CUN. CHEM.

2 1/9, 1297-1304 (1975)

A Flow-System Multiangle Light-Scattering

Instrument for Cell Characterization

G. C. Salzman, J. M. Crowell, C. A. Goad, K. M. Hansen, R. D. Hiebert,

P. M. LaBauve, J. C. MartIn, M. L. Ingram, and P. F. Mullaney

A flow-system cell-analysis instrument is described in (12). Some brain cell types have been shown to pro-
which cells from a heterogeneous population are char- duce distinct light-scatter patterns from cells on
acterized by their light-scatter patterns alone. As the slides (13). Several models have been proposed for
cells pass at high speed through a focused helium/neon identifying some types of mammalian cells from their
laser beam, the scatter pattern from each cell is sam-
light-scatter patterns (14, 15). Fourier optical tech-
pled simultaneously at up to 32 angles between 00 and
niques have been used to distinguish between normal
300 with respect to the laser beam axis, and the scatter
pattern for each cell Is transferred to a computer. A and abnormal exfoliated cervical cells, with use of
mathematical clustering algorithm is used to determine photomicrographs of selected cells (16).
the number of classes into which the cells can be divid- Two flow-system instruments have been used for
ed, and a linear separation algorithm is used to find the separating cells into different morphological classes,
boundaries between the classes. Preliminary results on based on the light-scatter pattern from each cell as it
exfoliated cells from gynecological specimens are pre- passes through a focused laser beam. In the first sys-
sented. This technique may be useful for automated tem, which is built around an existing cell separator
prescreening of gynecological specimens. (17), scattered argon-ion laser light is detected at
only two angles. This system has been used to physi-
Addftlonal Keyphrases: laser #{149}
categorization of cells from cally separate unfixed, unstained human leukocytes
vagina and cervix #{149}
cancer screening
into three classes: neutrophilocytes, monocytes, and
It has been established that the light-scatter pat- lymphocytes (18, 19). The topic of this paper is a sec-
tern produced by any single scattering object repre- ond flow system in which the light-scatter pattern for
sents a unique physical description of the morpholo- each cell is simultaneously sampled at 32 angles and
gy of the object. In principle, then, an appropriate each scatter pattern stored in a computer for analysis
sampling of the scatter pattern should permit a by a mathematical clustering algorithm. Each light-
unique description of the spatial properties of the scatter pattern is acquired by the computer in 250 his.
scattering object. This inversion process (i.e., conver- This system is described in detail, and preliminary
sion of experimental data to the characteristics of the results are presented for gynecological specimens.
scattering objects) represents a formidable problem
Materials and Methods
in mathematical physics. However, the method has
been successful in several histological applications Multiangle Light-Scattering Flow System
Figure 1 shows a schematic drawing of the mul-
Small-angle light scattering from both particles tiangle light-scattering system. A 5-mW helium-neon
and biological cells has been shown to be a measure laser beam (Spectra-Physics, Inc., Mountain View,
of their size (5-8). Light scattering from cells in sus- Calif. 94042; Model 124A, 632.8-nm wavelength) is
pension has been used as a method for identifying focused by a 15-cm focal length spherical lens onto a
bacterial cells (9-il) and as a technique for distin- cell sample stream, which is surrounded by a dis-
guishing between normal and virus-infected cells tilled-water sheath. The laser beam passes out of the
flow chamber (5) and strikes the center of a light-
Biophysics and Instrumentation Group, Los Alamos Scientific
Laboratory, University of California, Los Alamos, N. Mex. 87544. scatter detector array consisting of 32 concentric
Received April 16, 1975; accepted May 21, 1975. rings of photodiode material (Recognition Systems,

