0022- 1554/79/2701
THE JOURNAL Copyright ©
-0635$02.OO/O
OF HISTOCHEMISTRY 1979
The
by
AND
BioPEPR:
Society,
Inc.
A System D.
J.
ZAHNISER,
Physics
Vol. 27, No. 1, pp. 635-641, 1979 Printed in (ISA.
CYTOCHEMISTRY
Histochemical
P.
5.
Laboratory,
for
the
OUD,
M.
Automatic C.
Department
T.
Prescreening
RAAIJMAKERS,
ofPathology,
Received
G.
University
for publication
P.
of Cervical
VOOYS
ofNijmegen
June
R. T.
AND
Ntjmegen,
VAN
The
Smears’ WALLE
DE
Netherlands
23, 1978
A feasibility study has indicated that a Precision Encoding and Pattern Recognition (PEPR) cathode ray tube prescreening system for cervical smears can be both accurate and fast. Smears are prepared using a syringing technique and are stained with a Feulgen-type nuclear stain and a protein counterstain. The use of film as an intermediate step between the cells and Bio PEPR allows the scanning of fields as large as 8 x 8 mm. The morphological features of the cells are measured as directed by a hierarchical decision strategy. Additional programs detect artifacts, overlaps, and leukocyt.es. For clean samples, false positive and false negative rates on the cell level have been obtained that will allow acceptable smear level rates (10% false positive, 1% false negative). These rates have been reached without compromising the required speed goals of 120 to 180 smears per hr. The efficiency of the system is dependent on the quality of the smears. Measurements on a set of 192 routinely prepared smears indicate acceptable false negative rates and a false positive rate of about 18%. A reduction of this rate is expected with small improvements in cell preparation and measuring software, leading to the overall system efficiency required for commercial feasibility.
A modified Precision Encoding and Pattern Recognition (PEPR) cathode ray tube (CRT) scanning system is being used to study the feasibility of developing a fully automatic
more, field,
system
devices
screening
on
of high-
diagnostic photograph
cytology their
for the
have
been
energy negatives
prescreening
in use
physics made
at Nijmegen, general image basic features medical resolution
the
BioPEPR, system.
of a PEPR
analysis larger
early
smears.
1960’s
field
but
also
developed not many
into only
(1, 6, 7) while scanning systems. Furthermore,
the
ratory This
work
scanning
of
according became of the
with
cervical
cell
photonegatives
to
the
apparent glass slides
measurements of
Papanicolaou
fields
from
procedure
with
stained
(PAP).
It
need
to
I Project Wilhelmina
be scanned supported Fonds.
use of film simplifies reduces the number
to between by the
Netherlands
4 and
6 per Cancer
Society,
current
decentrally send the
state
smear.
sensitive
of
prepare resulting
and film
smears to each
with labo-
of development
and
and reports on the status cells and entire smears. A
photography,
scanning
wherein discussed:
hardware,
and
METhODS
AND
Ideally, a smear for automatic of cells at a reproducible density, to cellular
For
a number
A list of suspect then be returned
system, single
of each decrease
abnormalities,
and
high
scanning minimal nuclear:
cytoplasmic and cytoplasmic: background contrasts. Initial work with PAP-stained smears has shown the PAP technique to be unsuitable for automatic analysis. The PAP stain was not highly reproducible, and the contrast was not high enough. In this section, the methods currently used to prepare specimens for scanning are described. Upon collection, cells are preserved in a cold (4#{176}C) modified phosphate-buffered saline solution containing 1 10 mM NaCl: 27 mM Na2HPO4-2H2O: 6mM KH2PO4: 10% ethanol. Ordinarily, cells are left overnight at 4#{176}C before further handling. To break apart cell clusters, an automatic device based upon a peristaltic pump is used to force the cell suspension continuously through a loop of plastic tubing with 19-gauge syringe needles at both ends. This procedure is continued
soon
the scanning of fields that smear.
cell that
of the analysis is first presented basic aspects of the system are
preparation: a monolayer
a stain
artifacts,
that the scanning of photonegatives instead could lead to a significant improvement in
directly. The significantly
the
one
could and then
system. would
BioPEPR on both
analyze
it is envisioned
laboratories own slides
preparation, software.
Specimen should have
to
scale,
MATERIALS
the speed of the screening system, primarily because large fields could be scanned with the combination of a 9-inch CRT and 70-mm-wide roll ifim. Fields up to 64 mm2 can be scanned, as compared to fields of less than 1 mm2 with devices that scan slides device and
needed
after scanning. paper describes
specimen computer
the
smears
time a large
general summary each of the four
fields signifithe scanning
began
total
operation of the of measurements
and high of
of large fields plus the use of two hardware scanning modes and a hierarchical decision strategy have resulted in a system speed approaching that of the flow systems (4, 9) Because the system was originally designed for the scanning of film,
the
to a central scanning possible abnormalities
a
additional
from similar devices used in physics The BioPEPR system is capable of the multiple parameter analysis characteristic systems than such
in
measuring of photo(5). A PEPR system
has been It incorporates
system
PEPR
in the
for the scanning and from bubble chambers
adopted research. and
image cantly
since
renamed analysis
features
of cervical
all handling of the slides, as well as the focusing is done off-line, again resulting in a significant
FurtherKoningin
for
15 mm
various
at
chemical
a velocity
of 55
and enzyme
635
Downloaded from jhc.sagepub.com by guest on August 14, 2015
ml/min.
techniques
In
addition
are being
to
this
explored.
method,
Dithio-
ZAHNISER
636 threitol, glycine, hyaluronidase, but none has yielded significantly
and EDTA improved
have been tested disaggregation.
