IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 39, NO. 4.
APRIL 1992
W. E. Larimore, “Predictive inference, sufficiency, entropy, and an
asymptotic likelihood principle,” Biometrika, vol. 70, no. I , pp. 175-
181, 1983.
Y . Lee and M. Schetzen, “Measurement of the kernels of a nonlinear
system by cross-correlation,” Internat. J . Contr., vol. 2, pp. 237-
254, 1965.
R. Llinas, “The intrinsic electrophysiological properties of mam-
malian neurons: insights into central nervous system function,” Sci-
ence, vol. 242, pp. 1654-1664, 1988.
V. Marmarelis, “Practical identification of the general time-variant
nonlinear dynamic system,’’ in Proc. Internat. Con8 Cybern. Soc. ~
pp. 727-733, 1979.
W. Rugh, Nonlinear System Theory: The Volterra/ Wiener Approach.
Baltimore, MD: The Johns Hopkins Univ. Press, 1981.
M. Schwartz and C. Shapiro, “Recovery functions of human soma-
tosensory and visual evoked potentials,” Annals. New York Acad.
Sci., vol. 112, pp. 510-525, 1964.
R. Sclabassi, C. Hinman, J. Kroin, and H. Risch, “The modulatory
effect of prior input upon afferent signals in the somatosensory sys-
tem,” in Proc. Joint Automat. Contr. Conf., vol. 12, pp. 787-795,
1977.
R. Sclabassi and G. Noreen, *“The characterization of dual-input
evoked potentials as nonlinear systems using random impulse trains,”
in Proc. Pittsburgh Modeling and Simulution Conf., vol. 12, pp.
1123-1130, 1981.
R. Sclabassi, J. Kroin, C. Hinman, and H. Risch. “The effect of
cortical ablation on modulatory activity in the cat somatosensory sys-
tem,” Electroencephalogr. clinical Neurophysiol., vol. 64, pp. 3 1-
40, 1986.
1201 R. J. Sclabassi, D. N. Krieger, and T. W. Berger, “ A systems the-
oretic approach to the study of CNS function,” Annal. Biomed. Eng.,
vol. 16, pp. 17-34, 1988.
[21] R. Sclabassi, D. Krieger, J. Solomon, J. Samosky, S. Levitan, and
T. Berger, “Theoretical decomposition of neuronal networks,” in
Advanced Methods of Physiological System Modeling, V. Marma-
relis, Ed. New York: Plenum, 1989.
(231 J. Solomon, D. Krieger, T. W. Berger, and R. J. Sclabassi, “Re-
sponse prediction for the hippocampus of the rabbit,” in Proc. So-
ciety Neurosci., 1989.
[24] M. Verbaten, Investigations on Habituation in Humans. Utrecht,
Germany, Elinkwijk B. V., 1981.
[25] V. Volterra, Theory of Functionals and of Integral and Integro-Dif-
ferential Equations. New York: Dover, 1959.
[26] N. Wiener, Nonlinear Problems in Random Theory. New York:
Wiley, 1958.
[27] J. Woestenburg, M. Verbaten, and J. Slangen, “Stimulus informa-
tion and habituation of the visual event related potential and the skin
conductance reaction under task-relevance conditions,” Biol. Psy-
chol., vol. 16, pp. 225-240, 1983.
[28] J. Woestenburg, M. Verbaten, H. Van Hees, and J . Slangen, “Single
trial ERP estimation in the frequency domain using orthogonal poly-
nomial trend analysis (OPTA): Estimation of individual habitua-
tion,” Biol. Psychol., vol. 17, pp. 173-191, 1983.
A Model of Safe Levels for Electrical Stimulation
Robert V . Shannon
Abstract-A model is presented that represents a large body of data
on safety and damage levels of electrical stimulation. The predictions
Manuscript received March 12, 1991; revised September 27, 1991. This
work was supported in part by NIH Grant DC00409 and by FDA orphan
product Grant FD-R-000686.
The author is with the House Ear Institute, Los Angeles, CA 90057.
IEEE Log Number 9106412.
0018-9294/92$03.00 0 1992 IEEE
of the model are consistent with known principles of current flow and
known mechanisms of damage around stimulating electrodes. It is pro-
posed that limits on levels of electrical stimulation take into account
the location of the electrode relative to the stimulated tissue and these
limits can be computed algorithmically from the model.