CLINICAL CHEMISTRY, Vol. 21, No. 9, 1975 1297

 Top VIew
Detector elements

Flow chamber
Sort command

Pop-
11/20

FIg. 2. Block diagram of the data-acquisition electronics for

the multiangle light-scattering system
The pulse heights from each of the 32 logarIthmically amplified light-scatter
signals are sensed and held and multiplexed into an analog-to-dIgital convert-
er (ADC). The digitized signals constituting the light-scatter pattern for a sin-
gle cell are transferred to computer memory via a direct memory access In-
terface (DMA)
FIg. 1. Schematic drawing of the multiangle light-scattering
system showing a 5-mW helium-neon laser, a 15-cm focal
length spherical lens, a top view of the flow chamber, and
two of the light-scattering angles, O and 01
The face of the photodlode array Is shown In the lower half of the figire. The pulse again crosses the threshold as the cell leaves
Innermost ring Is a solid circle concentric with the center of the array.The
outside diameter ofthe largest photodlode ring Is 32mm the beam, the peak sense..and-hold circuits are locked
to prevent a second pulse from entering the system.
The master control unit then steps the analog multi-
plexer from amplifier to amplifier, passing each held
level to a fast eight-bit analog-to-digital converter
Inc., Van Nuys, Calif. 91406; Model WRD-6420A). (ADC) (21). The digitized output of the ADC is
Each ring detects the light scattered by a cell at a dif- transmitted to the computer memory via a direct
ferent polar angle, 0, which is the angle measured at memory access interface (Digital Equipment Corp.,
the sample stream/laser beam intersection inside the Maynard, Mass. 01754; Model PDP 11/20 computer,
flow chamber between the laser beam and the center Model DR11-B interface). When data from all 32 de-
of a ring. The flow chamber exit-window allows a tectors have been transferred to memory, the master
maximum scattering to be measured (5).
angle of 30#{176} control unit interrupts the computer program so that
Because the detector rings span 1800 in azimuthal the light-scatter event can be accepted. When an
angle, the rings are not sensitive to cell rotation 8192-work memory buffer has been filled with events,
about the detector axis. The importance of signal the buffer is written onto a disk memory. Approxi-
variation on a given ring with cell rotation about the mately 250 is are needed to acquire an event and to
flow stream axis has not been determined. The lower transfer its scatter pattern data to the computer.
half of the array contains 32 wedge-shaped detectors The logarithmic amplifier consists of an operation-
and is not used in this application. The entire appa- al amplifier network in a configuration to convert the
ratus rests on a pneumatically supported vibration current signals from the photodiode rings to propor-
isolation table (Newport Research Corp., Fountain tional voltage signals. Linear gain factors of 1, 3, and
Valley, Calif. 92708; Model RS-46-12) that reduces 10 and logarithmic spans of 1.5 and 3 decades are se-
microphonics and improves the system signal-to- lected by a front panel switch. A unique feedback
noise ratio. system is incorporated into this amplifier to allow
measurement of a scattered light pulse amplitude in
Data Acquisition and Storage the presence of an ambient light intensity a thousand
Figure 2 is a block diagram of the data-acquisition times greater than the pulse. When a cell passes
electronics. Each photodiode detector element is con- through the laser beam, light is scattered away from
nected to the input of a fast logarithmic amplifier the central photodiode disk (ring 1), producing a neg-
(20), followed by a peak sense-and-hold circuit. As a ative signal from this element. This signal is inverted
cell passes through the laser beam, each detector ele- before amplification and constitutes a quasi-extinc-
ment senses a pulse of scattered light. One of the ele- tion measurement.
ments is selected as a trigger to indicate the presence
of a cell in the laser beam. When the trigger signal ex- Data Analysis
ceeds a pre-set threshold as a cell enters the beam, A mathematical clustering algorithm has been de-
the master control unit activates all of the peak veloped for analyzing the 32-parameter data (22).
sense-and-hold circuits, so that the peak pulse This program examines the raw event file and finds
heights are captured and held. When the trigger the “hills” in a 32-dimensional space analogous to