(2, 3),
During the next step of the preparation, a sample of the cells is counted in suspension using a Coulter counter. The counter thresholds are set to count only cells larger than leukocytes. The sample is then centrifuged, the supernatant is removed, and enough carbowax
ET
AL.
Photography: Photonegatives fication of ten using a Ultra Micro large
field
lens.
To
obtain
very narrow bandwidth nm) light to give better
of the cells are made at a magniNikkor 28mm fl.8 high resolution,
optimal
results,
monochromatic
is used. The lens than 1 t resolution
light
is corrected over
of a
for green
a field
of up
(545 to 8 x
8 mm.
of cells than when using the carbowax method. After the cells have been deposited, the slides are air dried, and the cells are fixed in a methanol based fixative (85 parts methanol, 10 parts 40% formaldehyde, and 5 parts acetic acid). The cells are then stained using the Feulgen procedure with thionine (8) as the Schiff reagent. This stain is DNA-specific and has proved to be very reliable. Feulgen hydrolysis is performed for 1 hr using 5 N HC1 at room temperature. The cells are counter-stained using Congo red (0.25% in
A further advantage of the use of a photonegative for scanning is the relatively large depth of field of the lens used (± 4 z). This results in less focusing problems, which is especially important when using automatic focusing. In practice, this depth of field has proven to be large enough to tolerate the variations found within an 8 x 8-mm area of a standard microscope slide. At such a low magnification, it is necessary to use an extremely high resolution film. The Kodak Spectroscopic film type 649-GH has been used with great success. This film is capable of resolving 2000 lines/mm with high contrast objects. Using the present camera system, photonegatives are made on 70 mm wide roll film. The slide is held by a standard microscope table to allow easy positioning of the desired field. The film is held by a vacuum system for optimal flatness. A black and white television system monitors the field; this sytem is used for positioning the field as well as for automatic focussing of the image. Hardware: The basic hardware is modelled after the PEPR CRT system used in scanning bubble chamber photographs. A block dia-
70%
gram
(2%
polyethylene
glycol
in 50%
ethanol)
solution
is added
to ensure
a final distribution on the slide on the order of5O epithelial cells/mm2. An aliquot of the cell suspension is then placed on a glass slide, and the cells are smeared over the surface by hand using the side of a glass pipette. These techniques have consistently resulted in smears with at least 50 to 70% single cells and with much less than 1% nuclear overlap. Methods of centrifuging the cells directly onto the slide are being experimented with, but so far they have not yielded results better than smearing the cells onto the slide by hand. Precoating the slide
with
polylysine
ethanol
with
has
0.25%
also
acetic
been
acid).
tried
The
but
resulted
staining
in a greater
results
loss
in a brown-
This stain gives good results when photographing the preparations and also yields a color suitable for scanning by an analyst. Thionine and Congo red have absorption maxima at 600 and 500 nm, respectively. Measurements are made between these two maxima at a wavelength of 545 nm. ish-blue
nucleus
and
an orangish-red
cytoplasm.
FIG.
1.
Hardware
is shown
in Figure
1. In the
modified
BioPEPR
configuration
a
green phosphor CRT and a 2:1 demagnifying lens is used. Spot size on the film is on the order of 10 to 12 t. This spot size, combined with the xlO magnification onto the film, results in an effective 9-inch
system
random
configuration
resolution
access,
of about
allowing
of BioPEPR.
Downloaded from jhc.sagepub.com by guest on August 14, 2015
1 s. The
instantaneous
device
is capable
positioning
of completely
of the spot
to any
AUTOMATIC point
in the
field.
The
speed
of operation
PRESCREENING
is computer
controlled
OF up
to approximately 5 /zsec (in the cell plane). Two measuring modes are used. The first, the scanning mode, sweeps the spot of light and detects the crossing of a preset darkness threshold as light transmission through the film changes. Using this mode, the widths of objects can be measured extremely rapidly. During
the
sweep
the
is automatically relatively the
low
sweep
the
mode
is placed
second
light
the
by
a circuit
light
level.
into
mode output
of the
film
continually
tracks
the
two
one
buffer
an analog-to-digital conversion, obtained. In practice, the grey on the order of 64 levels. to a PDP 1 1/40 computer with
28K
of
disks
data
storage.
16-bit
This
for
the
on
the
film
photomultiplier
over
a period
terminal, directly
scan
darkness
at
used an
TV
film.
for
The
the
and
output
console
display,
and
latter
operates
on the TV and with
program
interactive
a graphics the
simultaneously of the
are
includes
a TV by
the CRT
of the
and
measuring
photomultiplier.
mode,
is digitized.
