAL THEORY
Take a body in three-dimensional space with spatial variable x (x, y, z) outward unit normal n. Suppose the body
has possibly inhomogeneous isotropic conductivity (x), permittivity (x), and permeability (x). A time-harmonic current
density J(x, t) J(x)ejt with angular frequency is applied
to the surface , and this results after some settling time in
(x, t) E(x)ejt and magnetic field H
(x, t)
an electric field E
H(x)ejt in the body. Maxwells equations then give us
E
H = ( j)E
y
y
1.5
7.324.0
170
1.31.5
1823
2.8
6.8
2128
3.55.5
H
E = jH
301
(1)
Assuming sufficiently small permeability and frequency, we

make the approximation E 0 and therefore E ,
where is the electric potential. We now define the complex
JdS = 0
With these conditions surface current density and surface potential are related by a linear operator, the transfer impedance operator R()J (referred to in the mathematical
literature as the Neumann-to-Dirichlet mapping). The operator R() represents a complete knowledge of boundary electrical data. In EIT we sample this operator using a system of
electrodes to apply current and measure potential.
The first problem that needs to be addressed is the theoretical possibility of determining from R(). Specifically, the
question Does R(1) R(2) imply 1 2? has been answered in the affirmative under a variety of smoothness assumptions for the i. For details of these results, including
the case where the i are complex, see Isakov (4). The closely
related problem of recovering the resistance values of a planar resistor network by boundary current and voltage measurement has been investigated by Curtis and Morrow (5) and
Colin de Verdiere (6).
For the case where is not negligible, Ola and Somersalo
(7) show that the electrical parameters and are uniquely
determined by a complete knowledge of boundary data n
E and n H, provided is not the resonant frequency.
The problem of actually recovering the admittivity from a
noisy, sampled boundary data is difficult for two main reasons: The problem is nonlinear and ill posed. Notice that the
potential depends on , so that Eq. (2) is a nonlinear equa-
302
tion for as a function of . The nonlinearity is illustrated

for a simple but typical example in Fig. 1.
The current is applied to the surface and voltage measurements made using a system of conducting electrodes. A typical EIT system with l drive electrodes D1, . . ., Dl and m measurement electrodes M1, . . ., Mm will have single-ended
digitally controlled current sources connected to all but one
drive electrode (multiple-drive system), or a single doubleended current source connected to the drive electrodes by a
system of multiplexers (pair drive system). The in phase and
quadrature components of the voltage are measured each of
the measurement electrodes.
A current pattern takes the form Ji Ii11 Iill,
where j is the normalized current density on the jth drive
electrode (which for the moment we will assume to be one on
the electrode and zero elsewhere), assuming the area of each
drive electrode is the same I1 Il 0. Let us assume
for simplicity that measurement electrodes are points and
that voltage is measured relative to Mm. Let j be the potential when Jj is applied. Then the measurements made are

Vij = j (Mi ) j (Mm ) =
j (xx )((xx Mi ) (xx Mm )) dS
(4)
We define the lead field i to be the potential that would arise

if a unit current were passed through the measurement elec-
1.5
0.5
2
log 10(s)
Figure 1. The simple example illustrates the typical sigmoid response of boundary impedance measurement to interior conductivity
change. Let be a unit height unit radius cylinder. Suppose that the
current density on the surface (using cylindrical coordinates (, , z)
is J(1, , z) cos and J(, , 1/2) 0. Let us assume a cylindrical
anomaly with radius r and conductivity

(, , z) =
60
80
100
120
1
2
3
4
5
6
Figure 2. The singular values of the sensitivity matrix give a clear

illustration of ill conditioning of the linearized inverse problem. The
singular values of S are the square roots of the eigenvalues of STS
arranged as a decreasing sequence 1 2 0. For a signal-tonoise ratio of S one would expect to be able to identify K conductivity
parameters, where K is the largest integer with 1 / K S. These
singular values were calculated using a two-dimensional finite element mesh, 16-point electrodes equally spaced, and trigonometric
current patterns. [After Breckon (11).]
trodes, i /n (x Mi) (x Mm). We can express the

voltage measurement in an integral form as

n dS =
Vij =
j i /n
i j dV
(5)
Now suppose that the admittivity is changed to and

the potential and lead fields change to j j and i i,
respectively, while the boundary current densities remain
fixed. We then have the expansion

n dS = i j dV
Vij =
j i /n
(6)

+ higher order terms
40
RECONSTRUCTION ALGORITHMS
Z(0.6,s)
2.5
20
1 r<1
s
0r
Then (1, , z) Z(s, r) cos , where Z(s, r) 1 mr2 /1 mr2 and

m (s 1)/(s 1). The figure illustrates the response of the boundary impedance Z to a change in conductivity contrast over six decades
for a radius of 0.6.
This gives the basis of a linear reconstruction algorithm. We

use a numerical model (FEM, for example) of the system with
admittivity m, which can be used to calculate predicted voltages Vmij. We measure experimental voltages Veij and set Vij
Veij Vmij, ignoring the higher-order terms. We can calculate a
correction by solving a system (m 1)(l 1) integral equations. This system of equations can be written matrix form
as V Sg, where V is the vector composed of the Vij and g
is the vector of components of with respect to some family
K
of independent functions k(x), (x) k1 gkk(x). Typically,
a pixel basis is used where each k is one in the kth pixel (or
voxel) and zero elsewhere, but other choices of basis are possible. The Jacobian or sensitivity matrix S then has elements

Sijk =
k i j dV
(7)
The linear system V Sg is highly ill conditioned (see Fig.

2), which means that it can only be solved with some regularization or smoothing of the admittivity. A simple example (Tichonov regularization) is to solve instead the system STV
(STS 2I)g for some small parameter . The resulting conductivity update can be added to an assumed background
admittivity to produce an approximate image. This simple linear reconstruction algorithm is similar to the NOSER algorithm used by Rensselaer Polytechnic Institute (RPI) (8). As
the inverse of STS 2I can be precomputed assuming a suitable background conductivity, the algorithm is quite fast
(quadratic in the number of measurements used). However,
as it is a linear approximation the admittivity contrast will
be underestimated and some detail lost (see Fig. 1). A fully
nonlinear algorithm can be implemented by recalculating the
Jacobian using the updated admittivity and solving the regularized linear system to produce successive updates to the
admittivity until the numerical model fits the measured data
to within measurement precision. This requires an accurate
forward model, including the shape of the domain (8,9) and
modeling of the electrode boundary conditions (10). The nonlinear algorithm is more computationally expensive as at each
iteration the voltages have to be recalculated and the linear
system solved.
There is still debate about the ideal current patterns Ji to
drive. For a given constraint on the allowable current levels,
an optimal set of current drives can be calculated. In the case
where the total dissipated power is the active constraint, the
optimal currents are as described by Cheney (12). In the case
of a two-dimensional disk where the unknown conductivity is
rotationally symmetric, these are the trigonometric current
patterns Iik cos ik, where k is an angular coordinate of the
kth drive electrode and 1 i l/2 (similar for sine). If the
active constraint is the total injected current, only pairs of
electrodes should be driven (13); on the other hand, if the
maximum current on each electrode is the only constraint,
then all electrodes should be driven with positive or negative
currents (Walsh functions). In medical applications the belief
that limiting the dissipated power is the most important
safety criterion has led to the design of systems with multiple
current drives.
TISSUE IMPEDANCE
Spectral information is of particular interest in medical applications, where it can improve tissue characterization. A variety of electrical models of tissue have been proposed to explain the variation of impedance with frequency, the most
widely used being the Cole plot. The tissue model in Fig. 3
would give rise to the Cole plot in Fig. 4. The difference between the model and experimental findings is explained by
assuming that the capacitive element has a complex reactance given by K( j)a, where 1 would be a standard
capacitor, but in tissue 0.8 typically. A different interpre-
Z(dc)
z(m)
S
Figure 3. Simplest tissue impedance model where Z(m) is the cell
membrane capacitance, S is the intracellular impedance, and Z(dc) is
extracellular impedance.
303
Imaginary
impedance
negative
Real
impedance
Z
Z(dc)
Depressed center of
impedance locus
Figure 4. Locus of impedance versus frequency for the simple tissue model.
tation of tissue impedance is that the capacitance is distributed, and this may give a similarly depressed Cole plot. A
comprehensive treatment is given in a review by Rigaud (14).
ELECTRONIC SYSTEMS
In any application, the impedance contrast in the region, the
size of the smallest distinguishable object, and the rate of
change of any impedance in the whole region set the measurement parameters for the electronic and computation tasks.
The majority of systems apply current and measure voltage, and, as the analysis shows, accurate current sources are
required in multiple-source systems. Single-current drive systems only require measurement of the current, and no adjustments will be required as long as the current stays constant
for the duration of a measurement set. This will be achieved
if the output impedance is much larger than the changes in
the impedance of the region being imaged. Multiple-drive systems may include current measurement and subsequent adjustment to generate a particular current field, but greater
operating speed is possible with deterministic current
sources.
High-output impedance current sources have been adapted
or developed for EIT in a number of ways. The RPI group
uses digitally adjustable negative resistance and negative capacitance circuits on each current source to obtain 64 M
(15).
The Oxford Brookes University group uses modified Howland sources to obtain 300 k at 160 kHz (Fig. 5). A calibration system is used to measure the output characteristics and
compensate the set current level for each source, thereby
allowing very precise currents to be set for as long as the
calibration coefficients remain constant.
Alternating currents (ac) are used for several reasons: dc
applied to the skin causes electrolytic action under the electrode and may result in ulcers. The maximum ac current that
may safely be applied is frequency dependent, with the maximum safe current increasing linearly with frequency between
1 kHz and 100 kHz. Since voltage data accuracy improves
with larger applied currents, this implies that the use of
higher frequencies may result in more accurate data. However, the accuracy of the data acquisition system may become
degraded at higher frequencies due to the effects of parasitic
capacitances. The most popular frequency range is from 10
304
measurement circuits are used or if less than ideal measurement patterns are used.
This approach is applicable to both the pair-drive excitation method and multiple drive, since both require the establishment of multiple current patterns and concurrent voltage
measurement for a full data set for the reconstruction of a
single image. With 32 drive electrodes, 31 independent adjacent pair-drive combinations are possible; 31 independent
trigonometric or optimal patterns can be set by multipledrive systems.
+VS
RF
C
VC
RI
+
RL
APPLICATION AREAS
Medical Imaging
VS
Figure 5. Supply current sensing current source, modified to stabilize the dc level at the output loop gain AL A0 RI /RL.
kHz to 1000 kHz (1621), although the system by Cusick (13)

operates down to 100 Hz.
The current source design was improved to overcome an
imbalance condition when an ac coupled load is used. The resistor RF (100 MW) provides a dc negative feedback path: C
(0.1 F) has, relatively, a negligible impedance at 10 kHz,
avoiding degradation of the ac performance of the supply-current-sensing current source (23).
Voltage Measurement
Electrode voltage signals are typically fed through coaxial cables with driven screens. A wide variety of instrumentation
amplifiers with suitable bandwidth and low noise performance is now available. Early EIT voltage measurement designs incorporated analog multipliers to demodulate (synchronously detect) and measure the in-phase and quadrature
components of the electrode voltages. The AD630 was frequently used. The output of the multiplier was low-pass filtered to obtain the impedance signal. There is necessarily a
compromise between the amount of carrier left in the filtered
signal and the response time of the filter to changes in the
impedance. Digital demodulation was introduced by the RPI
group and others, allowing better signal processingremoval
of interference, signal averaging, narrower bandwidth, and
adaptive filtering for different excitation signals. A nonuniform, oversampling method has been widely adopted: A digital signal generator is used to produce a reference sinusoidal
waveform with, for instance, 256 samples per cycle. Therefore, 256 independent sampling points are available and can
be used over 32 cycles to make 256 independent measurements from each of 32 electrode signals.
The duration of the sampling period for each electrode is
roughly the whole measurement period, and hence the bandwidth is the reciprocal of the total period. Nonsynchronous
noise will be filtered out. Synchronous noise will be minimized. The limitation with this method is the settling time
after switching a new electrode signal to the ADC. Using
video multiplexers, a sampling rate of 2.5 MHz is achievable;
this gives a measurement period of 3 ms for each current pattern that is applied. Higher speeds are achievable if parallel
The first EIT equipment to become available commercially

was the Sheffield Mark 1 2-D real-time difference imaging
system (16). This system is able to show cardiac and respiratory cycles by displaying the difference in impedance between
a reference measurement and the present measurement. The
choice of reference can be end expiration for the best imaging
of respiration or a point triggered from the electrocardiogram
(ECG) waveform for best cardiovascular imaging.
More recently the Sheffield group has developed a 3-D system (24) by using eight planes of electrodes on a subjects
chest. Differencing imaging is again employed, but the differences can be obtained by using multiple frequencies rather
than temporal differences. Advanced image presentation techniques allow isometric viewing of the combined images from
all planes. They have shown good resolution perpendicular to
the electrode planes by imaging phantoms with targets
clearly resolved on one plane and showing only faintly on the
adjacent planes.
The leading North American Medical EIT group at RPI has
achieved real-time imaging of human and dog chests using a
multiple-drive system with 32 electrodes, delivering current
and measuring voltage on the same electrodes. Real time refers to real-time data acquisition but slower off-line image reconstruction. Full-speed image replay is done later. The RPI
group has been able to show cardiac and respiratory cycles as
absolute impedance images and has performed physiological
experiments with altered blood flow to the left and right lungs
and imaged clear differences. The groups data also show concurrent rises in cardiac impedance with falls in lung impedance at the subjects heart rate.
Another static (or absolute) EIT system producing moving
images has been developed at Oxford Brookes University.
Real-time (as defined previously) imaging of human chests
has shown cardiac and respiratory cycles, but poorer localization of the impedance changes than the similar RPI system.
Principal component analysis applied to the data has been
used to separate the cardiac-synchronous in-phase and (expected) antiphase heart and lung impedance changes.
Novel configurations of the Sheffield system have been developed by Smallwood (25) and by Kotre (26) in which a planar array of electrodes is applied to the front of the abdomen
for monitoring gastric emptying. A modified back-projection
reconstruction technique is used to localize the impedance
changes close to the current-carrying electrodes, thereby
allowing a simple method of imaging.
Other principal medical EIT applications include a system
developed by Jossinet for investigating the detection of breast
tumors, where absolute measurements are needed, and the
localization of epileptic foci and detection of brain ischemia