1298 CLINICAL CHEMISTRY, Vol. 21, No. 9, 1975

those in a two-parameter space. The light-scattering Once a set of data has been clustered, the next
data stored on the computer disk contain no a priori problem is that of finding boundaries that character-
information about the types of cells that produced ize each of the clusters. Such boundaries allow a data
the data, or even any indication of how many cell point to be classified into one or another cluster
types are involved. The light-scatter signature of an based only on its position relative to the boundary
individual cell represents a point in some N-dimen- surface. The use of linear boundary surfaces is a clas-
sional space, where N may be as large as 32. The algo- sical technique in pattern recognition. This method
rithm is designed to determine whether these points relies on the use of linear surfaces of one dimension
fall into N-dimensional clusters and to determine the less than the space in which the points to be sepa-
number of such clusters. rated lie. For example, lines are used to separate
The clustering algorithm first produces a density classes of points in two dimensions, and planes are
estimate at each data point, where the density at a used in three dimensions. If, for two classes of points,
point reflects the frequency with which data points there exists a linear surface such that all of the first
occur in its vicinity. For example, one might estimate class lie on one side and all of the second class lie on
the density at a given point as the number of data the other side of the surface, then the two classes are
points (light-scatter signatures) lying within some said to be linearly separable. Even when the classes
fixed radius of the given point. This algorithm used of points are not linearly separable, the linear separa-
the “K-nearest neighbor” method, which estimates tion method can be used to generate a partial separa-
the density at a point by using the distance from the tion of the classes. In such cases, two types of classifi-
K-nearest of the other data points to the point in cation errors can occur. If the two classes being sepa-
question. The value of K is determined by the experi- rated are labeled “1” and “2,” an error may consist of
menter and is typically between 10 and 20 for this ap- either misclassification of a class 1 point as class 2 or
plication. The Euclidean distance function is used to of a class 2 point as class 1. The error rate for a given
measure the distance between two points in N-space. linear separation may be measured either as the over-
The density at each data point may be regarded as all fraction of errors made or, if one type of error is
the “height” of the point above the N-dimensional more serious than the other, as a weighed sum of the
space in which the data values lie. One can then con- fractions of errors of the two types. The “fixed incre-
sider the data points as points on a density surface in ment rule” method for linear separation (23) may be
N + 1 space, whose peaks are regions of high density used to search for a linear separation plane which
and whose valleys are the regions of low density. The minimizes the error rate. This linear separation algo-
algorithm finds the hills on this density surface. A rithm and an algorithm that finds a minimal set of
cluster is defined as the collection of all data points dimensions needed for a separation have been imple-
that lie on a given hill; each cluster corresponds to mented for the analysis of the light-scattering data.
one hill on the density surface.
The algorithm finds the hills in the following man- Cell-Sorting Procedure for Multiangle Data
ner. The data points are examined in order of de- Cell sorting based on the cluster analysis has not
creasing density. The first cluster is started with the yet been attempted. When it is implemented, data on
first point, the point at the top of the tallest “hill.” several thousand cells will be acquired and stored on
For each succeeding point, the algorithm determines the disk as described above, without any attempt to
whether there are any previously classified points sort the various cell populations. Then the clustering
among the K-nearest neighbors to this point. If so, algorithm will be applied to the data to find the sig-
the point is placed in the same cluster as its nearest nificant clusters, and linear boundaries characteriz-
classified neighbor. If not, a new cluster is started ing each of the clusters will be found by the linear
with this point. When the point at the top of the sec- separation algorithm. The dimension-reduction pro-
ond tallest hill is reached, the algorithm will find that gram will then eliminate those scattering angles not
this point does not lie in the K-neighborhood of any needed for the separation of the clusters. Data acqui-
previously classified point, and a new cluster is start- sition will then be resumed, and each scatter pattern
ed. The algorithm proceeds downwards, growing new will be checked for position relative to the boundary
clusters for each of the hills it finds until it reaches a surfaces. If the conditions for membership in the
saddle point between some number of hills. If any of class being sorted are met, an appropriately delayed
the adjacent hills is only slightly higher than the sad- sorting pulse will be issued by the program through
dle, the low hill is merged into the highest of the hills the interface to the sorting hardware.
neighboring the saddle. The distinction between a
low hill and a high hill is defined by the value of a Display Processing
merge criterion that is determined by the experi- The data-acquisition program produces a raw
menter. Small bumps on the density surface will be event file on the disk. Some of the light-scatter pat-
merged into the larger hills on which they lie. The terns from this file for a mixture of 8- and 10-zm di-
clustering algorithm groups data points into clusters ameter polystyrene spheres are displayed in Figure
without any prior knowledge of the process that pro- 3A. Three decades of log light-scattering intensity
duced the data points. are displayed in 256 channels on the ordinate vs. the