RKO5
hardware
a computer
a raster
Additional
transmittance
Two
looking
modulating is
memory. Additional
of
monitor
generating
BioPEPR
637
SMEARS
10 psec, and subsequently performing 256 levels of digital information are level discrimination of the device is The BioPEPR system is attached
consisting
during
memories. while
point
points
of the
recorded
sweep
the
discrete
signal
background
that of
from sweep.
of operation, at
of the
Information
one
of the data from a new
transmittance
integrating
darkness
background
allows the analysis busy collecting data
In the
varying
subtracted
CERVICAL
This
By
output
of
the
photomultipliers
measurement CRT
monitor
is then
of the
CRT
to
correct
used
phosphor. for
slowly
to correct
the
light
output
for any
of the
changes
A signal
is fed
back
varying
light
level
to the
in the
CRT. light
cathode
changes
and
e .
,‘
r
‘
‘I
#{149}‘
6,
,‘
R( ..
,,-
,.
4.Hb:/
.
0
#{149}
.
-,
.
.
“S
‘
V iF:*I..4;e
.
.
.
.
,.
.
4
.
.
-.
A
S
FIG. 2. BioPEPR measurements width ratio (H/W) is ten times the number of nuclei detected
for a) a superficial the actual value; and within the cytoplasmic
epithelial cell and density measurements area.
b) a moderately dysplastic cell. are in nonstandard, BioPEPR
Downloaded from jhc.sagepub.com by guest on August 14, 2015
Area measurements units. Overlap
are in correction
height: indicates
ji2;
of the
638
ZAHNISER
ET
AL.
NA
NA
NA 0
(IC
1 LC
#{149}
EP
UN 0
A2
100%
ON
III.II
0
OR
NA
IC
IC
AT
lIT.
0
0l\
0
Al,
NI)
0
UN ()
NI OF
0
NI
NI
NC
NI
0
0
0
OK 0
NC 0
OK 0
OF
AS
NC
0
0
0/
0
1
OK
0
AS
0
0
AS 0
II
i),’
AS/TOT
0
UN
NC
OK AB/TOT=
UN
1
NO
B
A
NA
UI,
A.,
‘SI
NA
NA
0
Ii:
II
I IN
A.’ ON
IF’ ..)./.
0#{149}/.
NIl
II” C)
NI
0
OK
NI”
0
I
0/
Ilk I
Up\ =
II I (I
NC 1
AS/TOT
I’ I
I III
NI
05 0
III
III’.
I I).,
TI’
1)
AS 0
‘IL
,.,r,, III!
I
Al’ (I
(I
D
C
FIG. 3. Paths taken through the decision tree by a) a normal cell with a small nucleus, b) an artifact detected by the shape criteria, c) a parabasal cell, and d) a carcinoma in situ cell. NA, nuclear area; QC, quality control; LC, leukocytes; EP, epithelial cells; A2, nuclear area (after re-thresholding); UN, unknown; OK, normal; NO, nuclear overlap (shape); NI, nuclear integrated darkness; NC, nuclear: cytoplasmic
ratio;
AB,
abnormal.
output of the monitor photomultipliers is also divided into the signal of the measuring photomultiplier. This latter correction removes any high frequency deviations due to impurities in the CRT phosphor. Software: The software used for the cell analysis is patterned after the basic morphologic features used by cytologists to discriminate between normal and abnormal cells. Programs have been developed to measure nuclear area, nuclear optical density, nuclear texture, and nuclear:cytoplasmic ratio. Additional programs have been designed to recognize artifacts, overlapping nuclei, and leukocytes. The complete measurements of a typical normal and abnormal cell are shown in Figure 2. However, the measurement of all of these parameters is not necessary to identify most types of normal cells. Because of this and to optimize speed, the programs are invoked through a hierarchical tree strategy whereby only the necessary measurements are made and analyzed. Figure 3 shows the tree in its present
passed
form.
At
to the right
each
of
the
if the result
branch
points
of the program
in
the
is above
tree,
control
a given
is
value
and to the used in the Because widths and
left if the result is below that value. tree are explained in the text below. of the high intrinsic speed of BioPEPR its capability of correcting electronically
variations,
nuclear
test,
at
the
top
area of the
(NA) tree
has
been
structure.
chosen The
The
in detecting object for background
as the search
abbreviations
for
highest
priority
nuclei
and
the
measurement of the nuclear area is done using 2-z sweep intervals. Nuclei smaller than a preset size of 56 z2 (approximately equal to the nuclear area of an intermediate cell) are immediately classified as being normal. A certain percentage (3%) of the cells thus classified as normal
pass
through
they are epithelial (LC). This is done preparation
containing
a quality
an
inadequate
uable cells or too many leukocytes further screening. The leukocyte atives due to the overlapping ofan
Downloaded from jhc.sagepub.com by guest on August 14, 2015
control
check
cells (EP) or whether they by measuring the nuclear number will
be sent
(QC)
to
are in fact : cytoplasmic
see
whether
leukocytes ratio.
of diagnostically to a cytoanalyst
A valfor
check helps to minimize false negabnormal nucleus with a leukocyte.