(27).
Impedance Spectroscopy
In parallel with EIT imaging, many investigators have examined the impedance spectrum of tissue in vivo by placing electrodes in or close to supposed homogeneous regions for tissue
characterization based on a lumped model for intra/extracellular current flow. In vitro measurements on isolated tissue
have also been used to develop the Cole models. It is expected
that when high-quality EIT data are available, it will be possible to detect pathological changes in tissues close to the skin
surface, such as breast tumors, or unsuccessfully implanted
kidneys.
Process Monitoring
Since the mid-1980s EIT techniques have been developed for
process monitoring in industrial applications involving the
identification of mixed components in pipelines, vessels, conveyers, and reactors. The excitation mode depends on
whether the substances to be monitored and the container are
conducting or insulating. Conducting substances in a conducting vessel are amenable to the techniques described for medical imaging as long as independent electrode sites are available on the container wall. A variant of this method is
electromagnetic induction tomography, in which images are
produced of eddy current loss in the materials. In the many
cases where the container is insulating, electrical capacitance
tomography (ECT) provides a rapid technique with a resolution comparable to EIT. Due to the large difference between
water and oil permittivity (40 : 1), ECT is particularly appropriate to the oil industry. Dual ECT systems placed a short
distance apart on a pipeline have been used to monitor flow
by cross-correlating image features.
Imaging improvements have come from the placing of electrodes on stirrers or other accessible fixings within vessels,
thereby overcoming the problem of poor resolution at the center of standard EIT images.
Geophysical Surveying and Archaeological Exploration
Electrical resistance tomography (ERT) can be used to image
principally the water content of soil and porous rocks. The
naturally occurring resistivity range is approximately 100
m to 10,000 m; in experiments, changes up to 10% are
seen when saline is added. Electrodes are inserted in deep
boreholes, current is passed between them, and the resulting
potentials are measured on either surface or other deep electrodes. The reconstruction methods have been adapted to consider the case where the electrodes are surrounded by conducting material. Three-dimensional imaging is best achieved
by having deep voltage measurement electrodes, at the cost
of getting them there. Similarly, anomalies in the resistance
image can be of use in detecting buried archaeological features. In this context, time and current density are not constrained and will allow better resolution than in the medical
context.
BIBLIOGRAPHY
1. R. E. Langer, On an inverse problem in differential equations,
Bull. Amer. Math. Soc., 39: 814820, 1933.
305
2. L. B. Slichter, The interpretation of the resistivity prospecting

method for horizontal structures, J. Appl. Phys., 4: 307322,
1933.
3. R. P. Henderson and J. G. Webster, An impedance camera for
spatially specific measurement of the thorax, IEEE Trans. Biomed. Eng., 25: 250254, 1978.
4. V. Isakov, Inverse Problems for Partial Differential Equations,
New York: Springer-Verlag, 1998.
5. E. B. Curtis and J. A. Morrow, Determining the resistors in a
network, SIAM J. Appl. Math., 50: 918930, 1990.
6. Y. Colin de Verdiere, J. Gitler, and D. Vertigan, Planar electrical
networks. 2, Commentarii Mathematici Helvetici, 71: 144167,
1996.
7. P. Ola and E. Somersalo, Electromagnetic inverse problems and
generalized Sommerfeld potentials, SIAM J. Appl. Math., 56 (4):
11291145, 1996.
8. H. Jain et al., Electrical impedance tomography of complex conductivity distributions with noncircular boundary, IEEE Trans.
Biomed. Eng., 44: 10511060, 1997.
9. W. R. B. Lionheart, Boundary shape and electrical impedance
tomography, Inverse Problems, 14: 139147, 1998.
10. K. Paulson, W. Breckon, and M. Pidcock, Electrode modeling in
electrical-impedance tomography, SIAM J. Appl. Math., 52: 1012
1022, 1992.
11. W. R. Breckon, Image reconstruction in electrical impedance tomography, Ph.D. thesis, Oxford Polytechnic, 1990.
12. M. Cheney and D. Isaacson, Distinguishability in impedance imaging, IEEE Trans. Biomed. Eng., 39: 852860, 1992.
13. B. M. Eyuboglu and T. C. Pilkington, Comments on distinguishability in electrical-impedance tomography, IEEE Trans. Biomed.
Eng., 41: 505, 1994.
14. B. Rigaud, J-P. Morucci, and N. Chauveau, Bioelectrical impedance techniques in medicine, Crit. Rev. Biomed. Eng., 24 (46):
257351, 1996.
15. R. D. Cook et al., ACT3: A high-speed, high-precision electrical
impedance tomograph, IEEE Trans. Biomed. Eng., 41: 713722,
1994.
16. D. C. Barber, B. H. Brown, and I. L. Freeston, Imaging spatial
distributions of resistivity using applied potential tomography,
Electron. Lett., 19: 933935, 1983.
17. P. J. Riu et al., Multifrequency static imaging in electrical-impedance tomography. 1. Intrumentation requirements, Med. Biol.
Eng. Comput., 33: 784792, 1995.
18. H. Griffiths and J. Jossinet, Bioelectrical spectroscopy from
multi-frequency EIT, Physiol. Meas., 15 (Suppl. 2A): A5963,
1994.
19. J. Jossinet, C. Trillaud, and E. T. McAdams, A distributed frontend tomograph for high-frequency electrical impedance imaging,
Proc. 8th ICEBI, Kuopio, Finland, 1992.
20. P. M. Record, Single-plane multifrequency electrical impedance
instrumentation, Physiol. Meas., 15 (Suppl. 2A): A2935, 1994.
21. Q. Zhu et al., An adaptive current tomography using voltage
sources, IEEE Trans. Biomed. Eng., 40: 163168, 1993.
22. G. Cusick et al., A system for impedance imaging epilepsy in ambulatory human subjects, Innov. Tech. Biol. Med., 15 (1): 3339,
1994.
23. C. W. Denyer et al., A high-output impedance current source,
Physiol. Meas., 15: A79A82, 1994.
24. P. Metherall et al., Three-dimensional electrical-impedance tomography, Nature, 380: 509512, 1996.
25. R H. Smallwood, Y. F. Mangnall, and A. D. Leathard, Transport
of gastric contents, Physiol. Meas., 15: A175A188, 1994.
26. C. J. Kotre, Subsurface electrical-impedance imaging using orthogonal linear electrode arrays, IEEE Proc.-Sci. Meas. Tech.,
143: 4146, 1996.
306
ELECTRICITY SUPPLY INDUSTRY
27. D. S. Holder, Electrical-impedance tomography with cortical or

scalp electrodes during global cerebral-ischemia in the anesthetized rat, Clin. Phys. Physiol. Meas., 13: 8798, 1992.
CHRISTOPHER N. MCLEOD
WILLIAM R. B. LIONHEART
Oxford Brookes University
ELECTRICITY METERING. See WATTHOUR METERS.

ELECTRICITY, STATIC. See STATIC ELECTRIFICATION.
PHYSIOLOGICAL MODELS SURVEY

PHYSIOLOGICAL MODELS, DEVELOPMENT
This article illustrates some basic principles of modeling in
the physiological sciences, with emphasis on a salient feature that receives more and more attention, namely the issue of variability, measurable as uctuations or oscillations
(13). Along these lines one may also consider the experimental application of so-called perturbation techniques
that induce instabilities in a system, while the observed
response reveals important information about the characteristics of such a system (4, 5). Also, circadian rhythms
and their implications for physiological processes are well
recognized and intensively investigated today, both in animals and in humans. Heart rate variability (HRV) refers
not only to variations in rhythm, but has also (possibly teleological) consequences for lling and ejection of the cardiac ventricles. In the future we may learn whether the
fundamentals of homeostasis pertain to oscillations rather
than to pure equilibrium. Is the oscillator a source for control or just a clockwork as already conceived by Descartes.
What matters most, entropy or order? And how can intrinsic rhythms be duly implemented in the physiological models?
Time patterns, therefore, seem to play a fundamental
role in normal functioning. There is emerging evidence that
independently operating systems within the body may yet
tend to be phase locked: for example, the coupling between
cardiac and locomotor rhythms (heart rate becomes entrained with walking and running cadence at some speeds)
is a physiological phenomenon (6). Similarly, although not
discussed in this article, a novel and fascinating area of investigation is the technological and even surgical coupling
of different tissues or organ systems. An example of the
therapeutic implications of the rst type is the auditory
biofeedback in the treatment of strabismus (cross eye). In
this process voltage signals proportional to vergence eye
position are transmitted from the eye movement monitoring system to an audio oscillator that converts the signal
to tones proportional to horizontal eye position, that is, the
patient hears the vergence eye movements. An example
of a surgical advance based on physiological models that
enhances our understanding of underlying mechanisms is
the cardiomyoplasty procedure in which a skeletal muscle is wrapped around a malfunctioning heart chamber to
assist in the process of contraction (7). These remarkable
applications demonstrate the potentially wide range and
impact on clinical practice, based on insight derived from
sound physiological models.
The functional importance of variability in physiological
modeling deserves attention. Already in 1865 the famous
French investigator Claude Bernard stated that in physiology we must never make average descriptions of experiments, because the true relations of phenomena disappear
in the average (8). Apparently, his remark has gained new
impetus over the last decade, and the notion of intrinsic
variability may turn out to provide a practical tool in distinguishing physiologic from pathophysiologic states.
A PRIMER OF CONTROL SYSTEMS IN PHYSIOLOGY

Physiology is the biomedical discipline that investigates,
describes, and interprets the mechanisms of functioning
pertaining to the normal (healthy) living being and intgrates this knowledge in the clinical sciences. Living systems and their components are continuously controlled
(and occasionally even perturbed) by internal and external stimuli. Therefore, a short discourse about physiological control systems seems appropriate.
Basically, biological control systems do not differ much
from technological controllers. Physiological control systems often feature the intrinsic property of small oscillations or uctuations, meaning that a moderate degree of
instability is desired in order to react to even small perturbations effectively. This prominent property occurs in various organ systems, for example, the variability of the heart
rhythm, skin temperature oscillations, the small uctuations of the resting transmembrane potential, the management of instability during dynamic exercise, and saccadic
movements encountered in vision research. Small uctuations actually may prevent sudden large overshoots that
could prove fatal.
The principle of minor instability for promoting smooth
operation is familiar to engineers who have developed advanced technological devices, as for example, the booster
engines of a rocket, which continuously rotate during takeoff to prevent tumbling. Also, from everyday life we know
that a stick placed on a nger will surely fall if one attempts to keep the hand rigid; small circular movements
(teasing the Coriolis force?) assist in keeping the stick in an
upright position. Experienced tennis players make clearly
visible lateral oscillating movements with their whole body,
in order to anticipate and initiate a rapid sprint in either
direction when the ball is being served by the opponent.
A control system is an arrangement of components (such
as sensors, comparators, and actuators) that are designed
to govern a particular process. A well-known and relatively
simple example is the heating control used in a room or
house. A typical diagram is presented in Fig. 1 and consists
of a heating unit that can be switched on and off by comparing the desired temperature (set point) with the actual
temperature sensed at the thermostat. Obviously, the dynamics of such a control system depend on many internal
and external factors including weather conditions, capacity of the heater, and possible routes for leakage. A similar
setup operates in biological systems, although the anatomical and physiological details are more complex. The simplest setup refers to a reex loop 1 in which a certain (effective) stimulus elicits a particular response according to a
standard protocol, for example, the blinking reex in which
the eyelid is quickly closed when a suddenly approaching
and threatening object is seen. Apart from temperature,
many other conditions and functions are controlled in living organisms, for example, locomotion, blood pressure and
ow, composition of body uids, release of hormones, and
vision.
Three features of the control system just described deserve further attention:
Figure 1. Diagram of a relatively simple control system. D is the

dial to set the desired temperature. C the comparator, H the heater,
T the thermostat, and S the switch to turn the system on and off.
mic axes, depending on the nature of the process(es) considered.

Various mathematical tools are used for the description
and analysis of physiological models and signals, including
time-domain and Fourier analysis (frequency-dependent
approach). The spectral power density provides a measure
of the distribution of the average power in a signal at different frequencies.
SIMULATION IN PHYSIOLOGY
Figure 2. A control system with a feedback loothat can be interrupted, thus permitting the estimation of the so-called open-loop
gain.
1. In order to keep a process under control, information must be received about the actual status of that
process, that is, the comparator must get feedback
about the outcome of its regulatory action. Feedback
rcuits result in either a negative or positive action;
in the terminology of physiology these effects are referred to as inhibitory and excitatory, respectively.
2. The onoff switch clearly introduces nonlinearities
in the behavior of the system. Such discontinuities
are not uncommon in biology. Sometimes a linearized
approach is selected for simplicity, which still yields
useful results.
3. The regulatory capabilities of a control system depend on the wiring (i.e., how the various elements
are connected), and the gain. When the feedback is
interrupted (a so-called open-loop state) during an experiment or due to a particular disease, the performance may be different from that under the normal
closed loop condition, (Fig. 2). Apart from how powerful a control system is (gain) one may also consider
the time course of a change imposed on the system.
This is generally done by calculating a time constant
that indicates the typical rate by which the system is
regulated.
MODELING IN BIOLOGY
Mathematical and graphical models are convenient because they aid in organizing the pattern of thinking and
thus facilitate communication among colleagues. Also, they
are relatively inexpensive and often applicable to the description and analysis of complex systems. In addition, they
potentially enhance insight by providing suggestions about
clues to critical experiments and validation of methods.
In the mathematical approach an analytical expression
is formulated that indicates how one variable depends on
other parameters. The equation may be precise, based on
a formal derivation, or may be a convenient approximation for the sake of simplicity. A graphical representation
may be multidimensional, with linear or (semi-) logarith-
Elementary models combined with knowledge about their

control systems can be fruitfully employed to simulations
in physiology. Such endeavors may mimic or even predict the outcome of a particular process under well-dened
conditions, either normal (i.e., physiological) or pathological. In general, simulations yield results that resemble
observed phenomena but do not necessarily reect physiological properties. For example, the arterial tree can be
simulated by a two-element electrical analog, the so-called
Windkessel (air bellows), consisting of a resistor (representing the presumed stiff peripheral vessels) and a capacitor
(which stores energy in systole and discharges during diastole; it is formed by the compliant aorta). A later improved model includes the characteristic impedance of the
aorta by adding another resistor in series with the previous two components. It was also shown that a nonlinear
three-element Windkessel model has no advantages over
the traditional version, (9) although the pressure dependence of compliance can be demonstrated.
Mechanical models simulating the systemic circulation
and successfully applied in animal experimental setups
were developed by Westerhof (10) and by Kerkhof (11)
An analog distributed model of the human systemic circulation was given by Westerhof in 1969, and in 1978 reduced by Noordergraaf (12). ORourke and Avolio proposed
a T-shaped model in relation to the anatomical design,
and other investigators developed single and multiple tube
models, thus lumping a small or large portion of the individual elements (13).
BASIC TYPES OF NEURAL CONTROL SYSTEMS