CLINICAL CHEMISTRY, Vol. 21, No. 9, 1975 1299

 A Table 1. Mean Scattering Angle at the Center
of Each Ring and the Total Polar Angle Subtended
by a Cell at Each Ring for a Cell-to-Detector
Distance of 30 mma
Total angle
a’ Ring no. Mean angle (#{176}) subtended (#{176})
0
1 0.00 0.14
2 0.28 0.11
3 0.43 0.11
4 0.58 0.11
5 0.74 0.12
6 0.89 0.12
7 1.1 0.13
B
8 1.2 0.14
9 1.4 0.16
10 1.6 0.18
11 1.9 0.20
12 2.1 0.23
13 2.4 0.26
a’
0
14 2.7 0.30
15 3.1 0.34
16 3.5 0.39
17 4.0 0.44
18 4.5 0.51
19 5.1 0.58
9 14 ‘9 24 20 5.7 0.66
Ring Number 21 6.5 0.75
Fig. 3. A: 100 randomly selected light-scatter patterns from a 22 7.3 0.83
mixture of polystyrene spheres 8- and 10-jm in diameter 23 8.2 0.93
The log light-scatter Intensity Is displayed on the ordinate, the detector ring 24 9.2 1.0
number on the abscissa. The scattering angle Is given interms of ring num-
bers in Table 1. About three decades of Intensity are shown. For the micro-
25 10.3 1.1
sphere measurements presented here, rIngs 1-4 and rings 29-32 were not 26 11.6 1.2
used 27 12.9 1.4
28 14.4 1.5
B: Cluster analysis of the microsphere mixture
Each cluster is enclosed by a paW of lines. Each solid line represents an ex- 29 15.9 1.6
cursion of one standard deviation from the center of the 1O-zm diameter 30 17.6 1.7
cluster, and each dashed line is that for the 8-am diameter cluster 31 19.3 1.8
32 21.1 1.8
aThe angles have been corrected for refraction effects of the 3-mm
thick quartz window whose inside face is 7 mm from the cell.

detector ring number on the abscissa. Two distinct

clusters are identified by the clustering algorithm. In
Figure 3B, each cluster is enclosed by a pair of lines
where each line represents an excursion of one stan-
dard deviation from the center of a cluster. Table 1 then screened for cytological study. The remainder of
gives the mean scattering angle at the center of each the specimen was centrifuged at 29 X g for 5 mm.
ring and the total angle subtended by a cell (at the The modified Davis solution was aspirated, and 5 ml
sample stream-laser beam intersection) at each ring of a 100 ig/ml solution of the DNA fluorochrome, mi-
for a cell-to-detector distance of 30 mm. The angles thramycin (24), was added to each specimen. The
have been corrected for refraction effects of the specimens were aspirated and expelled from a syringe
3-mm thick quartz window whose inside face is 7 mm four times each, with a 21-gauge needle (25), and al-
from the cell stream. Table 1 was used to determine lowed to stand for at least 20 mm before instrumen-
the scattering angle for Figures 3-10. tal analysis. Staining was done so that the cellular
DNA content could be determined by a fluorescence
Gynecological Specimens, Cell Preparation technique in a separate experiment.
Cervical and vaginal specimens were collected by Results
the Dacron swab technique and placed in 5 ml of
modified Davis solution (Dr. W. H. Tolles, SUNY, Gynecological Specimen Analysis
Brooklyn, New York, private communication). The For the data presented in this section, ring 1 at 00
Dacron swabs were dipped vigorously
up and down to was a quasi-extinction measurement, and rings 29-32
remove adhered cells, and 100 1d of the resulting were not used, so that the largest scattering angle was
specimen was used for a cytocentrifuge preparation, 18#{176}.
In all of the following figures, approximately
which was stained by the Papanicolaou method and three decades of log light-scatter intensity are dis-

1300 CLINICAL CHEMISTRY, Vol.21, No. 9, 1975

 0’
0

C
to is
Ring Number
Fig. 4. A: Some of the light-scatter patterns from a normal
gynecological specimen. B: The three clusters identified by
the clustering algorithm
Each line represents an excursion of one SD from the mean of one of the a’
0
clusters. The clusters are numbered 1 (short dashed lines), 2 (solId lines),
and 3 (long dashed lines)

A ‘0 15
Ring Number

Fig. 6. A, Overlay of clusters 1 from the two normal speci-

mens; B, overlay of clusters 2 from the two normal samples;
and C, overlay of clusters 3 from the two normal samples
0’
The similarity of each of these sets of clusters implies that these ttee clus-
0 ters are characteristic of a normal sample. The solid lines are for the first
normal sample, the dashed lines for the second