INTRODUCTION
Safety limits for electrical stimulation based on actual data from
animal experiments have led to a bewildering array of apparently
conflicting results for these limits due to the wide variety of stim-
ulating waveforms, electrode sizes, and charge densities. Mc-
Creery et al. [ 2 ] , [3] have presented one of the most complete sets
of data on safe and unsafe electrical stimulation levels for surface
electrodes on the cortex. They assessed damage to underlying tis-
sue from stimulation with 400 ps pulses over a range of current
densities and electrode shapes and sizes. Their data are schemati-
cally represented in Fig. 1. This figure summarizes results from
dozens of experiments showing the charge and charge densities at
which damage occurs. Observe that both parameters must be spec-
ified to assure that the stimulation is occurring in a region where
damage does not occur. We will demonstrate in this paper that a
simple relation can be used to determine these parameters for safe
stimulation. We will discuss the possible biophysical reasons for
this relation, and illustrate its implications for future electrode de-
signs and charge limits.
Previously proposed charge density limits have ranged from 15
to 65 pC/cm’/phase irrespective of electrode area [ I ] . For small
electrode areas, this range of charge densities would produce stim-
uli well below the damage limit shown in McCreery et al.’s data.
However, for large electrode areas, these charge densities would
probably produce damage. Thus, a limit purely based on charge
density would be quite conservative for small electrodes but might
actually cause stimulation-induced damage with large electrodes.
The boundary which McCreery found between safe and unsafe
charge injections at different charge and charge density levels can
be approximated by the equation
where D is charge density in pCoulombs/cm’/phase and Q is
charge in pCoulombs/phase. This equation describes a family of
lines for changing values of k , three of which are shown in Fig. I .
When k = 2, the straight line falls in an area where damage was
observed. If this line were used to define the limit of safe stimu-
lation, some stimuli would be expected to cause damage. The curve
for k = 1.5 defines a set of parameters where no damage was ob-
served and is used for all computations that follow.
Since D = ( I 7 ) / A = Q / A where I is current, T is the duration
of each phase of a biphasic pulse, and A is the surface area of the
electrode, (1) can be expressed as
log ( I T / A ) = k - log (IT)
which rearranges to
IT = ( A 1 o k ) 0 5 . (2)
Equation (2) describes the data of McCreery et al. through a simple
relationship between the three major variables in electrical stimu-
lation: current, pulse duration, and electrode surface area.
For disk-shaped electrodes of diameter d where A = ( r d ’ ) / 4 .
(2) can be written as
I = ( d / ( 2 ~ ) )1( ~0 ~ ) ~ ~ (3)
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 39, NO. 4, APRIL 1992 425

L loa(D)=k-loa(Q) size and space limitations of the target location for the electrodes,
Q the electrode perimeter and area should be as large as possible.
\
N
E The maximum charge density is not constant as a function of
U
\
electrode perimeter or area, but decreases hyperbolically as perim-
U eter and area increase. This result is due to the fact that stimulus
a
v
charge and charge density are cofactors in determining neural dam-
x age 131.
t
cn
c
a,
n DISCUSSION
a,
(5, It is important to note that the limit of safe stimulation is linearly
L
0 related to electrode diameter, not electrode area. This result is
L
0 probably due to the charge “building up” at the edges, to create
Charge p e r phase (pC/ph) higher charge densities around the perimeter of the electrode [4]-
161. McCreery et al.’s data were collected with cortical surface
Fig. I . Region of charge and charge density where neural damage was electrodes where stimulable tissue was in direct contact with the
observed by McCreery et al. [2] is represented by the hatched area. The
solid lines represent the proposed model with parameter k = 1 .O, 1.5, and electrode surface. Mathematical models and measurements [4]-[6]
2.0. have demonstrated that excess charge density at the electrode edges
can be reduced or eliminated by recessing the electrode into the
carrier, creating an active electrode surface at the bottom of a well,
and by flaring the opening of the recess. Edge effects might also
be eliminated by complex geometries which increase the effective
perimeter within the same surface area [6]. More basic research,
animal experiments, and modelling are necessary to determine if a
better electrode surface shape is possible. Fabrication feasibility is
another consideration in designing new electrode shapes to maxi-
mize the effective perimeter while holding the area constant.
The mechanism of tissue damage from electrical stimulation
probably depends on the charge density at the stimulable tissue.
McCreery and colleagues [2] suggest that the mechanism of dam-
age from electrical stimulation is probably due to overstimulation
and “mass action” effects, rather than to electrochemical byprod-
ucts or nonreversible reactions around the electrode surface. Thus,
relatively large charge densities near the electrode may not be
harmful to nonstimulable tissues, as long as they do not produce
X / 1 nonreversible reaction products.