Connections within the nervous system are realized by using synapses, which are electrical coupling sites in which
a specic transmitter substance released from the presynaptic ending will induce a voltage change at the opposing postsynaptic membrane. An excitatory signal induces
a depolarization, whereas an inhibitory signal causes hyperpolarization. This design permits the construction of a
variety of regulatory circuits, all based on the following
four prototypes (where the signal travels from left to right
side):
A divergent circuitry [Fig. 3(a)], in which distributes

a signal to various effectors
A convergent circuitry [Fig. 3(b)], which has an additive effect with possible threshold features
FLUCTUATIONS OF THE MEMBRANE POTENTIAL

Excitable tissues (such as nerve cells and muscle bers)
feature the property of a potential difference across the
membrane during the resting condition. This state is generally referred to as polarization. The electrical potential
stems from (unequal) distributions of several ions, mainly
sodium plus chloride (in the extracellular medium) and
potassium (within the cell). During rest a potential difference of about 80 mV (inside with respect to outside of
the cell) is maintained by the action of an ion pump. Small
spontaneous uctuations (possibly to be regarded as intrinsic perturbations) around this average potential have
been recorded and described as 1/f noise, that is, exhibiting
a spectral intensity obeying a relationship proportional to
f1 , where f is the frequency in hertz (14). The uctuations
imply that signal transmission in the nervous system is a
highly probabilistic process. The French scientist and 1965
Nobel prize winner Jacques Monod elegantly explained the
importance of probability in the eld of (molecular) biology
when he applied cybernetics to molecules (15). Thus the
phenomenon of variability is present at a cellular level and
represents a characteristic element in physiology.
CONTROL OF MOVEMENT AND LOCOMOTION
Figure 3. (a) A divergent (branching) synaptic circuitry, where

the circle denotes a neuron and the v shape ending a synaptic
connection; (b) a convergent circuitry; (c) a parallel circuitry; (d) a
reverberating system.
A parallel circuitry results in a repetitive sequence at

point Z after a single initiating impulse originating at
A, because every synapse introduces a certain delay
[Fig. 3(c)]
A reverberating circuitry with feedback capabilities,
possibly resulting in endless loops with a particular
clock frequency; the signal from point A travels to the
junction in B where some delay occurs, and then it is
rerouted via the switching neuron S and arrives again
at its point of origin at A [Fig. 3(d)].
The combination of one neuron plus one synapse is a
simplication introduced for convenience of explanation.
In fact, numerous synapses may connect to a single neuron,
permitting both temporal and spatial summation. These
phenomena illustrate the probabilistic nature of signal
transmission in the nervous system, where crossing a critical threshold for excitation is a matter of the delicate interplay of inputs from numerous sources. Another striking
example is formed by the neuromuscular synapse in which
miniature endplate potentials originate from spontaneous
release of transmitter molecules from presynaptic vesicles,
thus contributing to the likelihood of impulse transport.
The delicate interplay between macroscopic and microscopic structures such as muscle bers, bones, nerves, and
tendons permit the maintenance of posture as well as the
control of movements of humans and most animals. In contrast, some creatures such as spiders use no muscles to extend their legs. Instead, they extend them hydraulically,
and the blood pressures involved may exceed 450 mmHg.
For vertebrates, the so-called motor unit forms the basic entity for skeletal muscle action (Fig. 4). Such a unit
consists of a motor neuron (located in the spinal cord) plus
all muscle bers that it serves; physiologically this implies
that a distinct group of muscle bers contracts simultaneously once a single motor neuron res (i.e., generates
an action potential). A skeletal muscle consists of a multitude of such parallel ber groups, and total force development depends on the number of simultaneously active
motor units. At the level of muscle bers we also observe
the phenomenon of variability, called jitter, which refers to
the variation in the time interval between consecutive discharges of muscle action potentials of two bers from the
same motor unit.
Biomechanics is the branch of biophysics that is concerned with the effects of forces on the motion of a living organism or one or more of its parts, such as a leg.
Forces, torques, fulcrum, and lever action form the basic
ingredients of biomechanics. The center of mass (CM) of
the standing human body is centrally located in the abdomen at the level of the navel. The CM obviously shifts,
for example, when carrying a load with one arm. Performing dynamic exercise causes the body to become unstable.
Therefore, sports such as running, cycling, and skating imply the management of instability.
Because a complete description of a real biological system in terms of its movements is often extremely complex,
on the skin (transdermal route) is slow but has the advantage of being long-acting and rather constant. In particular,
in iontophoresis a drug is delivered transdermally by using
an electric eld to enhance the transport of small, poorly
absorbed ionic drugs across the skin surface with the advantage that only a low dosage of the drug is required (16).
All these considerations make clear that it is useful to develop models that incorporate the various (anatomical and
physiological) compartments that are spatial and chemical
in nature in order to predict the concentration and time
pattern at a target site dependent on the location of introduction of the substance, as well as the particular time
sequence (e.g., bolus versus repeated doses) of administration.
VISION RESEARCH AND EYE MOVEMENT
Figure 4. Schematic representation of a motor unit, consisting
of a motor neuron plus all connected muscle bers and intermediate structures such as collaterals of the efferent nerve ber and
neuromuscular synapses that functionally connect the nerve ber
endings with the corresponding muscle bers.
many researchers employ reduced models that nevertheless provide enough insight into the problem under investigation. A simple model of a control system encountered
in locomotion will be presented. In general, the current
contraction state of a particular muscle is reported to the
motor center (feedback loop), but the anticipated action is
also considered (feedforward control). Furthermore, information is routed to the antagonistic muscle, enabling a
smooth and balanced movement pattern. A defective coordination of movement (loosely referred to as clumsiness)
actually covers a spectrum of abnormalities. Thus, there
may be a number of potential defects in the control systems
that would involve substantial consequences. For example,
dysdiadokokinesis is a defect in which an individual lacks
the ability to perform rapid movements of both hands in
unison.
CLOSED LOOP DRUG DELIVERY
The eld of pharmacokinetics concerns the analysis of factors that affect absorption, distribution, and elimination of
drugs in the body. In contrast, pharmacodynamics refers
to interactions at the receptor site. Drugs may primarily
act locally (the so-called topical type such as eye drops or
inhaled aerosols) or they may more or less simultaneously
(intentionally or not) inuence many organs in the whole
body (i.e., systemic type). Obviously, temporal as well as
spatial considerations are relevant in pharmacokinetics.
The route of administration mainly determines where a
pharmacological substance exerts its primary action. Because of the frequently occurring side effects of almost all
drugs, it is also important to gather information about the
distribution and uptake of a drug at a site that is not the
primary target. Apart from the site of administration, it is
important to have some estimate about the optimal speed
of delivery at the desired location. Injection into the bloodstream is fast and potentially transient, while application
Various aspects of eye movement have been extensively

described in PHYSIOLOGICAL MODELS, DEVELOPMENT by T.
Bahill. Therefore, we focus here on our main theme, namely
uctuations, instability, and variability as observed in the
oculomotor control system. Basic concepts of the neuromuscular system controlling movement of hands and limbs
also apply to eye tracking. The task of permanently centering a moving image on the fovea is realized by a multilevel
system involving reex loops, volitional control, and predictors. Saccades are rapid successions of conjugate steps
of eye rotation that permit the positioning of a target image
onto the fovea. The eyes voluntarily move from one xation
point to another, as can be observed during reading. They
are preceded by a reaction time of about 0.20 s and follow
a typical course of rapid acceleration and subsequent deceleration with occasionally a small overshoot. Accommodation (focusing) is driven by a blur of the target image on
the retina. Smooth pursuit involves a slow but continuous
following movement needed to perform a smooth tracking
task. Disjunctive movements of both eyes permit vergence
(binocular xation system resulting in convergent or divergent movement). Accommodation and vergence form an interactive dual-feedback system. Physiological nystagmus
(min eye movements) are repetitive fast and slow movements that adds to the visual acuity by preventive bleaching while shifting to different receptors. Acceleration of the
body requires the vestibulo-ocular reex to become operational. It is believed that (white) noise enhances stability (as in HRV), while uctuations affect accommodation.
Hung described a nonlinear static model, containing the
depth of eld (as a dead-space operator for accommodation) as well as Panums (17) fusional area (as a dead-space
operator for vergence), and found that these operators are
able to account for the discrepancy between results using
the phoria and xation disparity methods (18).
RESPIRATORY CONTROL
During inspiration, air enters the lungs from the nasal passages (conchae) or the mouth via a branching system of
tubes ending in numerous small but highly elastic hollow
structures (alveoli). These elastic elements are in contact
with small blood vessels and are therefore the sites of gas
Figure 5. There is increased local obstruction at point C in case

of asthma, whereas in emphysema the trajectory between C and
the alveoli shows a permanent decrease of diameter.
exchange. Expiration implies transport in the opposite direction, from the lungs towards the nal tube (the windpipe or trachea). Expansion of the lungs is normally realized by muscular activity of both the diaphragm and the
intercostal muscles (19).
The dynamics of respiration are commonly described in
terms of a pressurevolume relationship (to study restrictive diseases such as interstitial brosis and pulmonary
edema) and derived quantities such as ow ( Q, to study
obstructive diseases such as lung emphysema) and compliance (the ratio of volume changes resulting from variations in pressure). In contrast to emphysema, asthma is a
reversible obstructive airway disease, because it is caused
by an increase in smooth muscle tone in the large bronchi.
Figure 5 illustrates the nature of these abnormalities in a
lumped parameter model, consisting of the thoracic wall (
T), an overall spherical elastic element with alveolar pressure ( Palv ) inside, atmospheric pressure ( Patm ), pleural
pressure ( Ppl ), total airway resistance ( Raw ), and C is the
usual point of collapse of the airways acting as a Starling
resistor (i.e., a collapsible tube affected by the pressure of
its surroundings). It can be derived that
The stimulus to breathing is controlled by a particular

area in the brain stem called the respiratory center. The
rhythmical contraction of the diaphragm and intercostal
muscles determine the inspiration sequences, that is, the
brain generates alternating cycles of ring and quiescence
in the responsible motor neurons.
CONTROL OF HEART AND CIRCULATION
The tracing of an electrocardiogram (ECG) is one of the
most popular icons displayed whenever referring to the
eld of medicine. From a clinical point of view, the measurement of the ECG is attractive because it involves a
noninvasive procedure that can be performed quickly using
either portable or xed equipment. The cardiac system can
best be described as a mechanical pump system that is triggered and synchromized by electrical signals. Because of its
noninvasive nature, the ECG is helpful in providing some
preliminary information on cardiac rhythm, electrical conduction, and its disturbances. Figure 6 shows the preferential pathways for conduction of the excitation wave, which
closely follows functional anatomy of the heart, that is,
starting at the sinus node in the right atrium, traveling
towards the apex of the ventricles, followed by a spread towards the outow tract. This route implies that the blood
Figure 6. A cross-sectional view of the heart, showing the two

ventricles and both atria, along with the conduction system for the
electrical impulse running from the sinus node towards the apex
and then upwards along the muscular walls of the ventricles.
is expelled from the ventricles similar to the optimal direction for squeezing out the contents of a tube of toothpaste,
that is, running from the very bottom up towards the opening at the opposite site.
The rhythm of the healthy heart is not constant but exhibits a certain degree of HRV (20). Many studies have employed advanced mathematical techniques to analyze the
cardiac rhythm and estimate the relative contribution of
the sympathetic and parasympathetic drives, respectively
(21, 22). But the perpetual change of the rhythm has also
profound mechanical consequences: An increase in cycle
length implies facilitation of ejection both by increased lling (i.e., elevated preload) and reduced opposing pressure
at the time of the valve opening (i.e., lower afterload), while
during the next beat with a shorter interval the opposite
applies. In other words, impeded and facilitated beats appear to alternate, thus possibly improving stability of the
complete circulatory system.
The left ventricle has often been modeled as a sphere
or prolate ellipsoid, but neither geometry conforms with
reality. Independent of geometrical assumptions, Beringer
and Kerkhof (23) have shown that a fairly linear relationship exists between end-systolic volume (ESV) and enddiastolic volume (EDV). This notion has been veried in
human patients and also in their experimental investigations when studying the volume regulation in physiologically operating chronically instrumented dogs (24). The
regression coefcients appear to be characteristic of ventricular volume regulation and are sensitive to inotropic
intervention (i.e., adrenergic agonists and blockers). This
relationship implies that a clinically important cardiac performance indicator, namely ejection fraction (EF) is inversely related to ESV (25). Furthermore, using the Sugamodel, myocardial oxygen consumption can potentially be
predicted from a single noninvasive determination of ESV
Quite similar to the description of pulmonary dynamics,
the dynamics of heart and vessels are also characterized by
pressurevolume ( PV) relationships. Flow ( Q) equals the
time derivative of a changing volume. Another frequently
employed derived index is elastance ( E), the reciprocal of
compliance, which is dened as the material property that

enables resisting deformation. In analytical form,
and nonlinear approaches. They are mentioned here to let

the reader become aware of the existence of these tools.
Fractal analysis
where V0 is the unstressed volume. This is analogous to