I
II played on the ordinate, and ring number is given on
the abscissa. Where collections of light-scatter pat-
terns for individual cells are shown, 100 randomly se-
lected patterns are always displayed. Three hundred
randomly selected cells were used for each cluster
analysis.
Some of the light-scatter patterns for individual
0’
0
cells from two normal gynecological specimens are
shown in Figures 4A and 5A. Each line represents the
scatter pattern from one cell. The three clusters iden-
tified by the clustering algorithm are shown in Fig-
ures 4B and 5B, respectively. Each line represents an
excursion of one standard deviation from the mean of
‘5
Ring Number
one of the clusters. The clusters are numbered 1
(short dashed lines), 2 (solid lines), and 3 (long
Fig. 5. A: Some of the light-scatter patterns from a second
normal gynecological specimen. B: The three clusters identi- dashed lines). In earlier work on two-angle light scat-
fied by the clustering algorithm tering from gynecological specimens and leukocyte
The clusters are numbered as in FIgure 48 preparations (unpublished), it was observed that leu-

CLINICAL CHEMISTRY, Vol. 21, No. 9. 1975 1301

 8’ a’
0

8’
a’
0

a’
0

5 tO iS 20 25
Ring Number
5 to Is
FIg. 7. Some of the light-scatter patterns from the gynecologi- Ring Number
cal specimens of three individuals diagnosed as having invas-
ive carcinomas Fig. 9. Overlays of clusters 1, 2, and 3 from the first normal
gynecological sample (solid lines) and those from the invasive
carcinoma samples (dashed lines), respectively
The good overlap of these clusters from the normal and abnormal samples
implies that these clusters intheabnormal sample represent normal cells

.,

kocytes scattered less light at a given angle than did

exfoliated cervical cells. This implies that cluster 3
represents the scatter patterns from leukocytes in the
normal gynecological specimens. The other two clus-
ters may then represent two types of normal exfoliat-
\ -

ed squamous cells: superficial and intermediate cells.

f ....-..‘
/ ‘-
..
- . S:::A
- %.)%.
:A.
I! #{149}‘‘r - tt’#{149}I. In Figure 6, each of the three clusters in the first nor-
‘S.... I-, ‘ 1t
Jr1 ‘. -
mal specimen are overlaid on the corresponding clus-
I A ters from the second normal sample. The good over-
I I I84
lap implies that these three clusters are common to
normal gynecological samples.
5 0 15 20 25 A collection of light-scatter patterns for cells from
Ring Number three gynecological samples diagnosed as invasive
Fig. 8. The four clusters resulting from the cluster analysis of carcinoma is shown in Figure 7. A merged file of
the combined light-scatter data from the three invasive carci- light-scatter patterns was created by combining the
noma samples
Clusters 1, 2, and 3 are enclosed by short dashed ‘1es (top), solid lines, and data from the three invasive carcinoma samples to
long dashed lines (bottom), respectively. Cluster 4 is enclosed by dotted lines obtain better statistics on each of the resulting clus-

1302 CLINICAL CHEMISTRY, Vol. 21, No. 9, 1975

 A Table 2. Fraction of Cells from Two Normal
(Ni and N2) and Three Invasive Carcinoma
(Id, 1C2, and lC3) Gynecological Specimens
Classified into Each of Three Categories According
to Light-Scatter Patterns from Individual Cellsa
Fraction of
8’ Normal Abnormal squamous
Speci- Leukocytes squamous squamous cells
men (%) (%) 1%) abnormal (%)
Ni 11.0 88.0 1.1
N2 4.5 93.5 2 2.1
lCl 6.5 86.5 7 7.5
1C2 26.5 65.0 8 11.6
1C3 59.5 6.5 34 84.0
B
a Note that the fraction of squamous cells that are abnormal (col-
umn 5) could be used to discriminate between normal and abnormal
samples.