AUTOMATIC Figure
3a shows
the
path
taken
through
the
tree
PRESCREENING by a typical
OF nuclear
normal
nuclei,
3) a normal
identify tree
having a “nucleus” the tree to ascertain
these
or 4) an
categories
abnormal
cell.
is demonstrated
The
below
process
by going
used
larger
immediately Objects they
checked exceeding
are
a 3002 as
of the
whose
nation
darkly
stained, thresholding
level
above
folded,
makes
tance
measuring
Fig. mate
size
reliable. again
the
the At
profile
where
to
set
of
the
measured,
new this
falling
a new
are
time
under
or large
not
into and cells
height:width
the
the the
using for
known,
this
intermediate
method the
a higher
resolution
cell cutoff
area
is
the smaller,
is designated
are
(OK). that
remain
FIG. 4. Frequency cutoff values used
large
after
distributions in the decision
the
re-thresholding
of the tree.
nuclear
are
integrated
sent
to the
darkness
are
occasional
of 1.25
for
(at
measurement
each of the
Using
size,
and
one
point
transmittance
as well
as average
and
times
criterion. designed
The
cells,
cells)
optical
mode
is measured.
dotted
higher
darkness
lines
cutoff
times
and
density
measuring density
half
to
(parabasal
metaplastic
The
one
might through
of the integrated
cells. tree.
two
that path
are
measures
distributions in the
half
and
the
darkness
tree
nuclei
(NI)
abnormal
used
shape
the
large
columnar,
two frequency values
by this
program
integrated and
the
in
with
objects
3b shows
as the
represent
corresponds
normal,
and
the
normal.
final program : cytoplasmic
amplification low contrast
is highly
nuclear
cutoff
cells
first
of normal
to one
scans
approxi-
is fitted
the
cigar-shaped
shown
cervical
nucleus.
integrated
A
nuclear the
two
nuclear
(see
to discard
normal
The
4 shows
lower
transmitprofile
the
an
a parabola
during
programs
intermediate,
a sample
to
algo-
optimal and
the
darkness
point
in the
position
sim-
the
a fixed the
for
in which
obtained
in order
between
a 1- step
Figure
have
data
remaining
cells.
within with
in the
a rethresholding
threshold
using
normal
always
some
abnormal
discrimi-
confusing
points
and
further.
cytoplasm
almost
is determined,
scanning
clusters; are
ratio
the tree The two discriminate
is
(NO)
to the
pass the parabola check. Figure for clusters and artifacts detected
otherwise
nucleus.
tree
uses
A2 cell
nuclei
to improper
which
the
already
the
a single analyzed
inflection
threshold
be
cytoplasms;
Because
nucleus
the
normal
are
Such
of the the
for
artifacts
due
branch-point
and
to
nucleus
check
first
discrimination.
A cell now Nuclei
large
area.
mode,
After
as normal
deeper
or overlapping of
2) indicate
and
passing
nuclear
a darkness
cytoplasmic
usually
artificially
between
background.
rithm
are
cutoff large
(UN)
now
of
size
if it is too
cutoff
appear
(thresholding)
plistic
initial
unknown
“nuclei”
nuclei
measurement
first
the
to see
classified
Many
cells
than
program
cell)
639
Objects not having a round or oval shape are designated and are not analyzed further. The program also checks
unknown
to
through
of the
nucleus.
structure. A “nucleus”
fL.
cell,
with a suspiciously large area is sent if it is 1) an artifact, 2) a cluster of
SMEARS
overlap
side
cell.
An object deeper into
CERVICAL
easily
(NC) locates the cytoplasm and calculates the ratio. Because of the large size of the cytoplasm, made in steps of 2 i. Here the computer boosts the analog of the detection system to make the detection of the cytoplasm
easier.
superficial
epithelial
darker
cytoplasms
detected.
very
high
cells
falling
between
253
normal
cells
Downloaded from jhc.sagepub.com by guest on August 14, 2015
cutoff
A
integrated
and
127
two
lightly may
of the of
darkness, the
The cell 15% and
cytoplasm
diagnostically is used
a cutoff
integrated
abnormal
stained
be incorrectly for
darkness
cells.
Dotted
but
borderline
cells
those
of 24%
of an
measured, cells is used
having for
cutoffs.
The
lines
indicate
a
those former
the
640
ZAHNISER
is set
low
this
because
is most
cytoplasm of
integrated
an
two
and
cells
located
nuclei
to
ratio,
or an
lie within
is then
provide
up
structure.
cells.
Figure
cytoplasmic
Figure 3d shows
that of
S ingle
the
nuclear:
due
a normal
cell.
Using
Normal
how
many
Abnormal
the
the
typically
Total
nuclear:cytoplasmic
1270
TABLE
cells
going
deep
often
followed
taken
by a carcinoma
into
the
Smear
Sampling
strategy: and
false
identify
possible
To
estimate
negative
rates
shortcoming
a lower
of
the
Results
BioPEPR
classification
by parabasal
in the
limit
BioPEPR
programs
Analyst:
in situ
for
the
false
system,
and
Total
or decision
logic,
Normal
Suspect
rate
according
from
photonegatives
normal and two abnormal In the test measurement that BioPEPR measured types.