Hookes law (lengthtension relationship), which states
that tension (i.e., force per unit area) equals Youngs modulus times the relative change in length.
Spontaneous oscillations of arteries have also been observed, with a period ranging from 45 s to 60 s for the radial artery diameter under resting conditions, resulting in
an almost twofold change in distensibility (26). A general
model of the properties of a muscular blood vessel at varying levels of contraction predicts that the vessel becomes
unstable at high levels of contraction (27). A lumped model
of the arterial tree can also be considered, similar to the
lumped model for the lungs as presented before. Imagine
that the arterial bed with all its branches is replaced by a
single vessel with a diameter similar to the human aorta
and encompassing an identical hemodynamic resistance (
R). How long would this hypothetical noncollapsible aorta
be? From Poiseuilles law we know that R = 8l/r4 . If the
radius r = 1 cm, the blood viscosity = 3 cP, and R is the
mean driving pressure divided by mean ow, that is, 100/5
mmHgmin/L, then the length l turns out to be approximately 200 m.
Obviously, the real circulation consists of various vascular beds that supply blood to the various organs such as
brains, kidneys, skeletal muscles, and intestines. One powerful regulatory mechanism used to accommodate varying
needs in a single organ is based on the principle of redistribution of ow within the body by local vasodilation and
constriction of arterioles. Another phenomenon, called autoregulation, is based on vasoactivity that maintains the
ow at a particular level even during acute changes in pressure.
Arterial blood pressure is well controlled by several systems. Figure 7 presents a survey with emphasis on time
course and gain for each subsystem involved (28). Hypertension can then be interpreted as a disorder in which
the set point is altered while the control systems continue
to regulate towards an erroneously determined set point.
This view also explains the chronic nature of a condition in
which blood pressure even at rest is elevated. Going back to
Fig. 1 this abnormality would be equivalent to a situation
in which the heater feedback system continues to maintain
a given (wrong) set point.
FRACTAL AND WAVELET ANALYSIS OF
CARDIOVASCULAR PARAMETERS
The design and function of the cardiovascular system is
commonly described in terms of a muscular pump and a
network of branching vessels. This traditionally implies an
interpretation in terms of the laws of physics (12, 13). More
recently, analysis using the fractal properties of physiological observations has been employed. This appears to have
a great potential for the study of physiology at scales of
resolution ranging from the microcirculation to the entire
organism. Other new techniques include wavelet analysis
The values of the measured properties of many physiological systems seem random. This view originates from the
tools that have been used when analyzing the details of organs. Nowadays there are new methods to analyze seemingly random experimental data, such as a time-series approach. These methods use many of the properties and
ideas of fractal analysis. The mathematician Mandelbrot
chose the word fractal to denote objects or processes with
multiple-scale properties, that is, the ever ner subdivisions of objects or processes as they are viewed at progressively higher magnication. Until recently, our ability to
understand many physiological systems was hampered by
our failure to appreciate their fractal properties and to interpret scale-free structures (29).
Wavelet analysis
Fourier analysis (i.e., decomposition of a signal into sinusoidal waveforms) is not suitable for signals with discontinuities. This type of computational headache can be successfully treated by wavelet (ondolettes) analysis, a powerful technique developed by Mallat and Meyer in France
(30). Later, Daubechies (at Princeton) discovered the dual
family, representing the high-frequency range and the
smooth parts (low frequencies). Wavelet analysis owes its
efciency to the fast pyramid algorithm, reducing the number of calculations by down-sampling operations, that is,
steps that remove every other sample at each operation
(halving the data each time). The method has vital applications in compression of signals and images, in addition to
noise reduction (a common problem in biomedical recording, in particular in magnetic resonance imaging). The current popularity of wavelet analysis may be exemplied by
the commercially available software Wavelets for Kids
developed at Duke University.
Nonlinear analysis
Heart rate (HR) and blood pressure (BP) are controlled
by several central nervous system oscillators and different control loops (see Fig. 7). Interactions among these
units may induce irregular time courses in the processes
they govern, but the underlying subprocesses also include
deterministic behavior. These irregular time courses can
be more accurately characterized by dynamic nonlinear
analysis rather than by linear time series. Typically, onedimensional time series data are transformed into multidimensional phase-space plots, thus lling selected regions.
Two essential aspects of such plots include:
1. The correlation dimension, which is a measure of the
complexity of the process studied, that is, the distribution of points in the phase space
2. The Lyapunov exponent, a measure of the predictability of the process, quantifying the exponential
divergence of initially closed state-space trajectories.
Figure 7. Gain of the various arterial pressure control mechanisms as a function of time after the onset of a disturbance.
(Reproduced from A. C. Guyton, Human Physiology and Mechanisms of Disease, 5th ed. Philadelphia: W.B. Saunders, (1992,
with permission.)
Signicant nonlinear dynamics of HR uctuations were

found in rabbits and piglets. Such studies may improve the
understanding of underlying physiological processes.
ENDOCRINOLOGY
Cells communicate by means of electrical coupling (e.g.,
direct contact interaction or neurocontrollers) and chemical interaction (i.e., endocrine control often for the longdistance type). These communication networks regulate vital processes such as growth, differentiation, and
metabolism of tissues and organs in all living systems. The
nervous system and the endocrine system are linked by the
integrating function of the hypothalamus, a specic region
of the brain.
The specic endocrine contribution to regulation forms
the subject of this section. It involves signal generation, propagation, recognition, signal transduction, and response to a particular stimulus. In endocrine signaling, the
pertinent cells release substances called hormones. The
main endocrine glands are the hypothalamus, the anterior
and posterior pituitary, the thyroid, parathyroid, adrenal
cortex and medulla, pancreas, and the gonads (i.e., ovaries
or testes). Hormones can be chemically classied into three
groups: steroid hormones (e.g., estrogen and progesterone),
peptide or protein hormones (such as insulin, prolactin,
and the various releasing hormones), and a category derived from the amino acid tyrosine (i.e., thyroxine and triidothyronine). These substances are usually transported
via the bloodstream. Therefore they are distributed within
the whole body. Lifetime in plasma may range from seconds to days. In order to ensure a selective action at a
special site, target cells exhibit sensitivity for a particular hormone and only in these cells will an appropriate
response be induced. The ability of target cells to exclusively respond to specic hormones depends on the presence on the membrane of receptor proteins that are unique
for a particular hormone. The action may take place on the
cell surface (with or without an intracellular mediator
the second messengers, substances that serve as the relay

from the plasma membrane to the biochemical machinery
inside the cell), or only within the interior of the cell, at the
nucleus level.
Besides negative and positive feedback circuits, the endocrine system features rhythmic control by pulsatile release of certain hormones leading to circadian variations.
Also, endocrine control can be pharmacologically modied
with the use of synthetic hormones. The modern eld of
molecular biology is also concerned in part with many
aspects of the physiology of the endocrine system at the
(sub)cellular level.
BIBLIOGRAPHY
1. B. W. Hyndman, R. I. Kitney B. M. Sayers Spontaneous
rhythms in physiological control systems, Nature, 233:
339341, 1971.
2. J. A. Palazzolo, F. G. Estafanous, P. A. Murray, Entropy measures of heart rate variation in conscious dogs, Am. J. Physiol.,
274: H1099H1105, 1998.
3. V. Shusterman, K. P. Anderson, O. Barnea, Spontaneous skin
temperature oscillations in normal human subjects, Am. J.
Physiol., 42: R1173R1181, 1997.
4. K. B. Campbell et al., Left ventricular pressure response to
small-amplitude sinusoidal volume changes in isolated rabbit
heart, Am. J. Physiol., 273: H2044H2061, 1997.
5. P. Schiereck H. B. K. Boom, Left ventricular force-velocity relations measured from quick volume changes, Pugers
Arch.,
379: 251258, 1979.
6. R. L. Kirby et al., Coupling of cardiac and locomotor rhythms,
J. Appl. Physiol., 66: 323329, 1989.
7. M. A. Radermecker et al., Biomechanical characteristics of unconditioned and conditioned latissimus dorsi muscles used for
cardiocirculatory assistance, Cardiovasc. Surg., 5: 516525,
1997.
8. C. Bernard, An Introduction to the Study of Experimental
Medicine, New York: Dover, 1957, p. 135.
9. R. Fogliardi, M. di Donfrancesco, R. Burattini, Comparison of

linear and nonlinear formulations of the three-element windkessel model, Am. J. Physiol., 271: H26612668, 1996.
10. N. Westerhof G. Elzinga, P. Sipkema, Articial arterial system
for pumping hearts, J. Appl. Physiol., 31: 776780, 1971.
11. P.L.M. Kerkhof et al., Design of an artical afterload; the
Hofkessel, Automedica, 7: 7782, 1986.
12. A. Noordergraaf, Circulatory System Dynamics, New York:
Academic 1978.
13. J. K.-J. Li Arterial System Dynamics, 2nd ed., New York: New
York University Press, 1999.
14. H. E. Derksen, Axon membrane voltage uctuations, Acta
Physiol. Pharmacol. Neerl., 13: 373466, 1965.
15. J. Monod, Le Hasard et al Necessite, Paris: Editions du Seuil,
1970.
16. C. A. Zakzewski et al., Iontophoretically enhanced transdermal delivery of an angiotensin-converting enzyme inhibitor
in induced hypertensive rabbits, Cardiovasc. Drugs Ther., 6:
589595, 1992.
17. P. L. Panum, Physiologische Untersuchungen uber
das Sehen
mit zwei Augen, Kiel: Schwersche-Buchhandlung, 1858.
18. G. K. Hung, Quantitative analysis of the accommodative convergence to accommodative ratio: Linear and nonlinear static
models. IEEE Trans. Biomed. Eng., 44: 306316, 1997.
19. D. L. Fry R. E. Hyatt, Pulmonary mechanics, Am. J. Med., 29:
672689, 1960.
20. M. L. Appel et al., Beat to beat variability in cardiovascular variables: Noise or music? J. Am. Coll. Cardiol., 14:
11391148, 1989.
21. M. Malik A. J. Camm (eds.), Heart Rate Variability, Armonk,
NY: Futura, 1995.
22. M. Pagani et al., Power spectral analysis of heart rate and
arterial pressure variabilities as a marker of sympathovagal
interaction in man and conscious dog, Cir. Res., 59: 178193,
1986.
23. J. Y. Beringer P.L.M. Kerkhof, A unifying representation of
ventricular volumetric indexes, IEEE Trans. Biomed. Eng., 45:
365371, 1998.
24. P. L. M. Kerkhof, Beat-to-beat analysis of high-delity signals
obtained from the left ventricle and aorta in chronically instrumented dogs, Automedica, 7: 8390, 1986.
25. P. L. M. Kerkhof, Importance of end-systolic volume for the
evaluation of cardiac pump performance, in E. I. Chazov, V. N.
Smirnov, and R. G. Oganov (eds.), Cardiology, An International
Perspective, New York: Plenum, 1984, pp. 13391352.
26. D. Hayoz et al., Spontaneous diameter oscillations of the radial
artery in humans, Am. J. Physiol., 264: H2080H2084, 1993.
27. C. M. Quick et al., Unstable radii in muscular blood vessels,
Am. J. Physiol., 271: H2669H2676, 1996.
28. A. C. Guyton, Human Physiology and Mechanisms of Disease,
5th ed., Philadelphia: Saunders, 1992.
29. J.B. Bassingthwaighte, L.S. Liebovitch, B. J. West, Fractal
Physiology, New York: Oxford University Press, 1994.
30. A. Bruce et al., Wavelet Analysis, IEEE Spectrum, 33 (10):
2635, 1996.
READING LIST
31. F. W. Campbell, J. G. Robson, G. Westheimer, Fluctuations of
accommodation under steady viewing conditions, J. Physiol.,
145: 579594, 1959.
32. T. Coleman, Mathematical analysis of cardiovascular function,
33. H. Collewijn, F. van der Mark, T. C. Jansen, Precise recording
of human eye movements, Vision Res., 15: 447450, 1975.
34. J. Darnell, H. Lodish, D. Baltimore, Molecular Cell Biology,
New York: Scientic American Books, 1986.
35. P. H. Forsham (ed.), The Ciba Collection of Medical Illustrations, Vol. 4: Endocrine System, Summit, NJ: CIBA/Novartis,
1981.
36. P. A. Insel, Adrenergic receptorsevolving concepts and clinical implications, N. Engl. J. Med., 334: 580585, 1996.
37. J. Jalife (ed.), Mathematical approaches to cardiac arrhythmias, Ann. NY Acad. Sci., 591: 1416, 1990.
38. A. T. Johnson, Biomechanics and Exercise Physiology, New
York: Wiley, 1991.
39. E. R. Kandel, J. H. Schwartz, T. M. Jessell (eds.), Essentials
of Neural Science and Behavior, Norwalk, CT: Appleton and
Lange, 1995.
40. P. L. M. Kerkhof J. Y. Kresh, Cardiovascular function analysis,
in D. N. Ghista (ed.), Clinical Cardiac Assessment, Interventions and Assist Technology, Basel: Karger, 1990, Vol. 7, pp.
125.
41. M. Levy, The cardiac and vascular factors that determine systemic blood ow, Circ. Res., 44: 739747, 1979.
42. E. Maddox, Investigations in the relation between convergence and accommodation of the eyes, J. Anat. Physiol., 20:
475508, 1986.
43. V. Z. Marmarelis, Advanced Methods of Physiological System
Modeling, New York: Plenum, 1994, Vol. III.
44. J. F. Martin, A. M. Schneider, N. T. Smith, Multiple-model
adaptive control of blood pressure using sodium nitroprusside,
45. T. A. McMahon, Muscles, Reexes and Locomotion, Princeton,
NJ: Princeton University Press, 1984.
46. S. Permutt R. Riley, Hemodynamics of collapsible vessels
with tonethe vascular waterfall, J. Appl. Physiol., 18:
924932, 1963.
47. D. A. Robinson, The oculomotor control system, a review, Proc.
IEEE, 56: 10321049, 1968.
48. L. B. Rowell J. T. Shepherd, Handbook of Physiology, New York:
Oxford University Press, 1996, Sec. 12.
49. T. C. Ruch H. D. Patton (eds.), Physiology and Biophysics,
Philadelphia: Saunders, 1965.
50. A. Vander, J. Sherman, D. Luciano, (eds.), Human Physiology,
5th ed., New York: McGraw-Hill, 1990.
51. P. Welling, PharmacokineticsProcess and mathematics,
Washington. DC: American Chemical Society, 1986.
52. M. W. Whittle, Gait Analysis, Oxford: ButterworthHeinemann, 1991.
REVIEWS
53. E. J. Crampin, M. Halstead, P. Hunter, P. Nielsen, D. Noble,
N. Smith and M. Tawhai: Computational physiology and the
physiome project. Exp. Physiol. 89 ( 2004) 126.
54. T.S. Deisboeck and J.Y. Kresh (Eds): Complex Systems Science
in Biomedicine. Springer, 2006.
55. A. A. Cuellar, C. M. Lloyd, P. F. Nielsen, D. P. Bullivant, D.
P. Nickerson, and P. J. Hunter: An Overview of CellML 1.1, a
Biological Model Description Language. Simulation 79 (2003)
740747.
56. G. Goel, I-C. Chou and E.O. Voit: Biological Systems Modeling
and Analysis; a Biomolecular Technique of the Twenty-rst
Century. J. Biomol. Techn. 17 (2006) 252269.
Relevant URLs:
Physiologic modeling: http://nsr.bioeng.washington.edu/
Physiome:
http://www.physiome.org/Models/
and
http://www.ngi-nz.co.nz/applications/physiome.html
PETER L.M. KERKHOF