0’
0

ous as an error in the other direction, since no abnor-

mal cells are observed in the normal specimens. By
this criterion, 5% of the abnormal cluster is misclassi-
5 10 ‘5 20 25
Ring Number fied into cluster 2, while 1% of the normal cluster 2 is
misclassified into the abnormal cluster. Similarly,
Fig. 10. A: cluster 4 from the invasive carcinoma samples
Since this cluster did not appear In either normal sample, one can speculate 15% of the abnormal cluster is misclassified into nor-
that It represents abnormal squamous cells mal cluster 3, and 4% of cluster 3 is misclassified into
B: some of the scatter patterns from cells belonging to
the abnormal cluster.
cluster 4 When the dimension-reduction program is applied
to the data, it is found that only 11 of the 28 dimen-
sions (corresponding to 11 of the 28 rings) are needed
to effect linear separation with the above error rates.
ters. The algorithm identified four clusters in these Based on the data presented here, these two separa-
data, as shown in Figure 8. Clusters 1, 2, and 3 are en- tion planes are sufficient to characterize three popu-
closed by short dashed lines (top), solid lines, and lations of cells in a gynecological sample: leukocytes,
long dashed lines (bottom), respectively. Cluster 4 is whose scatter patterns lie “below” both planes; ab-
enclosed by dotted lines. Three clusters in the abnor- normal squamous cells, whose scatter patterns lie be-
mal sample overlapped quite well with those from the tween the planes; and normal squamous cells, whose
normal samples, as shown in Figure 9. This good scatter patterns lie “above” both planes.
overlap implies that these clusters in the abnormal The simple mathematical descriptions of the two
specimens represent normal cells. The fourth cluster planes enable the computer to quickly classify any
is assumed to correspond to abnormal cells in the in- scatter pattern into one of these three groups. Two
vasive carcinoma samples. This cluster is shown in hundred randomly selected cells from each of the five
Figure bA, and some of the light-scatter patterns gynecological samples considered in this preliminary
from this cluster are shown in Figure lOB. analysis were classified in this manner as either leu-
The next step in the analysis involves the separa- kocytes, normal squamous cells, or abnormal squa-
tion of clusters of scatter patterns by linear boundary mous cells. The results are summarized in Table 2.
surfaces. Of the four different classes of scatter pat- When the sorting hardware is operational, it will be
terns identified by the clustering algorithm, two possible to accomplish this classification in real time
classes appear to correspond to normal squamous and to physically separate the cells in each class for
cells, one class to leukocytes, and a fourth class to ab- positive identification by a pathologist.
normal squamous cells. The linear separation algo-
rithm discussed in the data-analysis section is used to Discussion
find the boundary planes separating the abnormal The preliminary evidence
given here indicates that
cluster (Figure 1OA) from clusters 2 and 3 of Figure 4. a gynecological specimen
diagnosed as invasive carci-
Because there is some overlap between the clusters, noma cai be distinguished from a normal gynecologi-
perfect linear separation is not possible. The mis- cal specimen based only on the angular distribution
classification of a scatter pattern from one of the nor- of light scattered by individual cells from each speci-
mal clusters in Figure 4 as a member of the abnormal men. Although the research instrument described
cluster 4 in Figure 8 is judged to be four times as seri- herein is probably unsuitable for day-to-day use in