The
normal
parabasal, (78)
(1276)
and
mildly
dysplastic,
and
carcinoma
in situ
of nuclear
area,
has
been
is passed
left
maximum tively
shape, recorded
diagnosis
or
made
two node,
by to
cells
severely
the
dys-
measured
density,
disk
of the
a preset
and
node
cutoff
along
the
chosen This
relevant
cells
To
tree
obtain
have
classified
normal.
19) the
The
missed. ±
this
procedure
proved
efficiency
to be necessary
these
affect
program smears
most of
specimens cells,
efficiency
has begun. are presented.
32 abnormal the
smears outpatient
from patients of the abnormal
were patients
were
used
false
negative
rates,
an
iterations
of
optimal
this
cell level
rates
contain
an
overlapping
impression
full-smear
of how
measurement
the results of the first set of 192 from 160 normal smears and The
normal
according
BioPEPR.
smears to the
The
were
The
a carcinoma
remaining
smears
in
cancer.
a first
estimate
on
smear
the
or invasive
situ
The
abnormal
results
were
of the level
present
under
smears
were
cancer
from realistic
and
these
192
false
positive
shown
matrix
in Table
as normal superficial cervical
I. Of the
and cell,
for the
one
smears and
1276
normal
6 were incorrectly 3 intermediate
columnar
cell).
cells,
1270
were
results
in a false
cells, classified
cells
(3 out
of
negative
rate
for
this
sample
is
minus
and false an average
for each of 160 normal II). A fraction representing false
(truly
negatives)
optimal smears. classified were
characterized
parakeritosis,
and
the
abnormal
two one
by the
senting metrium.
plus
total
number
leukocytes and arA cutoff value of 1%
metaplasia.
The
smears dysplasia
folded heavy
remainor rolled cytolysis.
were incorrectly and the other
classirepre-
a highly differentiated adenocarcinoma of the The former was characterized by an extremely ratio,
few abnormal The analysis upon the number structure.
the
latter
was
endolow
characterized
by
cells to pass the 1% cutoff. time for each smear is of course dependent of cells which must go deeper into the tree
Typically,
probably
and
of
false
between the normal and smears, 18% were thus inOf these 28 false positive by slight abnormalities: trichi-
squamous
of the 32 abnormal representing a mild
32 abnormal number
cells,
divided
discrimination Of the normal as suspect.
and
negative of 6000
20
to
containing be necessary
photonegatives)
per
40
sec
up to scan
woman,
are
to 4000
needed
cells.
about thus
to
15,000
taking
analyze
one
In practice, cells
it will
(four
to six
1 to 2 mm.
DISCUSSION cells
is
classified
classified as abnormal cells, one parabasal cell,
This
dysplastic
ing 11 contained large clusters of leukocytes, cytoplasms, darkly stained cytoplasms, or Only fled;
calculated
incorrectly
level false positive “dirty” samples,
conditions.
on single
being
78 abnormal
cells (3 out of 36). None of carcinoma in situ cells were
of diagnostically valuable cells (excluding tifacts) was calculated for each smear.
The
measurements
32
error
6 were
3 mildly
objects
photonegative
RESULTS A confusion
smear realistic
too
above-described
abnormal
and
false
nuclear:cytoplasmic
collected
earlier diagnosed as having abnormalities. smears were dysplastic, and of those one-
endometrial
to make
they get
extensive
prepared by
dysplastic. with
To
analyzed.
measured
mildly with
artifacts.
clinic,
obtained One-half
patient
few
at the
are not clean;
and
were
and
from
to arrive
In this report, Photonegatives
method,
half
a
rates.
In practice, nuclei, cluster
from
Only
160
Of the
dysplastic or the
overall
cells were analyzed smears (see Table
monas,
obtained.
the
0. 19%,
6 included
estimate for more
false
were
28 30
3.0%.
gave the abnormal correctly smears, 17
rates
±
distribution.
as abnormal
These
point in the tree and the other representing the abnormal cells. By comparing the two distributions, and in particular their area of overlap, an “optimal” cutoff value was chosen at each node. This was continued throughout the tree and fmal cell level false positive and negative
of 0.47
and 3 moderately severely dysplastic
positives, at
reached
sample
to a binomial
abnormal-classified
interacprogram
parameter
that
as
To rates
nuclear:
in the
value.
cutoff value was graphics package. normal
for the
7.7
values
computer
at each
of
the
abnormal
a cytoanalyst.
distributions
representing
cell
control
according
frequency one
the
previously,
right
each optical
onto
four
intermediate,
The
dysplastic,
nuclear
system efficiency, each the help of a simple
with
displays
moderately
were explained
superficial,
cells.