I I I I We suggest that damage limits be considered in terms of three
I I 1 1
categories: near-field as described above, when neurons are adja-
600 - cent to the electrode, mid-field, and the far-field. In the mid-field
500 - case the stimulable tissue is far enough away from the electrode
400 - contact surface so that the current field is uniform and damage might
300 - be related to electrode surface area. This case might be comparable
200 - to recessed electrodes in that the charge distribution, which is con-
100 - centrated around the electrode edges near the electrode, would be
more diffuse as the distance from the electrode increases [4]-[6].
I I I I
0 1 2 3 4 5 6 7 0 1 2 3 4 Within the stimulable tissue the charge density might be uniformly
Electrode Perimeter ( m m ) Electrode Area (mm2) distributed and its magnitude reduced according to the inverse
Fig. 2 . Proposed maximum allowable amplitude, charge, and charge den- square of the distance from the electrode. In most applications of
sity fork = 1.5. stimulating electrodes a reasonable estimate is usually available of
the distance between the electrode surface and the stimulable tis-
sue. Thus, a fixed charge density limit might be appropriate if the
Let us explore the implications of (3) for the proposed limit of k electrode is some distance from the stimulable tissue, while the
= 1.5. Fig. 2 presents plots of the maximum pulse amplitude, present model might be appropriate if the electrode is immediately
charge/phase, and charge density as a function of electrode disk adjacent to stimulable tissue.
pcrimetcr and electrode disk area. Somewhat surprisingly, there is In the far-field case, the electrode is a large distance from the
a linear relation between electrode perimeter and maximum charge. stimulable tissue and can be considered as a point source. In this
This linearity leads to a point of diminishing returns in choosing case the surface area of the electrode and thus charge density at the
clcctrode size. While the largest electrode diameter may be desir- electrode surface is not important. Assuming that nonreversible re-
able to allow the maximum charge delivery, we are limited by the action products are not generated and that nonstimulable tissues or
available space for the electrode, and the electrode area increases fluids near the electrode are not damaged by the charge densities
as the square of the diameter. The calculations presented in Fig. 2 near the electrode, only the charge density at the target tissue would
also show that longer pulse durations can be used-at maximum am- be of importance. In this case, the important stimulus parameter
plitude as the electrode perimeter is increased. Thus, within the for damage is not charge density at the electrode, but rather the
426 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING. VOL. 39. NO. 4. APRIL 1992
amount of charge per phase in each pulse, reduced by the inverse
square of the distance between the electrode and the target neurons.
ADDITIONALCONSIDERATIONS
McCreery et al. [3] measured damage after stimulation for only
7 h by a 50 Hz 400 ps/phase biphasic pulse train. At the present
time we do not adequately know how to extrapolate from this data
to predict damage thresholds for longer stimulation durations, or
for higher stimulation rates, or even for different pulse durations.
For these reasons, we must be very conservative in setting our cri-
tenon for safe limits of stimulation in patients. For the stimulus
conditions of McCreery et al., k = 1.5 is a conscrvative limit and
has been used in all calculations in this report. A more comprehen-
sive model of safe levels for electrical stimulation would also ac-
count for the effects of pulse rate, pulse duration, stimulus duty
cycle, and duration of exposure. In the simplest case these factors
might simply change the value of k at which damage is observed,
although it is possible that these factors might change the nature of
the function.
The present model is based on damage observed in cortical neu-
rons. It is possible that other neurons would show a different pat-
tern of damage, due to differing metabolism, packing density, and
fiber diameter. However, the implications of the model discussed
above are in accord with known mechanisms of neural activation
and damage from electrical stimulation, and with principles of cur-
rent flow around electrode surfaces.
CONCLUSION
The present model could be used as an initial framework for de-
signing future animal safety studies. Based on the electrode ge-
ometry and surface area used, and considering the estimated dis-
tance between the stimulable tissue and the electrode, the safe
charge limit predicted by the model could provide a testable hy-
pothesis of the safe limit for electrical stimulation. Designers of
electrodes for prosthetic stimulation either could optimize the elec-
trode size and geometry based on the placement site and expected
distance from electrode to target tissue, o r could design electrode
arrays for the worst case of stimulable tissue in contact with the
electrode surface.
REFERENCES
P. A . Leake, D. K . Kessler, and M. M. Merzenich, “Application and
safety of cochlear prostheses,” in Neural Prostheses: Fundamentd
Studies,W. F. Agnew and D. B. McCreery, Eds. Englewood Cliffs,
NJ: Prentice-Hall, 1990, pp. 253-296.