CLINICAL CHEMISTRY, Vol. 21, No. 9, 1975 1303

clinical laboratory, this technique may have broad tered by bacteria and similar-sized biological objects. J. Theor.
Biol. 18, 133 (1968).
clinical applicability. Once a set of scattering angles
11. Koch, A. L., and Ehrenfeld, E., The size and shape of bacteria
adequate to distinguish an abnormal gynecological by light scattering measurements. Biochim. Biophys. Acta 165,
specimen from a normal sample has been identified, 262 (1968).
a hardwired flow-system instrument (or one based on 12. Cram, L. S., and Brunsting, A., Fluorescence and light scatter-
ing measurements on hog cholera-infected PK-15 cells. Exp. Cell
a microprocessor chip) could be constructed to in-
Res. 78, 209 (1973).
clude an inexpensive 2-mW helium/neon laser, an 13. Meyer, R. A., Haase, S. F., Podulso, S. E., and McKhann, G.
array of single photodiodes mounted at the appropri- M., Light scatter patterns of isolated oligodendroglia. J. Histo.
ate angles, and a set of gated scalers to tally the frac- chem. Cytochem. 22, 594 (1974).
tion of abnormal cells. Flow rates of 60 000 cells/mm 14. Brunsting, A., and Mullaney, P. F., Differential light scatter-
ing: A possible method of mammalian cell identification. J. Colloid
would ensure good statistics for the analysis. The ma- Interface Sci. 39,492 (1972).
chine could be calibrated by comparison with cyto- 15. Brunating, A., and Mullaney, P. F., Light scattering from coat-
technician counts. ed spheres: Model for biological cells. Appi. Opt. 11,675 (1972).
16. Kopp, R. E., Lisa, J., Mendelsohn, J., et al., The use of coher-
ent optical processing techniques for automatic screening of cervi-
This work is being performed (a) under interagency agreement calcytologic samples. J. Histochem. Cytochem. 22, 598 (1974).
YO1-CB-10055 between the Energy Research and Development 17. Steinkamp, J. A., Fuiwyler, M. J., Coulter, J. R., et al., A new
Administration and the Laboratory of Pathology of the National multiparameter separator for microscopic particles and biological
Cancer Institute and (5) directly under the auspices of the ERDA. cells. Rev. Sci. Inst rum. 44, 1301 (1973).
IS. Salzman, G. C., Crowell, J. M., Martin, J. C., et al., Cell classi-
References fication by laser light scattering: Identification and separation of
1. Wyatt, P. J., Differential light scattering: A physical method for unstained human leukocytes. Acta Cytol. (1975), to be published.
identifying living bacterial cells. Appi. Opt. 7, 1879 (1968). 19. Crowell, J. M., Salzman, G. C., Mullaney, P. F., and Martin, J.
C., High-speed optical analysis of microscopic particles. In: Pro-
2. Latimer, P., Moore, D. M., and Bryant, F. D., Changes in total
ceedings of the Technical Sessions at Electro-Optics ‘74 West and
light scattering and absorption caused by changes in particle con-
the International Laser Exposition, San Francisco, California (No-
formation. J. Theor. Biol. 21, 348 (1968). vember 5-7, 1974). [Obtainable from Industrial and Scientific
3. Serber, B. A., and Latimer, P., Extinction efficiencies of large Conference Management, 222 West Adams Street, Chicago, Ill.
latex spheres. J. Colloid Interface Sci. 23, 509 (1967). 60606.]
4. Latimer, P., and Tully, B., Small-angle scattering by yeast cells. 20. Model 74 Quad-Logarithmic Amplifier and Peak Sense-and-
A comparison with the Mie predictions. J. Colloid Interface Sci. Hold Circuit, Los Alantos ScientificLaboratory drawing No, 4Y-
27,475 (1968). 89754 (1974). U. S. Atomic Energy Commission patent, Docket No.
5. Mullaney, P. F., Van Dilla, M. A., Coulter, J. R., and Dean, P. 5544-777. [The instrumentation drawings given in Refs. 20 and 21
N., Cell sizing: A light scattering photometer for rapid volume de- are available from the National Technical Information Service, U.
termination. Rev. Sci. Inst rum. 40, 1029 (1969). S. Department of Commerce, 5285 Port Royal Road, Springfield,
6. Mullaney, P. F., and Dean, P. N., Cell sizing: A small-angle Va. 22151.1
light-scattering method for sizing particles of low relative refrac- 21. Model 75 Fast ADC, Los Alamos Scientific Laboratory draw-
tive index. Appi. Opt. 8, 2361 (1969). ing No. 4Y-168406 (1975).
7. Mullaney, P. F., and Dean, P. N., The small-angle light scatter- 22. Goad, C. A., A clustering algorithm for mixtures of monotone
ing of biological cells, theoretical considerations. Biophys. J. 10, densities. IEEE Trans. Inf. Theory (1975), submitted.
764 (1970). 23. Minsky, M., and Papert, S., Perceptrons, An Introduction to
S. Arndt-Jovin, D. J., and Jovin, T. M., Computer-controlled cell Computational Geometry. M.I.T. Press,Cambridge, Mass., 1969.
(particle) analyzer and separator. Use of light scattering. FEBS 24. Crissman, H. A., and Tobey, R. A., Cell-cycle analysis in 20
Lett. 44, 247 (1974). minutes. Science 184, 1297 (1974).
9. Koch, A. L., Some calculations on the turbidity of mitochondria 25. Wheeless, L. L., Jr., and Onderdonk, M. A., Preparation of
and bacteria. Biochi,n. Biophys. Acta 51,429 (1961). clinical gynecological specimens for automated analysis. J. Histo-
10. Koch, A. L., Theory of the angular dependence of light scat- chem. Cytochem. 22,522 (1974).

1304 CLINICAL CHEMISTRY, Vol. 21, No. 9, 1975