For
from
care was taken to ensure of a wide variety of cell
included
cells.
fields
72 were
metaplastic
nuclear
ratio
the cell type
As
each
cells
included
cytoplasmic
with
smears. of these cells, enough of each
endocervical,
plastic,
containing
2
a set
cells
obtained
132
Abnormal
to
of measurements on single cells has begun. To date, 1354 preclassified normal and abnormal cells have been measured by BioPEPR. The were
78
II
Classification
Normal
positive
1276
72
ratio
cell.
one
6 6
present.
path
Suspect
Normal
measures
for
Results classification
Analyst:
to clusters
touching
paths
I
Classification BioPEPR
program The
Cell
overlapping
for
negatives
TABLE
the
the
3c shows path
estimate
are
typical
of the
AL.
cell
nuclei
3d show
indicated feature
detection
on false
area.
if additional
3c and
tree
down
already
additional
a better
cut
as a mask,
the
scaled
Figures
the
abnormal
cytoplasm
An
allows
thus
has
cell.
routine
cytoplasms
of abnormal
darkness
abnormal
measuring
cytoplasmic
the
the
likely
ET
(1 and
positive
results
representative smears; they and sion smears,
of the
false negative is that if the
measurements
ofthe results thus provide
desired
Downloaded from jhc.sagepub.com by guest on August 14, 2015
rates BioPEPR smear
done
on
that can be obtained a lower limit for the
on the cell level. is presented level
false
single from false
cells
are
“clean” positive
The primary concluwith well-prepared
positive
(10%)
and
false
AUTOMATIC
negative
(1%)
rates
apparatus and In principle, preted
level
types
assortment
of
cells,
parabasal
positive
rate
of 0.47%
with false
a much positive
rate, cells
it is important to note that and 8.3% of the moderately none
severely
us
begin
presence
From the
to of
data,
the
rate
the
of about
smear system. desired. that
derive
suggest
preparation and analysis This is approximately The experience from false
of clusters and overlaps should false positive rate. In addition, of
give
an improvement
weighting
certainty
the factor
analysis is being
metrial more sional
same
as our
test
of
sample,
For
a
a smear
cells, false
the
the
abnormality percentage
a higher negative
or carcinoma
in
screening
level, there
a slightly or moder-
false negative are missed,
rate of 8.3 to and therefore,
due
nuclei,
clusters
some
preliminary
that
a smear
level
with
present
the
false
rate
can
on
objects quality
it should
We would
like to thank
in the
efficiency of each
To date no in-depth study has been of the BioPEPR system in detecting endometrium. The system has been
rates; suspect
development
of a
measuring feasibility.
occain a
system apThe system
to screen
2 or 3 smears
per
Riet
Harbers-Hendriks
of smear
ofsmears;
the
for her expert
preparation
staffof
the
techniques
Laboratory
their
especially neering
and
of
work
on
hardware
support,
and superb
in developing
Husain for
of the science construction;
and
due group
the BioPEPR
GJ, Todd gynecologic
P, Sattilaro specimens:
25:513,
automated
system.
RF: The Chemical
systems for Cytology.
dispersal agents.
of cells from J Histochem
JA: A sample
preparation
1977
Page-Roberts
OAN,
the engifor their
CITED
PH, Wied GL: High resolution processing cancer. The Automation of Uterine Cancer of Cytology, Chicago, 1976. p 144
Tutorials 2. Escobar human
the
of Nijmegen cell samples.
Special thanks are High Energy Physics
work
in
for Cytopathology
development
the camera system. technical staffof the
work
and
cervical
BA, Millet cancer
screening.
Acta
Cytol
22:15,
1978
4. Mullaney PF, Steinkamp JA, Crissman HA, Crowell JM, Martin JC, Salzman GC, Cram 5: An introduction to flow systems for cell analysis and sorting. The Automation of Uterine Cancer as
Cytology,
5. Pless
is
Tutorials
IA: PEPR.
High
Energy
6. Poulsen
the
7.
a 8.
made of the efficiency abnormalities of the optimized for use in a
of Cytology,
1976.
p 90
of the International Conference on Dubna, 1964, p II 409 LH, Cahn RL, Louis C, Toussaint G: High
Oliver
analysis
of cervical
cells-A
progress
report.
J Histo-
chem Cytochem 25:689, 1977 Tanaka N, Ikeda H, Ueno T, Watanabe S, Imasato R: Fundamental study on automatic cyto-screening
Y, Kashida for uterine
cancer. utilizing
(CYBEST) 21:85, 1977
Van
III. the Duijn
New pattern
system of recognition
P: A histochemical method positive substances.
its use in a bi-color acid 9.
Chicago,
Proceedings
Physics,
RS,
resolution
by
be possible
development
for
a method cell
the
faculty
3.
also lead to a reduction in the a modification ofthe 1% criteria
results studied.
the
The
to endo-
detection of the is represented
and the Departments of Gynecology of the Universities and Utrecht for their help in obtaining and classifying We would also like to thank the Technical Services
of
a reliable method in suspension onto in the recognition
scrape.
method
entirely limited by the speed of the computer. computer power and/or more efficiently writ-
Cytochem
positive
be expected
warrant
LITERATURE
techniques of the BioPEPR a factor of two higher than these measurements indicates positive
not
a cervical
sampling
speed of the for commercial
ten programs, min.