Vrije Universiteit Medical
Center, Amsterdam
TELECOMMUNICATION EXCHANGES
HISTORICAL BACKGROUND
The telecommunications network including both public

and private parts is one of the most important systems
created by humans in modern civilization. The network
enables people to communicate between continents at all
times of the year and allows thoughts and ideas to be exchanged between families, companies, and governments.
Several phone calls have changed history. The network
also saves lives every day. An example is when mobile subscribers in their cars call emergency numbers when witnessing car accidents.
Due to the networks great importance the requirements
on the components that constitute the system are several.
The more central parts of the network must always be
available. The most important type of such a central component in the classic telecommunication network is the
telecommunication exchange. The exchange enables many
calls to be switched and alternative paths to be taken if a
path in the network has failed, and hence should be very
reliable. This reliability is probably the most important requirement of the exchange. Other requirements are that
the exchange should be able to cope with all signaling standards in a network. It should also be able to coexist with all
equipment in the network, irrespective of its age or fabrication. This imposes great requirements on backward compatibility.
Note that a telecommunication exchange is sometimes
also called a central ofce or switch. There are both public
exchanges and private branch exchanges. An exchange is a
node in a telecommunications network that includes functions for access, control, switching and charging of calls.
These parts may be physically separated and distributed,
also to places outside the central ofce site; and act as
subnodes connected via signaling protocols, for example between access, control and switch resources. A call in a modern context is not just about phone calls but is a general
connectivity service for two or more parties that need access to communication bandwidth over a shorter or longer
period of time. Connection services can be established not
only between peer users but also from peer users to central
servers.
A telecommunications exchange is one of the most complex systems created by humans. State-of-the-art hardware and software technology is used to create very large
systems in terms of the number of connected subscribers
and number of switched calls. The development cost of an
exchange is very large. Several millions of hours of hardware and software development are invested each year. An
exchange can connect hundreds of thousands of subscribers
and switch a million of calls at the busiest hour during the
day. And all of this is in real time, meaning short setup time
and low delay of the voice. The quality of service requirements is thus great. Whether or not the call is made between two different continents, the call has to be switched
through with practically no delays and provide an acceptable speech quality.
The basic service associated with traditional telephony is

the enabling of the bidirectional voice communication dialogue between two persons located at different but xed
places. To enable such voice communication each person
uses a telephone set equipped with a transmitting device
starting with (1) a microphone that translates acoustic energy to electric energy and (2) a receiving device ending
with a loudspeaker that does the reverse transformation.
Each person who has access to telephony also wants to
be able to select whom to connect to and have a dialogue
with. For the establishment of such a connection service,
one must be able to signal from each place to any other selected place that a connection is wanted between one calling and another called person. The traditional call is hence
actually directed to a place where a called person is supposed to be located. To enable the sending and reception
of alert signals the telephone set is equipped with a device
that can generate some form of electrically encoded signal
and a device for reception and transformation of such a
signal to an acoustic signal. For a basic two party dialogue,
two telephone sets are connected, usually electrically via a
pair of wires. In the very beginning the second wire was implemented by an earth connection; this, however, resulted
in a high level of cross-talk when several such unshielded
wires came close to each other.
To start with, telephones were mainly used between two
or a few locations such as between a shop owners ofce,
workshop, and home. A simple n-way switch could be used
to select who to talk to within such a small mesh structured
private net. Later when several families and organizations
in a town had such connections, the communication possibilities were extended. This was done by connecting all
wires in a star structure to a central ofce equipped with
an exchange consisting of a manual switchboard and an
operator. Hence, each telephone in such a local area was
connected via a single wire or a pair of wires to the switchboard. Such a pair of wires is often called a subscriber
(or access) line or loop. The switchboards themselves were
then interconnected via other pairs of wires, called trunks.
To enable voice signals to be carried with less distortion
and over longer distances (by use of inductor coils and ampliers respectively) the trunk lines soon came to use four
wires, one twisted pair in each direction.
The rapid growth of the telephony network during the
early days may well be explained by the fact that the network technology had large similarities with the technology already used for telegraphy. But perhaps most important, the telephone set replaced the need for a telegraph operator knowing Morse code and thus simplied
the humanmachine interface. To establish a call, the calling person had rst to send a ring signal to the operator and
then tell the thus alerted operator whom to be connected
to. If the call was local, the operator then could send a ring
signal to the person being called and ask this person for
permission to set up a connection to the person calling. The
operator could at this point make clear how or by whom the
call was to be paid. A call could then be established by connecting the two pairs of wires to each other. To simplify the
operators work the manual switchboard was designed to
Telecommunication Exchanges
make it simpler for the operator to supervise the lines that

were busy and to handle both the request for, establishment and the ending of a call. If the call was non-local, the
operator rst searched for and selected a free trunk in the
right direction and then called one of the operators at the
receiving end of the trunk (using it as a signaling trunk)
and asked for help to establish the call. Such searching,
routing and forwarding tasks presume that that the operator had some knowledge about the networks topology. For
a long-distance call a chain of operators had to be involved
via a chain of trunk line links before the called person could
be reached and asked for permission to set up the call via
voice trunks along the same path.
To set up a long-distance call, a number of resources
must be free. Those resources mainly were operators and
voice trunks. If some resource was lacking, the operator
could serve the customer by organizing a waiting list or
job queue and then set up calls when resources eventually
became free. The operators had to keep track of each job
by help of a written job record including time stamps for
charging purposes and also provide time supervision of the
ongoing calls to guarantee that resources allocated to a call
were released even if the calling parties forgot to send an
end-of-call alert signal.
When the number of subscribers increased, quite a
large number of operators could work in the same exchange ofce. Different techniques were then used to ensure that they could share the workload and coordinate the
setup of calls between subscribers connected to different
switchboards. Parallel processing of call attempts could be
achieved, for example, by distributing the incoming access
attempt from a subscriber to a non-busy switchboard. Another example is that one single operator could handle the
setup of a local call within a large exchange consisting of
many switchboards by use of multiple point technique
that is, by having switchboards all outgoing lines connected
to all other switchboards in the ofce.
When the demand for telephony increased, the number of operators swelled. Pressure grew to decrease the
costs for human switchboard operators but also the time to
set up calls. Requirements on increased personal integrity
were also a reason to try to automate the call setup procedure. The rst automated switches were based on the
use of electromagnetic coils, effectuating drive mechanisms
and contact points. These electromagnetic devices were, in
turn, controlled by signals generated from the telephone
set by use of a dial. The dial could generate sequences of
current pulses, where the number of pulses corresponded
to a dialed decimal digit. A subscriber was given a telephone number related to a corresponding access line. The
digits used to represent this number controlled the behavior of the switch. The signals were rst decoded and used
directly to control the switch movements in a decadic way
and somewhat later indirectly via registers. The use of registered signals (1) reduced the requirements on timing of
the signals, mechanical precision, and preventive maintenance and (2) increased the exibility by making number
translations possible.
All decisions to be made were not controlled by digits.
For example, a number of trunk lines between two exchanges could be treated as a group and the digits used
merely to select and seize a trunk group going in the

right direction. Then one could do a search (or hunt) for
a free trunk line and, when available, select one within
this group. When no trunk line was free, another possible
route could be tried. However, the number of alternative
routes was limited both due to economic reasons and because the networks soon came to be built more or less hierarchically to simplify the coordination work needed to
establish non-local calls. If no free route could be found,
a blocking situation occurred and the call attempt had to
wait for resources to become free. The operator could then
use different methods to supervise resource release events
and to handle the queue of waiting call attempts.
It is interesting to note that the tasks performed by a
human operator, such as searching for, reservation, monitoring and management of resources and charging records,
were very similar to what the control system of a modern exchange does. There are also interesting analogies
between the call routing, redirection and answering services provided by a human operator and what today can be
provided by the control system of an automated exchange.
The network has evolved from very simple bidirectional
communication links via small private mesh and star networks that as soon as signal regeneration and amplication technology permitted were interconnected via transit
networks to more public, global and hierarchical network
structures. The number of individual trunk lines could also
later be reduced by multiplexing techniques allowing several calls to share a physical line. This evolution started
with frequency division and has evolved via digital time
division toward many different combinations of frequency,
phase, time and code division. Multiplexing creates a logical network layer on top of the physical transmission media
implementing a number of logical lines (or channels) and
hence a more efcient use of each physical line.
Logically, signaling has always been separated from the
voice connection. A trend has been to clarify this by a separation into a signaling trunk network and a voice trunk
network. However, the signaling network may in practice
use reserved logical lines or channels multiplexed on top
of the same physical lines as the voice network.
Voice encoding, with its inuence on transmission and
switching, has evolved from analog; via digital to compressed digital representation and the signal encoding has
evolved from simple current pulses, via frequency-encoded
signals to digital message records.
Manual exchanges handled by human operators were
quite exible and intelligent in many ways since the primitive alert signals simply could be complemented by verbal
communication between subscriber and operator that is,
human to human. Automation required a predened signaling scheme, including not only simple alert signals but
also encoding of the phone number of the called line. To
make the automated exchanges able to provide large exibility and more advanced services, the rst decadic control of an exchange directly from signals representing digits evolved via register mapped control to stored program
control of exchange behavior. The automation possibilities
were increased further by introduction of larger signal alphabets and protocols capable of more than just alert signals and digits.
A great step from a functional point of view was the introduction of radio transmission for cellular coverage and
wireless access to/from mobile terminals. This allows a call
to be directed to a mobile terminal carried by a person
rather than just to a xed place (terminating a wire or
ber). Technically it was not new to reuse the radio spectrum by dividing a geographical area into regions, in this
case called cells, but connecting the cells covered by base
stations to the switched telecommunication network was
new and created many new challenges and opportunities.
Other steps in this direction are: digital subscriber lines;
new call services (often for redirection of calls) sometimes
substituting what a manual operator previously could give
help with; introduction of personal numbers that in conjunction with mobility services can help to make it easier
to reach a specic person rather than a phone terminal and
also can increase competition among operators if the personal phone number becomes a property of a person rather
than an operator; the merger of telecommunication and
data communication networks that enables new multimedia communication services.
A recent such network convergence that will simplify
the evolution of streamed and real-time multimedia services is the sharing of a more and more common infrastructure for xed and mobile voice, data and video services. In
this new setting some of the functions of a telecommunication exchange become obsolete and other more dedicated
nodes become more important. For example, circuit switch
functions may be replaced by packet forwarding while new
forms of access control, authorization, authentication, address location and charging services will be more important. This development along with international standardization is also believed to increase the competition between
different equipment suppliers.
An exchange can include support for almost all functions in a telephone network. However, one can also distinguish specialized network nodes. Early examples of
these specialized nodes are local exchanges and transit
exchanges. Today one can also distinguish other types of
nodes such as network access nodes; switch, call and service control nodes; mobile switching nodes; and network
database nodes (for example, databases for: number translations; handling of subscriber service proles; location information supporting mobility of subscribers, authentication/authorization data, equipment data). Other nodes are
different types of information service, media content, etrade transaction and access right handling servers.
THE EXCHANGE FROM A NETWORK PERSPECTIVE

The main purpose of a telecom network is to interconnect
telephone users, devices and services. Since all users does
not need to use the network all of the time, many network
resources can be shared among several users.
A large exchange is a means to share switching resources by keeping many such resources in one common
pool, thus providing statistical gains. A large exchange also
will reduce network costs by enabling the sharing of network resources and reducing the number of lines, transmission capacity and external signaling needed if split into
smaller exchange nodes or several specialized nodes.

The concept of an exchange can be related to the functions of a network node where many telecommunication
lines are connected. However, the traditional functions associated with an exchange are these days sometimes distributed and located also to other node types in the network.
Functional and Structural Perspectives on Networks
A practical approach is to look at networks from the operators perspective. From this point of view a telecommunications network can be divided into four major networks: an
access network, a transport network, a signaling and control network, and a management network. In more recently
developed networks one often complements this view with
service and media networks.
However, for different purposes and reasons, one may
also talk about other types of networks that are related to
each other in different ways. A reason to do this is that one
often needs to look at networks both from functional and
from structural perspectives.
In abstract functional perspectives, one views the network from the external environment as if it were a black
box and focuses mainly on the services that the network
provides. Functionally, a network can be dened as a set of
points at which a set of functions are provided and at which
certain properties of each such function can be measured.
A network function at one level may be implemented by
using a function or a set of functions at the adjacent lower
level (see Fig. 1).
In structural perspectives, one takes a closer, more internal, detailed and structural (white box) view and focuses
on matters related to structurally and physically measurable properties, including how the network functions can
be partitioned, distributed in space, and allocated to network nodes. Structurally, one can hence dene a network
as a set of network nodes, each with a set of allocated node
functions, placed and interconnected via logical interconnection points in this way enabling the node functions to
cooperate so as to implement the network functions (or network services) to be provided.
Using the functional perspective of networks, one can
distinguish some network concepts related to telecommunication exchanges. For example: an access network provides access related functions at a set of network access
points; a service network provides service (e.g., call) related functions at a set of service points; a mobile network
provides location and mobility related functions at a set
of location points; a signaling network provides signaling
functions at a set of signaling points; a connection network provides connection or switch related functions at a
set of connection points; a management network provides
management functions at a set of management points; a
transport network provides transport functions at a set of
transport points; and so on a media network provides information functions at a set of media content provisioning
points.
The functions at the network points mentioned above
are functionally separated but can for economical or other
reasons be colocated in the same physical node or site. The
Figure 1. Network using functions from a lower-level network.
networks are related to and use each other in many ways.

For example: the signaling network is used to enable functional inter-work between the other networks; a subscriber
can, by sending signals via the access network, obtain access to functions provided by the service network that establishes these services using (by sending of signals to) the
connection network(s); the transport network is used by all
the other networks as a bearer for transport of raw data.
Purpose of Networks and Network Design
As discussed above, there are several types of networks
seen from a functional point of view. A purpose of this partitioning or layering into several functional networks is
to help us to build network architecture with well-dened
functional areas and interfaces between these that will
simplify extensions and modications to be done. Another
purpose is to support the building of cost-effective networks, both from a purchase and from a life-cycle point
of view, with low operation and maintenance costs.
The end users of a telephone network have a more holistic view: They want to be able to contact each other so that
they can talk when they are physically far away. Hence, the
main purpose of a telephone network is to establish communication between people. In a mobile multimedia network this interest extends to having mobile access wherever the user is located also to media servers including
ordinary databases and le systems as well as audio and
video stream content.
End users of telephony not only want to talk, they also
want to talk inexpensively and get a good quality of service.
This requires efcient use of resources and a good network
design. The costs to operate a network depend much on how
the subscribers are distributed in the geographical area
that the network is aimed to cover. Charging policy, trafc
intensity, and trafc patterns such as length and locality of
calls are other factors that all will inuence the design of
the network. Most networks are built in a more or less hierarchical structure to collect, concentrate, transmit, and
connect the trafc such that higher-capacity transmission
and switching equipment can be utilized. This reduces the
costs for transmission and switching since many users can
then use the equipment more efciently and with less risk
for blocking due to statistical multiplexing effects. The degree of concentration that can be utilized is inuenced by
the trafc intensity and the desired service quality level.
Another important cost factor is that the total physical
length of the transmission path can be reduced in this way,
and this is important since the cost for digging ditches and
use of pipes is a large contributor to the overall network