D. B. McCreery, W . F. Agnew, T . G. H. Yuen, and L. Bullara
“Comparison of neural damage induced by electrical stimulation with
faradic and capacitor electrodes,” Ann. Biomed. Eiig., vol. 16, pp.
463-481, 1968.
D. B. McCreery, W. F. Agnew, T. G. H. Yuen, and L. Bullara,
“Charge density and charge per phase as cofactors in neural injury
induced by electrical stimulation,” IEEE Truns. Biomed. Eng., vol.
BME-37, pp. 996-1001, 1990.
J. T. Rubinstein, F. A . Spelman, M. Soma, and M. F. Suesserman,
“Current density profiles of surface mounted and recessed electrodes
for neural prostheses,” IEEE Trans. Biomed. Eng., vol. BME-34, pp.
864-875, 1987.
M. F. Susserman, F. A . Spelman, and J . T. Rubinstein, “ I n vitro
measurement and characterization of current density profiles produced
by nonrecessed, simple recessed, and radially varying recessed stim-
ulating electrodes,” IEEE Truns. Biomed. Eng.. vol. BME-38, pp.
401-408. 1991.
0018-9294/92$03.00 0 1992 IEEE
[6] X. Xue and D. J . Anderson, “Finite element analysis of electrical cur-
rent fields induced by stimulating electrodes,“ Abstracrs ARO M i d
winter Res. Meer.. D. J . Lim, Ed., p. 134, 1991.
Microcomputer-Based Portable Long-Term
Spasticity Recording System
Dejan Tepavac, James R . Swenson, Jerome Stenehjeni,
Ivan Sarjanovii., and Dejan PopoviC
Abstract-A device for long-term monitoring of muscle activity
(EMG) with surface electrodes and method of its application are de-
scribed in this paper. This device is called a microcomputer two-chan-
ne1 EMG monitor. The device can be used for up to 24 h monitoring
of EMG activity, followed by data transfer to a host computer for sig-
nal analysis. This device records amplified, rectified, and integrated
EMG activity. Shorter recording time allows shorter sampling periods
suitable for different other EMG analysis. Recording of spontaneous
EMG in complete spinal cord injured subjects was the original reason
for the design of the long-term monitor. These recordings were used
for estimation of spasticity in complete spinal cord patients.
INTRODUCTION
It is known that spasticity may be a difficult and complex prob-
lem in the motor rehabilitation of subjects with an upper motor
central nervous system lesion. There are several definitions of
spasticity. Summarizing published material we can state that fol-
lowing phenomena are included in behavior called spasticity: 1 )
increased resistance of skeletal muscles to passive movements: 2)
clonus; 3) increased stretch reflex; and 4) mass reflexes as a path-
ological form of motor hyperactivity [24].
There is no widely used objective method for measurement of
spasticity. The need for therapeutic intervention in the management
of spasticity relies upon subjective data derived from the history
and physical examinations. In the assessment of the degree of spas-
ticity there is frequently a poor correlation between the patient’s
complaints and the physical finding. Several systems were sug-
gested as possible tools for quantification of spasticity. We will
mention representatives of these systems: I ) goniogram analysis of
the freely swinging leg, [ 2 ] , [3]; 2) spring-loaded stimulation of
clonus at the ankle and the recording of resultant motor activity
[ 7 ] ;3) muscle force recording, invoked by the reflex hammer as a
stimuli [13]; 4) measuring resistance to passive movement of the
extremity [16]; 5) measuring the ratio of maximum H reflex to
maximum M response [ 181; 6) measuring the H a n d stretch reflex
Manuscript received June 8, 1990: revised August 19, 1991. This work
was supported in part by the Department for Physical Medicine and Re-
habilitation, University of Utah. Salt Lake City. The work conducted in
Belgrade was supported by the Research Council of Serbia. Belgrade, Yu-
goslavia.
D. Tepacac was with the Faculty of Electrical Engineering, Belgrade
11000, Yugoslavia. He is now with The Miami Project to Cure Paralysis.
School of Medicine, University of Miami, Miami, FL 33136.
J . R. Swenson is with the School of Medicine. University of Utah. Salt
Lake City, UT 84112.
J. Stenehjem is with Sharp Rehabilition Center. San Diego. CA 92123.
I. SarjanoviC is with the Rehabilition Institute Dr. Miroslav Zotovic.
Belgrade, Yugoslavia.
D. PopoviC is with the Miami Project to Cure Paralysis. School of Med-
icine, University of Miami, Miami, FL 33136, on leave from the Faculty
of Electrical Engineering. Belgrade 11000, Yugoslavia.
IEEE Log Number 9106413