1. Bartels cervical
etc.
limits
may
scrape. The that desired is now increased
ideas,
to the
of cells,
scrape
analysis that guarantees where endometrial cancer
classification
is not a large relatively few
arise
using
cervical
abnormalities complex cases
speed With
with
this are
that
program
of the
cervical proaches
641
SMEARS
ACKNOWLEDGMENTS
cells
situ
and false negative rates. criteria of 1% suspicious
be achieved
in the
inter-
making
on the
achieved.
disaggregation methods improve and found for randomly depositing the cells glass slide. Furthr software improvements
should
types,
complications
can
18% can
scale
smear
entire abnormal smear is unlikely. the 192 “complete” smears allows
positive decision
results
a reduction
level cells
overlapping
one
CERVICAL
large
15.8% of the mildly dysplastic dysplastic cells were missed,
smear because
A cell a few
examine
artifacts,
these
the
of an from
smear level false Using the sample
in a smear,
etc.
be Each
that BioPEPR will have containing only mildly
cells. On however,
misclassification The data collected to
cell
be
dysplastic
abnormal cells present. 15.8% means that only the
cells, would
missed. We can infer efficiency for smears
ately dysplastic number of cells,
scanning
should
sample.
higher percentage of parabasal level may be expected. As for
of the
existing
rates
in the
of a test statistic for deciding difficult. For a smear with the
intermediate
were lower
the
efficiency
cell
false
but
using
OF
insensitivity
of the
different
a
development of a smear
be obtained
software. the cell
in terms
contains
can
PRESCREENING
schiff
Wheeless LL, Hardy system for automated
Downloaded from jhc.sagepub.com by guest on August 14, 2015
automated method.
apparatus Acta Cytol
specific thionine-SO2 for deoxyribonucleic acid J Histochem Cytochem
JA, Balasubramanian cytopathology. Acts
reagent and and periodic 4:55, 1956 Slit scan flow
N: Cytol 19:45,
1975
THE JOURNAL Copyright ©
-0635$02.OO/O
OF HISTOCHEMISTRY 1979
The
by
AND
BioPEPR:
Society,
Inc.
A System D.
J.
ZAHNISER,
Physics
Vol. 27, No. 1, pp. 635-641, 1979 Printed in (ISA.
CYTOCHEMISTRY
Histochemical
P.
5.
Laboratory,
for
the
OUD,
M.
Automatic C.
Department
T.
Prescreening
RAAIJMAKERS,
ofPathology,
Received
G.
University
for publication
P.
of Cervical
VOOYS
ofNijmegen
June
R. T.
AND
Ntjmegen,
VAN
The
Smears’ WALLE
DE
Netherlands
23, 1978
A feasibility study has indicated that a Precision Encoding and Pattern Recognition (PEPR) cathode ray tube prescreening system for cervical smears can be both accurate and fast. Smears are prepared using a syringing technique and are stained with a Feulgen-type nuclear stain and a protein counterstain. The use of film as an intermediate step between the cells and Bio PEPR allows the scanning of fields as large as 8 x 8 mm. The morphological features of the cells are measured as directed by a hierarchical decision strategy. Additional programs detect artifacts, overlaps, and leukocyt.es. For clean samples, false positive and false negative rates on the cell level have been obtained that will allow acceptable smear level rates (10% false positive, 1% false negative). These rates have been reached without compromising the required speed goals of 120 to 180 smears per hr. The efficiency of the system is dependent on the quality of the smears. Measurements on a set of 192 routinely prepared smears indicate acceptable false negative rates and a false positive rate of about 18%. A reduction of this rate is expected with small improvements in cell preparation and measuring software, leading to the overall system efficiency required for commercial feasibility.
A modified Precision Encoding and Pattern Recognition (PEPR) cathode ray tube (CRT) scanning system is being used to study the feasibility of developing a fully automatic
more, field,
system
devices
screening
on
of high-
diagnostic photograph
cytology their
for the
have
been
energy negatives
prescreening
in use
physics made
at Nijmegen, general image basic features medical resolution
the
BioPEPR, system.
of a PEPR
analysis larger
early
smears.
1960’s
field
but
also
developed not many
into only
(1, 6, 7) while scanning systems. Furthermore,
the
ratory This
work
scanning
of
according became of the
with
cervical
cell
photonegatives
to
the
apparent glass slides
measurements of
Papanicolaou
fields
from
procedure
with
stained
(PAP).
It
need
to
I Project Wilhelmina
be scanned supported Fonds.
use of film simplifies reduces the number
to between by the
Netherlands
4 and
6 per Cancer
Society,
current
decentrally send the
state
smear.
sensitive
of
prepare resulting
and film
smears to each
with labo-
of development
and
and reports on the status cells and entire smears. A
photography,
scanning
wherein discussed:
hardware,
and
METhODS
AND
Ideally, a smear for automatic of cells at a reproducible density, to cellular
For
a number
A list of suspect then be returned
system, single
of each decrease
abnormalities,
and
high
scanning minimal nuclear:
cytoplasmic and cytoplasmic: background contrasts. Initial work with PAP-stained smears has shown the PAP technique to be unsuitable for automatic analysis. The PAP stain was not highly reproducible, and the contrast was not high enough. In this section, the methods currently used to prepare specimens for scanning are described. Upon collection, cells are preserved in a cold (4#{176}C) modified phosphate-buffered saline solution containing 1 10 mM NaCl: 27 mM Na2HPO4-2H2O: 6mM KH2PO4: 10% ethanol. Ordinarily, cells are left overnight at 4#{176}C before further handling. To break apart cell clusters, an automatic device based upon a peristaltic pump is used to force the cell suspension continuously through a loop of plastic tubing with 19-gauge syringe needles at both ends. This procedure is continued
soon
the scanning of fields that smear.