costs. Microwave links are used to reduce cost where it is
too expensive to dig ditches. For high trafc routes, direct
routes are often used, especially if the distances are short,
such as in metropolitan areas. Alternative routes are also
used to improve reliability, but these increase the cost and
for that reason are usually not used in the access network.
Using the structural perspective of networks, one can
for a specic area identify a distribution of subscribers that
must be interconnected in a reasonably efcient wayfor
example, in a star- or ring-structured physical access network. One can then allocate a number of network nodes of
different basic types that are needed, both to deal with expected maximum trafc loads and to interconnect these in
a way that keeps transmission costs close to a minimum.
Typical nodes in xed and mobile telecommunication networks are illustrated in Fig. 2.
We can now go back to the external functional network
perspective and hence to the different functions and examine how these are distributed in the physical network.
This distribution can be guided by different principles. One
principle that has to do with costs is that simple and cheap
functions that are used often and for long periods of time
should be located close to the subscribers. Complex expensive functions used seldom and for short periods of time
should be placed more centrally in the network and thus
be shared by many users. However, with todays technology and using mainstream components, complex must not
mean expensive and hence a complex function if implemented as an integrated circuit can often be placed close to
the subscribers, for example voice coders. Furthermore, the
effects of distribution on reliability, signaling, and maintenance must also be considered when deciding on function
distribution over network nodes.
From the above discussion one can see that there is no
absolute denition of how networks shall be implemented
or what a network node shall contain. Rather, there are
several possible congurations that can fulll the requirements, because not only does the distribution of subscribers
as well as their quality of service (QoS) requirements and
trafc patterns differ signicantly but also implementation costs for different types of solutions differ and change
over time.
Going back to and analyzing what an exchange is, one
can see that many of the functions carried out in the different logical networks have traditionally been placed in
an exchange node and can still be located in such a node.
On the other hand, these functions and hence the exchange
Figure 2. Typical nodes in a telecom network.
can also be more or less distributed and placed in more specialized networks and nodes. With well-dened functional
areas and interfaces between these it is possible to congure networks and network nodes in many ways. However,
for a node or cluster of nodes to be regarded as an exchange
it must have some basic call access, control, switching and
connection handling capability.
Routing and Switching Techniques
The task of nding a path from source to destination is
called routing. The task to follow such a route from source
to destination guided by end-to-end address information is
called route selection or forwarding. Each node that participates in this task does not know the whole path but must
be able to analyze the destination address and nd out in
which direction the path should go and also nd such a
path with a trunk that is free to use.
To forward information (a bit, a voice sample, a message,
a cell, a packet or a frame) from a xed or mobile input terminal (address, channel, line or trunk) to a selected output
terminal (address, channel, line or trunk) of an exchange
where the route selection or forwarding is controlled by
local address information (based on the selected path) is
called switching.
The telephone exchange will use network topology or
routing information to prepare a connection. Hence, routing precedes the establishment of a full circuit switched
connection and is done only once while switching is done
for each voice sample (using traditional circuit switching)

during a connection.
To guarantee that all resources needed to enable communication in real time are available, they can be reserved
by setting up a circuit connection along a route between
the circuit end points. Information (voice and data) to be
exchanged between such end points in a network can then
be sent along the circuit connection that is routed and set
up before the actual information exchange starts and then
released when the information exchange ends. This guarantees that all resources needed to enable voice and data
communication in real time are reserved beforehand.
For historical reasons, circuit connection techniques
are often associated with the synchronous transfer mode
(STM) and pulse code modulation (PCM) using a 125 s
frame rate for transmission of byte (8-bit) encoded voice
samples.
Another technique, called asynchronous transfer mode
(ATM), uses small packets called cells with a given xed
size (53 bytes) divided into a small header (5 bytes) and
payload (48 bytes). The header does in this case not contain
the address but rather path and channel identiers, where
a path is a bundle of channels. Routing is done in a setup
mode where the destination address is sent as payload in
a cell (hence used to carry control information). The trafc
can then be switched at the path or channel level cell by
cell in each node along the paths and channels routed and
set up in advance from sources to destinations.
A third technique usually used for data communication

is to send information in packets of a reasonable size, e.g.
as short Internet Protocol (IP) packets, add addressing information and other control information to the packet, and
send it via a route toward its destination address. Each
node on the way to the destination participates in the forwarding of the packet. For information that needs to be
divided into several packets, forwarding is done for each
packet in each node. To get deterministic packet transport
delays circuit like reserved routes can be established using
multi protocol label switching (MPLS) techniques.
Analog systems such as AMPS and NMT are sometimes

called rst generation mobile networks while DAMPS,
GSM and PDC are referred to as second generation mobile
networks, and are all based on cellular digital technology.
The third generation systems based on code division
multiple access (CDMA) and wideband code division multiple access (WCDMA) IMT-2000 and UMTS are designed
to support variable speed multimedia communication.
Typical network nodes in the switching part of the
PLMN are:
The mobile services switching center (MSC) controls
Network Standards
The classic telecommunications network available all over
the world is the public switched telephone network (PSTN).
It is basically designed to allow the transmission of speech
between two or more users and services related to that.
However, this network is also used for facsimile trafc and
data trafc via modems. Examples of services are alarm
calls and abbreviated dialing; call forwarding and threeparty calls.
Integrated services digital network (ISDN) is an evolution of PSTN that gives the subscribers access to integrated or combined services. ISDN integrates different
telecommunication services into the same network that
transports voice and data in digital form between network access points. The main advantage of the evolution
from analog to digital end-to-end communication is safer
and more exible transfer of information. ISDN provides
a wide range of services divided into bearer services and
teleservices. ISDN is based on the digital telephony network using ordinary two-wire subscriber lines, 24-or 32channel PCM link structures, and Signaling System No. 7.
Integrated access implies that the user has access to both
voice and non-voice services through a single subscriber
line, whereas combined access implies the use of several
subscriber lines. Services include voice, facsimile, and computer connections.
There are two types of user-network accesses dened by
ITU-T:
Basic Rate Access. A basic rate access is used for low

trafc load. It normally includes one 16 kbps signaling
channel (D) and two 64kbps communication channels
(B).
Primary Rate Access (T1/E1). The primary rate access
handles higher trafc loads. It normally includes 23
or 30 communication channels (B) and one signaling
channel (D).
The public land mobile network (PLMN) is used here as
an acronym for a set of networks based on standards such
as advanced mobile phone system (AMPS) and its digital
version (DAMPS), Nordic mobile telephony (NMT), global
system for mobile communication (GSM), and personal digital cellular (PDC) with the primary objective to provide
communication to and from mobile subscribers connected
to the xed network via radio. The radio interface is implemented by the mobile terminals and the base stations. The
base stations are end points in the xed (wired) network.
the radio base stations and the calls within the PLMN
and calls to and from other telephony and data communication networks such as PSTN and ISDN.
The home location register (HLR) contains subscriber
information such as which supplementary services
are activated and information regarding in which
MSC-area the subscriber is currently located.
The visitor location register (VLR) is a database with
information of the locations of the mobile stations in
the area controlled by the MSC. The VLR also fetches
information from the HLR so that the call setup can
be performed without using the HLR each time.
The media gateway (MGW) is a switching/routing/
transferring node in the UMTS transport network,
to facilitate communication between RNCs, between
RNCs and the core network nodes, and between RNCs
and O&M nodes.
The base station controller (BSC), called radio network controller (RNC) in UMTS, coordinates and controls a number of radio resources usually located in
base stations and some interwork functions such as
handover between the cells, covered by the these base
stations.
The Radio Base Station (RBS) is responsible for radio
transmission/reception in one or more cells to/from
the User Equipment (UE).
The intelligent network (IN) is an architecture aimed

at making a telecom network work as one uniform system
where new network services can be easily developed, introduced, and made available in the network from a central
service control point (SCP). IN aims at a logical separation of signaling, call, connection, and transport. The idea
is that local and transit exchanges use number analysis
and ask the SCP to handle the call, unless it is a simple IN
service that can be handled locally. The SCP then executes
a corresponding service script that results in orders to the
exchange regarding how to proceed with the call. For this
communication the intelligent network application protocol (INAP) is used. New services are described as service
scripts or building blocks consisting of functional components and developed by the operator in a service creation
environment.
The telecommunication management network (TMN) is
an architecture and standard portfolio in the area of operation and maintenance. It denes functional areas and
protocols for management in general terms and also more
specic information models for how to monitor and manage
these areas.
The open systems interconnection (OSI) reference
model is a standardized layered model of how computer
systems can be interconnected and interoperate that has
had an inuence on the way signaling networks and protocols are viewed and built. The model denes seven layers:
application, presentation, session, transport, network, link,
and physical.
The telecommunication information networking architecture (TINA) is an international collaboration for dening an open architecture for telecommunication systems. It
focuses on the software architecture. To some extent this
effort can be seen as an attempt to put together some other
standardization efforts such as OSI, IN, and TMN from a
software architecture point of view.
THE FUNCTIONS OF AN EXCHANGE
The telecommunications exchange is a multi application
digital switching product that offers its main services to
its subscribers but also services to the operator of the exchange. One example of a service offered to the subscribers
is telephony calls, and a service offered to the operator
is the ability to charge for such services by registering of
charging data in the exchange.
A telecommunications network offers various services
to the users and the operator. ITU-T has divided these services into two main categories:
A bearer service for transport of speech or data in the

network between the user interfaces. The transport of
speech should be done in real time and without distortion or alteration. The function of the bearer service
corresponds to the OSI levels 13 for transport, routing, and safeguarding of the information through the
network.
A teleservice is a complete communication service
that combines the information transfer of the bearer
service with terminal services, such as information
processing functions. A teleservice corresponds to the
OSI levels 17. Some teleservices are tied to a special bearer service, whereas others can utilize different bearer services. Examples of teleservices are telephony, facsimile, and computer connection.
The bearer services and the teleservices are divided into
basic and supplementary services. Telephony is an example of a basic teleservice, and call waiting is an example of
a supplementary service that gives users additional functionality. In general, the supplementary services provide
additional capabilities that rely on basic services to be
used.
Examples of teleservices are:
Telephony. The normal two-way voice communication

between two users is the most fundamental service.
Facsimile. This teleservice allows the connection of

facsimile machines.
Voice Mail. This service offers the subscribers the possibility to forward calls to a central location in the net-
work. The subscriber can later check the system for

unanswered calls and listen to recorded voice messages.
Basic Telephony
Below is a brief description of the main functions required
by the exchange in order to set up, maintain, and disconnect a basic telephone call between two mobile or xed subscribers.
Subscriber Signaling. In order to set up a call, the calling subscriber alerts the exchange that there is a new call
attempt and then sends the dialed number. For a digital
(mobile or ISDN) access, the alert and the digits are all
sent in one message, in order to save bandwidth resources
and decrease the delay for call setup (see Fig. 3). For an
analog access, the alert is made by lifting the handset, and
the exchange replies by a dial tone. The dialed numbers are
then sent one by one. In both cases, the exchange replies
with a tone to the calling subscriber when the status of the
called subscriber has been checked.
Number Analysis. The A-number (the calling subscriber)
and the B-number (the called subscriber) are analyzed. The
result is then used as input to charging and routing analysis. For analog subscribers, each digit in the B-number
arrives as the subscriber dials, whereas all digits can be
sent at once for digital mobile or xed subscribers.
Subscriber Category and Service Analysis. The exchange
must check whether the calling subscriber has any particular service invoked. Some of the services must be analyzed
early in the call setup, such as blocking of outgoing calls.
The services implemented in software (programs and
data) are executed either in the local exchanges used by
the calling and called subscribers, in other cases in a transit exchange or in a separate exchange that handles intelligent network (IN) services, a service control point (SCP).
In the latter case, the local exchange will need to check
whether to request the service from the SCP or not.
Charging. There are two classic charging methods, pulse
metering and detailed billing. Detailed billing, also called
toll ticketing, enables an operator to specify the characteristics of each call very extensively.
Charging can be divided into two steps: analysis and
output. The result of the analysis is the charging method
(toll ticketing, pulse metering or at rate) and the charging
rate, depending on a number of call data set by the operator.
The output includes formatting of the charging data along
with output on a reliable storage medium, locally or in a
charging and maintenance center.
Routing Analysis. Finding a path from source to destination is called routing and is made mainly on the B-number.
Usually there are two or three (and sometimes up to ten)
alternative routes to select among. The selection of a route
(also called trunk group) is guided by priority and load status information. If the rst route fails or is overloaded, the
next alternative is selected. There are more sophisticated
Figure 3. Subscriber signaling, ISDN.
routing algorithms that dynamically choose a link in order to minimize the congestion in the network. These dynamic routing algorithms can be either local or central;
the local provides results by using data available in its
own exchange such as previous success rates on different
link choices, while the central algorithms collect input data
from other exchanges in the network.
Connection. The connection is the through-connection
of two normally 64 kb/s circuits, one in each direction, in
the hardware devices, and particularly in the switch fabric.
The connection is required to be with limited probability
of blocking, from end-to-end. This means that the switch
fabric must add very low blocking probabilities, in order
to fulll the end-to-end requirements for calls that pass
several transit exchanges. The connection also must be well
synchronized with the rest of the exchange and with the
rest of the network, in order to handle the digital speech
connections properly.
Trunk Signaling. Trunk signaling enables a call to be connected between subscribers in separate exchanges. The basic data in all trunk signaling systems are alert messages
that a call is to be connected or disconnected, along with
routing information, mainly the relevant parts of the dialed
digits. Modern signaling systems can transmit all types of
datafor instance, in order to support detailed billing, advanced network services, and transparent user data.
Early signaling was made on the same line where the
speech was transmitted, rst by decadic pulses and later
by tones of different frequencies. A still common such inband signaling system is multi-frequency signaling, where
a combination of two tones is sent to a tone-receiver in

the exchange. Modern signaling is based on digital message passing. The globally dominant signaling system is
the Signaling System No. 7.
Subscriber Services
When an exchange receives digital data and is computercontrolled, almost any communication service can be performed, and a large number have also evolved. The service
software is located in the terminals, in the ordinary exchanges, service control points and network databases. The
most cost-efcient location depends on the type of service.
For some services where the logic is local (such as abbreviated number, also called speed dialing), it is most efcient
to store the translation between abbreviated number and
real number in the calling terminal. For more intelligent
network services such as virtual private networks, freephones, or universal personal numbers, it is preferable to
locate the service logic in a network node, in either (1) the
ordinary exchanges such as the mobile switching center or
local exchange or (2) a network database such as the SCP
or HLR. The advantage of central service control is that the
introduction of new services and features is simplied. In
addition, some features require consistent data for the entire network, such as the information in the HLR regarding
where a called mobile subscriber is located.
More powerful protocols enable more advanced services
to be implemented in the network. At the same time, there
are an increased number of services that are implemented
in the terminals, and related data is sent transparently
through the network between the end users. As an example, ISDN has spread slowly while data trafc over the
telecom network has increased much more rapidly, where

the services are executed in the end-users computers.
A few common telephony subscriber services, implemented in an exchange, are as follows:
Freephone. The call is free of charge and paid by the

called party. Often the freephone service can be directed to various physical subscribers depending on
time, date, and trafc.
Conference Call. More than two parties take part in
a call.
Transfer Services. The call is transferred to another
telephone immediately or when the called number is
busy or not replying.
Universal Personal Number. One phone number is
used regardless of which mobile or xed physical connection a person is using.
Call Completion Services. When the called party is no
longer busy or nonreplying, the call is reinitiated.
Virtual Private Network. A group of subscribers, for
instance a corporation, form a private network with
their own charging and telephone numbers.
dover between channels in the same cell.

Location Update. When the mobile phone is turned on
and running in idle mode, it listens to control messages
indicating which area the closest base station sending control messages belongs to. When a border is passed it sends
a message to the mobile switching center (MSC), indicating that the subscriber has moved to another location. The
information is stored in the MSC, in a visitor location register (VLR), and it is also stored in the home location register
(HLR) if the area is handled by a new MSC.
Paging. Locating a mobile subscriber within the network is called paging. This is done by requesting the HLR
in which VLR/MSC and location area where the subscriber
is located, and then sending a paging message on the page
channel to the cells in that area.
Roaming. When subscribers move to another operators
network than their own, the network can page the subscriber and then set up a call to their new location.
Operation and Maintenance
These services are also effective over a network, not only

when both parties reside in one exchange.
Cellular Mobile Telephony
Cellular mobile telephony differs from basic telephony
since the subscribers can move freely within areas covered
by the radio access network. As a result, the exchanges
in a cellular system must keep track of where the subscribers are located and nd free radio channels to use for
new calls and during calls since these are shared among
all subscribers in an area.
The radio frequency spectrum made available for mobile telephony is a scarce resource that is reused by dividing the space in small areas called cells. Usually the
frequency spectrum is also divided into frequency bands,
and these bands can in turn be time or code divided into
channels. Cells that are not too close to each other can, due
to the use of limited power levels, share frequency bands
and channels without disturbing each other.
Handover. Handover means to change or switch connection from one cell to another with better radio transmission
quality during an ongoing call. The handover decision is
based on measurements of received signal quality in up
and down links. Handover can be made several times during a call. This and the fact that handover decisions require
the collection and analysis of measurement data contribute
to making cellular mobile telephony quite processing resource consuming compared to xed telephony.
Channel Allocation. Channel allocation, aims at nding
free frequency, time and/or code divided channels within a
cell and then allocating such free channels to calls. The allocation logic gives an ongoing call higher priority than a new
call. Channel allocation is closely related to the handover
function and especially intercell handover, which does han-
A modern telecommunications exchange should offer operation and maintenance functions that guarantee a high
quality of service to the operators and the subscribers. Operation is the normal everyday running of the exchange.
This includes activities to adapt the exchange to continuously changing demands. Examples of operational activities are:

Connection and disconnection of subscribers

Change of subscriber data
Collection of charging data
Collection of statistics
Maintenance is the prevention, detection, localization,

and correction of faults. The faults can be detected automatically by the exchange or reported to the operator by
the subscribers or other exchanges in the network. Examples of maintenance activities are:
Fault detection, testing, and repair of exchange hardware, for example, trunk lines or subscriber lines
Fault detection, auditing, recovery and correction of

exchange software and data
Checking of disturbance indicators in various parts of

the exchange such as the power system or the control
system
In a modern telecommunications exchange there are
several approaches to maintenance. One is preventive
maintenance that involves a set of routine tasks to check
for faults before they occur and requires a high level of
effort to achieve a certain grade of service. Another is corrective maintenance where faults are dealt with as they
occur. This requires a more selective and limited effort but
may result in a less consistent quality of service. The best
10
is to have a balance between the preventive and corrective

maintenance.
Statistics are used to supervise trafc and performance
in the network and to recongure an exchange to handle
more or less trafc. Particularly for the conguration of
location areas and cells within a mobile network, large
amounts of trafc data are used as facts to support corrective actions.
All operation and maintenance activities should have
minimal impact on the trafc handling of the exchange.
The ideal situation is an exchange where every subscriber
can make a call at any time no matter what happens to the
system. In order to achieve this, the system should be robust to operator errors and allow the performance of maintenance and software and hardware upgrades without affecting the execution of trafc events.
An exchange should be able to be operated and maintained remotely. The ability to access exchanges remotely
using an operation and support system (OSS) ensures a
high level of service to subscribers and low costs for the operator. A centralized operation also contributes to a more
reliable operation of the exchange and a reduction in personnel.
ARCHITECTURE OF COMPUTER CONTROLLED
EXCHANGES
The architecture of computer controlled exchanges is inuenced to a large extent by the architecture and technology
of both the switching system and the (central) control system and how they are related via decentralized (regional)
control systems (see Fig. 4).
A modular architecture can lower the costs of system
handling and make it easier to adapt the system to the
changing world of telecommunications. In a truly modular
system each module is fully decoupled and independent of
the internal structure of other modules. There are different
forms of modularity, for example:
Application modularity. Make it easier to combine

several larger applications in one node.
Functional modularity. The system dened in

terms of functions rather than implementation units.
Functions should be possible to add, delete, and
change without disturbing the operation of the system.
Software modularity. The software modules should
be programmed independently of each other, and they
should interact only through dened interfaces and
protocols. In this way, new or changed modules can be
added without changing existing software.
Hardware modularity. Supports that new hardware can be added or changed without affecting other
parts of the exchange.
On the highest level the system architecture of the
exchange can be divided into various application modules in analogy to how telecommunications nodes interact and communicate, using protocols enabling modules to
be added or changed without affecting the other modules.
Typically the implementation of a telecommunications exchange can be divided into:
Application modules. These implement various

telecommunication applications much like virtual
nodes using standardized interfaces to other application modules. Application modules act as clients to
resource modules.
Resource modules. These modules coordinate the
use of common resources available to applications
by means of well-dened interfaces to the users. Resource modules act as servers to application modules.
Control modules. These modules are responsible for
the operating system functions, input-out-put functions, basic call service functions, and so on.
Application Modules
The application modules implement various telecommunication applications and have standardized interfaces to
resource modules. In general an application consists of access and services. Examples of application modules are:

Analog access module

Digital access module
Mobile access module
PSTN user services module
ISDN user services module
MSC user service module
Home location register (HLR) module
Resource Modules
The resource modules typically handle and coordinate
the use of common resources and may contain both software and hardware. The most important part is the group
switch. Trunks and remote and central subscriber switches
(RSS and CSS respectively) are connected to the group
switch. The trunks are used to connect the switch to other
switches, to data networks, mobile base stations, etc. The
subscriber switch handles the subscriber calls and concentrates the trafc (see Fig. 5).
Group Switch. The main function of the group switch
is selection, connection, and disconnection of concentrated
speech or signal paths. The group switch often has a general structure.
The overall control of the group switch is performed by
the central processor system. The regional processors take
care of simpler and more routine tasks, such as periodic
scanning of the hardware, whereas the central control system handles the more complex functions. Associated functions included in the group switching resource module are
network synchronization devices and devices to create multiparty calls.
Subscriber Switch. The subscriber switch handles selection and concentration of the subscriber lines, its main
functions are as follows:
11
Figure 4. Typical architecture of

stored program controlled exchange.
Figure 5. Switch architecture.
Transmission and reception of speech and signaling

data to and from the subscriber equipment (for example, on-and off-hook detection).
Multiplexing and concentration of the subscriber
lines, to save hardware and make more efcient use
of the communication links between the subscriber
stage and the group switch.
The architecture should be modular and enable to combine PSTN and ISDN access in the subscriber stage. The
subscriber switch can be colocated with the group switch
in the exchange (central subscriber switch, CSS) or located

at a distance from the exchange (remote subscriber switch,
RSS).
Remote Subscriber Multiplexer. The remote subscriber
multiplexer (RSM) is an add-on subscriber access node,
used in the access network, which can cater small groups
of subscribers. It provides both mobile and standard telephony connections. The RSM multiplexes and concentrates
the trafc to the central or remote subscriber switch but
does not carry out trafc switching functions.
12
Trunk and Signaling. This resource module includes the

circuits for connecting trunks and signaling devices to the
group switch. The module should handle the adaptation to
different signaling systems, namely common channel signaling as well as various register and line signaling systems.
Trafc Control. This resource module contains the trafc
handling and the trafc control functions of the exchange.
This module is responsible for nding the most suitable
route between calling and called subscribers and of verifying that call establishment is allowed.
Operation and Maintenance. This resource module enables tasks such as supervision of trafc, testing of the
transmission accessibility and quality, and diagnostics and
fault localization of devices or trunks.
Common Channel Signaling. This resource module includes the signaling terminals and the message transfer
part (MTP) functions for common channel signaling systems such as SS7.
Charging. This resource module is used in exchanges
that act as charging points. Both pulse metering and specied billing (toll ticketing) can be offered. It should be possible to charge both calls and services, and the charging
should be based on:
Usage
Provision/withdrawal of subscriber services and supplementary services
Activation/deactivation of subscriber services and

supplementary services
Control Modules
The primary function of the control module is to provide
the real-time processing and execution environment required to execute software in application modules and resource modules used to perform trafc-handling functions
and call services. The processing can be centralized where
one processor takes care of all tasks, or distributed where
the processing of information is distributed over several
processors.
Execution of telecom software imposes stringent realtime requirements on the control system. Calls appear
stochastically, short response times are needed, and overload situations must be handled. The main control modules
are: the central processor(s); the data store to store call
data; and the program store to store the actual programs.
In order to achieve an efcient overall control system, it
can be divided into:
Central control. One or several processors that perform the non-routine, complex program control and
data-handling tasks such as execution of subscriber
services, collection of statistics and charging data, and
updating exchange data and exchange conguration.
Regional control. A set of distributed processors
that perform routine, simple and repetitive tasks but
also some protocol handling. They are of different

types optimized for their main tasks, for example,
input/output processing. They often have strict realtime and throughput requirements, for instance for
protocol handling, and may have customized hardware support in the form of ASIC, FPGA and DSP
circuits for this purpose.
Switching Techniques
Until about 1970 most switches were analog and based on
electromechanical rotor switches and crossbar switches.
Since then the utilization of digital techniques has become dominant. Digital switches have been based on synchronous time and space multiplexed circuit switch technology. For pure data communication, packet switching
technology is often used. In order to be able to support
voice and data sharing a common infrastructure, an asynchronous transfer mode (ATM) has been developed. Synchronous time and space multiplexed circuit switch technology is based on synchronous transfer mode (STM) carrier and synchronization technology (STM transport technique is also often used as a carrier for ATM).
A typical STM-based switch architecture is made up of
a combination of time and space switches (T and S, respectively). A space switch connects physical lines by changing
of positions in the space domain, see Fig 6a and a time
switch changes the ordering sequence of data (voice) samples by changing of positions in the time domain, illustrated in Fig. 6b.
The elements T and S can be combined in several ways
to realize a switch and make it congurable in many ways.
Usually time switching is used in input and output stages
and space switching is often used in central parts of a
switch. This basic time-space-time (TST) switch structure
can be used both in subscriber and in group switching
stages. The rst part of a TST switch is a time switch, which
interchanges time slots between the external incoming digital paths and the space switch. The space switch connects
the time switches at the input and the output. The last part
of the TST switch is a time switch, which connects the time
slots between the external outgoing digital paths and the
space switch (see Fig. 7).
The time switch moves data contained in each time slot
from an incoming bit stream to an outgoing bit stream but
with a different time slot sequence. To accomplish this, the
time slot needs to be stored in memory (write) and read out
of the data store (DS) memory and be placed in a new position (read). The reads and the writes need to be controlled,
and the control information needs to be stored in a control
store (CS) memory as well. The timing of DS and CS is
controlled by a time controller (TC). Examples of control
actions are time slot busy or time slot idle.
A typical space switch consists of a cross-point matrix
made up of logic gates that realize the switching of time
slots in space. The matrix can be divided into a number of
inputs and a number of outputs and is synchronized with
the time switching stages via a common clock and a control
store.
Switch architectures based on asynchronous transfer
mode (ATM) handle small packets called cells with a given
13
Figure 6. Examples of switching in space (a) and

time (b).
Figure 7. Typical STM switch with TST

structure.
Figure 8. Typical ATM switch structure with input

and output ports and control space switch.
xed size (53 bytes) divided into header (5 bytes) and payload (48 bytes). The header contains virtual path identiers
(VPIs) and virtual channel identiers (VCIs), where a virtual path (VP) is a bundle of virtual channels (VC). Trafc
can be switched at the VP or VC level cell by cell. Associated
with a VP or a VC is a quality of service (QoS) contract. In

order to be able to guarantee the switching of cells according to the contract without unacceptable cell loss a number
of queues are used at the input ports and output ports of the
switch. Between input and output ports (including buffer
14
queues and cell multiplexers and demultiplexers) a space

matrix often is used, as in Fig. 8. However, other types of
central switching structures sometimes are used, such as
fast ber buses and Banyan networks.
Other switching techniques are used in a telecommunication exchange in addition to STM and ATM. These
include Ethernet and Rapid IO switches as well as IP
routers, each with their own characteristics. RapidIO is
highly efcient internally in nodes with hard latency and
real-time requirements and for small packets, while Ethernet switches show good enough performance for high
throughput for large packets. Embedded IP routers, for IP
forwarding, are useful for nodes that border IP networks
or are part of these. Sometimes, several switches can be
needed inside a node.
CHARACTERISTICS REQUIREMENTS ON THE
EXCHANGE
Availability
High availability of the telecom network and associated
services is the single most important operator and subscriber requirement. Normal requirements on maximum
unavailability are in the order of one or a few minutes
of subscriber unavailability per year. This includes downtime due to faults in the exchange and in the transmission equipment and software, but also unavailability due
to planned software upgrades and often also accidents outside the control of the vendor, such as res, damaged transmission cables, and incorrect operation of the exchange.
Several methods are used to increase the availability of
the exchange to the subscriber:

Redundancy, including fault-tolerant processors

Segmentation
Diagnostics of hardware and software faults
Recovery after failure
Handling of overload
Disturbance-free upgrades and corrections
Robustness to operator errors
Each of these is treated briey below.

Redundancy. In order to cope with hardware faults, redundant hardware is used for those parts of the switch
that are critical for trafc execution. Requirements are in
the order of 1000 years for the mean time between system
failures (MTBSF).
Specically, current technology requires that a redundant processor is available in synchronized execution (hot
standby), ready for a transparent takeover if a single hardware fault occurs in one processor. An intelligent fault analysis algorithm is used to decide which processor is faulty.
In a multiprocessor system, n + 1 redundancy is normally
used, where each processor can be made of single, double,
or triple hardware. When one processor fails, its tasks are
moved to the idle (cool standby) processor. A similar redundancy method is based on load sharing, where the tasks of
the failed processor are taken over by several of the other
processors that are not overloaded themselves.

The group switch hardware is also normally duplicated
or triplicated, because it is so vital to the exchange functions. The less central hardware devices, such as trunk devices, voice machines, transceivers, signal terminals, and
code receivers, are normally pooled, so that a faulty device
is blocked from use and all users can instead access the
remaining devices until the faulty device is repaired.
Segmentation. To avoid system failure a fault must be
kept isolated within a small area of the exchange. This is
done by segmentation of hardware, with error supervision
at the interfaces. In software, the segmentation is made
by partitioning of and restricted access to data structures;
only owners of data can change the data, where the owner
can be a call process or a function.
Diagnostics of Faults. After the occurrence of a fault in
hardware or software, the fault must be identied and localized, its effect restricted, and the exchange moved back
to its normal state of execution. For this to work, the diagnostics must be extensive and automatic. The exchange
must be able to identify the faulty software and hardware
and must be able to issue an alarm, usually to a remotely
located operator.
Recovery After Failure. After a fault has been detected,
the effect should be restricted to only the individual call or
process (for instance, an operator procedure or a location
update by a mobile subscriber) or an individual hardware
device. This call or process is aborted, while the rest of the
exchange is not affected. The recovery must be automatic
and secure. In a small fraction of events, the fault remains
after the low-level recovery, or the initial fault is considered too severe by the fault handling software, so that a
more powerful recovery procedure must be used. The process abort can be escalated to temporary blocking of hardware devices or software applications and, if required, result in the restart of an entire processor or a number of
processors. If the restart fails in recovering the exchange
into normal trafc handling, new data and software are
loaded from internal or external memory.
Handling of Overload. The exchange is required to execute trafc literally nonstop and when offered more trafc
than it can handle, the rejection of overow trafc should
be made gracefully. ITU requires, that an exchange that
is offered 150% of what it was designed for should still
have 90% of its maximum trafc handling capacity. The exchange must also be able to function without failure during
extreme trafc loads. Such extreme loads can be both short
peaks lasting a few milliseconds or sustained overload due
to failures in other parts of the network. Overload handling is accomplished by rejecting excess trafc very early
in the call setup, before it has used too much processor time
or any of the scarce resources in the switching path. Figure 9 shows the overload performance with and without an
overload control function.
Disturbance-Free Upgrades. Both upgrading of software
packages and the replacement or extension of hardware to
15
Figure 9. Overload control.
the exchange must not disturb ongoing trafc execution.

This should be true both for fault corrections and when
new functions are introduced. The architecture must thus
allow trafc to be executed in redundant parts, while some
parts of the exchange are upgraded.
Robustness to Operator Errors. Security against unauthorized access is accomplished by use of passwords and
physically locked exchange premises. The user interface
part of the exchange can supervise that valid operation
instructions are followed for operation and maintenance,
and it can issue an alert or prohibit other procedures. Logging of operator procedures and functions for undoing a
procedure can be used. If a board is incorrectly removed
from the exchange, the exchange should restrict the fault
to that particular board, use redundant hardware to minimize the effect of the fault, and then indicate that this
board is unavailable. A simple user interface with on-line
support makes operator errors less probable.
Grade of Service. The real-time delays in the exchange
must be restricted to transmit speech correctly. Packetswitched connections have problems achieving good realtime speech quality for this reason, especially during heavy
usage. Circuit-switched networks have so far given the
best real-time performance regarding delays and grade
of service for voice and video, compared with packet data
networks. ATM switching (due to its short and xed size
cell/packets) also fullls the grade of service requirements
and also denes service classes for different requirements.
Scalability
There is a need for scalable exchanges regarding capacity,
from the very small (such as base stations and local exchanges in desolate areas) to the very large, mainly the
hubs (transit exchanges) and MSC:s of the networks and
exchanges in metropolitan areas. Furthermore, there is
sometimes a requirement for downward scalability regarding physical size and power consumption, particularly for
indoor or inner city mobile telephony.
The following are the common system limits for downward scalability of an exchange:
The cost to manufacture and to operate in service will

be too high per subscriber or line for small congurations.
The physical size is limited by the hardware technology and by the requirements for robustness to the environment, and what is cost efcient to handle.
The power consumption is limited by the hardware
technology chosen.
The following are the common system limits for upward
scalability of an exchange, each treated briey below:

(Dynamic) real-time capacity

(Static) trafc handling capacity
Grade of service (delays)
Memory limits
Data transfer capacity
Dependability risks
Processing Capacity. New more advanced services and

techniques requires more processing capacity, this trend
has been valid for (a) the replacement of analog technology with digital processor controlled functions and (b) the
development of signaling systems from decadic and multifrequency to packet mode digital signaling, including trunk
signaling protocols such as ISDN user part (ISUP) together
with mobile application part (MAP) and transaction capability application part (TCAP) and the development of mobile telephony. In the charging area, the trend from pulse
metering to detailed billing affects the call capacity.
The number of calls per subscriber has also increased
due to lower call costs from deregulation and due to the
use of subscriber redirection and answering services.
Trafc Capacity. It is very important to design and congure hardware and software correctly, in order to minimize hardware costs, and at the same time ensure suf-
16
ciently low congestion in the exchange and in the network.

Normally, the switch fabric is virtually non-blocking, and
the congestion occurs in other resources, such as the access lines, trunk lines, and other equipment. The relation
between congestion probability and the amount of trafc
is well known, if all devices are accessible and the trafc
follows a Poisson process; that is, the times between offered calls are independent and exponentially distributed.
In some cases, the probability can be calculated explicitly.
In more complex device congurations with non-Poisson
trafc, the congestion probabilities are most easily calculated by simulation techniques.
Memory. The amount of memory per subscriber line or

trunk line is another way to measure the complexity of a
telecom application. The trend in this area is similar to that
of processing capacity, and the same factors are responsible
for the large increase in memory needs. Due to the realtime requirements, fast memory is used extensively, and
secondary memory is only used for storage of backups, log
les and other data where time is not critical.
Transfer Capacity. A third part of the switching system capacity is the data transfer from the exchange to
other nodes, for example, to other exchanges and network
databases, billing centers, statistical post processing, and
nodes for centralized operation. There has been a growing
demand for signaling link capacity due to large STPs, for
transfer capacity from the exchange to a billing center due
to detailed billing and large amounts of real-time statistics, and to transfer capacity into the exchange due to the
increased amount of memory to reload at exchange failure.
Dependability Risks. Although dependability has increased for digital exchanges, there is a limit as to how
large the nodes in the network can be built. First, the more
hardware and the more software functions assembled in
one exchange, the more errors there are. The vast majority
of these faults will be handled by low-level recovery, transparent to the telecom function, or only affect one process.
However, a small fraction of the faults can result in a major
outage that affects the entire exchange during some time.
As an example, assume that the risk of a one-hour complete exchange failure during a year for one exchange is
1%. If we add the functionality of an SCP to an HLR node,
then we more than double the amount of software, and
presumably the number of faults, in the node. The risk of a
major outage should be larger in a new exchange introducing new software with new faults. Only if unavailability
due to completely stopped trafc execution is much less
than the total effect of process abortions and blocked hardware devices can we build exchanges of unlimited software
complexity.
The second reason for a limited complexity from a dependability point of view is that network redundancy is
required and only can be used if there are several transit
exchanges in the network.
Life-Cycle Cost
Since the 1980s, the operating cost has become larger than
the investment cost of an exchange. Thus, the emphasis on
efcient operation and maintenance has increased, regarding both ease of use and utilization of centers that remotely
operate a number of exchanges that are not staffed. For
ease of use, the telecommunication management network
(TMN) was an attempt by ITU to standardize the operator
interface. After several years this standard is still not used
much. Instead, the operator interface is to a large extent
dependent on the exchange manufacturer as well as the
requirements from the telecom operator company. Several
open and proprietary user interfaces are common.
For central operation, more robust methods of remote
activities have evolved. Software upgrades and corrections,
alarm supervision and handling, collection of statistics and
charging data, and handling and denition of subscriber
data are all made remotely. Transmission uses a multitude
of techniques and protocols. Open standard protocols have
taken over from proprietary protocols.
In addition, important parts of the life-cycle cost are
(a) product handling for ordering and installation and (b)
spare part supply.
EVOLUTION TRENDS
Technology Trends
Due to the effects of semiconductor process scaling, improved chip fabrication yield, and increasing numbers of
connectivity layers the storage capability of memory and
the execution speed of processors has doubled every 18
months during the last 40 years, following the so called
Moores law. This exponential growth of transistors-perchip will continue, but will force new hardware architectures such as chip multiprocessors and systems on chip
in order to keep energy use within reasonable limits. The
development of optical bers, including the introduction
of wavelength multiplexing, is perhaps even faster. Thus
there are many factors that lead to cheaper nodes and
higher bit rates in the network. At the same time, digital
coding and compression techniques have been improved
that makes it possible to transmit voice with traditional
telecom quality using only a fraction of the bandwidth that
is used today. These developments are changing the design
of both nodes and networks.
It is also important to provide increased interoperability between network standards since the end users do not
want to be concerned about where a person is physically
located or to which network the person is connected. The
introduction of universal personal numbers can solve this
and lead to a convergence of xed and mobile telephony.
The possibility of accessing short message services (SMS),
fax, and e-mail via xed and mobile devices and so-called
Internet telephony are examples of services that illustrate
the needs for interoperability and convergence of telecommunication and datacommunication.
High bit rates at a low price, combined with the demand
for real-time multimedia services, indicates that the network must either become more exible or consist of sev-
17
Figure 10. Converged network architecture with decoupling of access,

transport, control and service functions.
eral different but interoperating networks. Packet, cell, and

new, more exible circuit switching techniques supported
by new signaling protocols (such as MPLS) can solve these
needs. In order to integrate service execution, control and
connectivity horizontally across multiple access networks,
a layered architecture approach can be used using a common transport layer based on IP and Ethernet technology over ber rather than delivering single services such
as voice telephony or data access (vertically integrated
networks). This architecture supports efcient IP packet
based transport of both signaling and payload, which is not
possible with classic switches. By doing this one introduces
a single IP infrastructure that can handle all network services, such as xed and mobile communications.
Network Convergence
The evolution and convergence towards a common core infrastructure is sometimes called the New Generation Network (NGN) architecture, see gure 10.
This network evolution is supported by techniques for
separation of control and switching such as the media gateway control protocol H.248. Call session control functions
and protocol collections such as H.323 enable call setup
including coded and compressed voice calls and choice of
coding standard to be used. The Session Initiation Protocol
(SIP) has a similar but more limited scope for the setup of
communication sessions between two parties and selection
of coding standard using the Session Description Protocol
(SDP).
The IP Multimedia Subsystems (IMS) dened by the
European Telecommunications Standards Institute (ETSI)
and the 3rd Generation Partnership Project (3GPP) allow
video and other media forms to be exchanged and charged
on a session by session basis from peer to peer in a way
similar to classic phone calls.
So-called softswitches or media gateway controllers include call session control functions for handling of voice
calls and other session oriented services. They are also responsible for that sessions can be connected via a physical
switch or media gateway (MGW) and handle the signaling
between network nodes and other networks. The call session control function establishes a call or session, and fur-
ther manages its reserved connection path resources end

to end, for example through an ATM or IP backbone network and for media stream processing. The MGM provide
physical switching and interfaces to access nodes and other
networks.
BIBLIOGRAPHY
1. R. L. Freeman, Telecommunication System Engineering, 3rd
ed., New York: Wiley, 1996.
2. D. Minoli, Telecommunications Technology Handbook, Boston:
Ar-tech House, 1991.
3. M. Schwartz, Telecommunication Networks: Protocols, Modelling and Analysis, Reading, MA: Addison-Wesley, 1987.
TONY LARSSON
ALEXANDER KOTSINAS
BJORN
KIHLBLOM
Halmstad University, Halmstad,
Sweden
3 Scandinavia, Stockholm,
Sweden
Ericsson Networks, Stockholm,
Sweden

05 Biomedical Engineering

Загружено:

Сведения о документе

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

05 Biomedical Engineering

Загружено:

Авторское право:

Доступные форматы

338

Nerve. A sketch of a typical neuron is given in Fig. 1(a),

Node of Ranvier Schwann cell

bers (typically 5 m/sec) and its diameter lies between 10 m

Figure 1. (a) A motor-neuron is shown with typical

Excitable CellsMembrane Structure

Figure 2. The figure shows the membrane that

Even though the resting potassium permeability exceeds

Source. Hodgkin-Huxley (2).

charge at the inside surface of the membrane and thereby

Each ionic current based on diffusion and electric field has

where Di is Ficks constant and x is the direction across the

where d is the membrane thickness; the outward electric field

To add Eq. (3) and Eq. (4), a relationship between Di and ui

+ PNa [Na]o + PCl [Cl]i

where RT/F is replaced by 25.2 for T 293 C and hence Vm

chloride is close to its Nernst potential in this illustration.

Figure 4. Four configurations for patch clamping are described. The

where EK is the potassium equilibrium (Nernst) potential.

where is the probability of a transition from open to closed

models have been formulated from experiments conducted

INa = gNa (Vm ENa )

(a) Unitary K currents

(b) Ensemble average

where gl is an experimentally determined constant and El is

We may interpret this to describe a functional potassium

and considered m as an activating gating variable and h an

In these expressions we have 0 m 1 and 0 h 1 while

where n n /(n n) and n 1/(n n). The initial value

where Im(t) is the total measured transmembrane current as

The remaining values are needed to implement a Hodgkin

and this is provided by high energy phosphates (ATP). The

where m RmCm is the membrane time constant.

where IT VT /Rm is the threshold current amplitude for T

which predicts a linear relationship with duration for the

where ri is the intracellular resistance per unit length. (The

I0 /(4eR2) where the extracellular medium has a uniform

where ri (a2i)1. In Eq. (31) we replaced i by vm since for

p(x x ) i + (y (x, yy,)z)+ (z z )

where we have written out

where Iv is a volume source density. In the literature Iv is also

fer (surely it cannot be zero everywhere since there is a

which is Poissons equation. Even though we understand how

The field from surface sources have useful properties, which,

Surface Sources and Field Discontinuities

where KS(0) is the value of the surface-source density at the

Since a surface source occupies zero volume, the S obtained

Eq. (37) we get

Eq. (37) we now have

where positive z is directed from 1 to 2.

The partial derivative in Eq. (39) is known as a directional

In spherical coordinates (1/R) aR /R2 where aR is a unit

where p I0dl is the dipole moment (namely the product of

where we use dS dS because the dipole orientation for a

where we utilized aR az z/ 2 z2 because the axis of the

The vector double-layer strength is n where n is the unit

Figure 7. The single cell lies in an unbounded volume conductor.

and also satisfies Laplaces equation. From Eq. (35), we

so that the solution for from Iv has the interesting property

an inequality that reflects both the nonzero transmembrane

a result that is obtained by associating region 1 in Fig. 7 with

and substituting Eq. (50) results in

As noted earlier we see that the conductivity is absent from

membranes if its resistivity (which is high) and thickness