cell that
of the analysis is first presented basic aspects of the system are
preparation: a monolayer
a stain
artifacts,
that the scanning of photonegatives instead could lead to a significant improvement in
directly. The significantly
the
one
could and then
system. would
BioPEPR on both
analyze
it is envisioned
laboratories own slides
preparation, software.
Specimen should have
to
scale,
MATERIALS
the speed of the screening system, primarily because large fields could be scanned with the combination of a 9-inch CRT and 70-mm-wide roll ifim. Fields up to 64 mm2 can be scanned, as compared to fields of less than 1 mm2 with devices that scan slides device and
needed
after scanning. paper describes
specimen computer
the
smears
time a large
general summary each of the four
fields signifithe scanning
began
total
operation of the of measurements
and high of
of large fields plus the use of two hardware scanning modes and a hierarchical decision strategy have resulted in a system speed approaching that of the flow systems (4, 9) Because the system was originally designed for the scanning of film,
the
to a central scanning possible abnormalities
a
additional
from similar devices used in physics The BioPEPR system is capable of the multiple parameter analysis characteristic systems than such
in
measuring of photo(5). A PEPR system
has been It incorporates
system
PEPR
in the
for the scanning and from bubble chambers
adopted research. and
image cantly
since
renamed analysis
features
of cervical
all handling of the slides, as well as the focusing is done off-line, again resulting in a significant
FurtherKoningin
for
15 mm
various
at
chemical
a velocity
of 55
and enzyme
635
Downloaded from jhc.sagepub.com by guest on August 14, 2015
ml/min.
techniques
In
addition
are being
to
this
explored.
method,
Dithio-
ZAHNISER
636 threitol, glycine, hyaluronidase, but none has yielded significantly
and EDTA improved
have been tested disaggregation.
(2, 3),
During the next step of the preparation, a sample of the cells is counted in suspension using a Coulter counter. The counter thresholds are set to count only cells larger than leukocytes. The sample is then centrifuged, the supernatant is removed, and enough carbowax
ET
AL.
Photography: Photonegatives fication of ten using a Ultra Micro large
field
lens.
To
obtain
very narrow bandwidth nm) light to give better
of the cells are made at a magniNikkor 28mm fl.8 high resolution,
optimal
results,
monochromatic
is used. The lens than 1 t resolution
light
is corrected over
of a
for green
a field
of up
(545 to 8 x
8 mm.
of cells than when using the carbowax method. After the cells have been deposited, the slides are air dried, and the cells are fixed in a methanol based fixative (85 parts methanol, 10 parts 40% formaldehyde, and 5 parts acetic acid). The cells are then stained using the Feulgen procedure with thionine (8) as the Schiff reagent. This stain is DNA-specific and has proved to be very reliable. Feulgen hydrolysis is performed for 1 hr using 5 N HC1 at room temperature. The cells are counter-stained using Congo red (0.25% in
A further advantage of the use of a photonegative for scanning is the relatively large depth of field of the lens used (± 4 z). This results in less focusing problems, which is especially important when using automatic focusing. In practice, this depth of field has proven to be large enough to tolerate the variations found within an 8 x 8-mm area of a standard microscope slide. At such a low magnification, it is necessary to use an extremely high resolution film. The Kodak Spectroscopic film type 649-GH has been used with great success. This film is capable of resolving 2000 lines/mm with high contrast objects. Using the present camera system, photonegatives are made on 70 mm wide roll film. The slide is held by a standard microscope table to allow easy positioning of the desired field. The film is held by a vacuum system for optimal flatness. A black and white television system monitors the field; this sytem is used for positioning the field as well as for automatic focussing of the image. Hardware: The basic hardware is modelled after the PEPR CRT system used in scanning bubble chamber photographs. A block dia-
70%
gram
(2%
polyethylene
glycol
in 50%
ethanol)
solution
is added
to ensure
a final distribution on the slide on the order of5O epithelial cells/mm2. An aliquot of the cell suspension is then placed on a glass slide, and the cells are smeared over the surface by hand using the side of a glass pipette. These techniques have consistently resulted in smears with at least 50 to 70% single cells and with much less than 1% nuclear overlap. Methods of centrifuging the cells directly onto the slide are being experimented with, but so far they have not yielded results better than smearing the cells onto the slide by hand. Precoating the slide
with
polylysine
ethanol
with
has
0.25%
also
acetic
been
acid).
tried
The
but
resulted
staining
in a greater
results
loss
in a brown-
This stain gives good results when photographing the preparations and also yields a color suitable for scanning by an analyst. Thionine and Congo red have absorption maxima at 600 and 500 nm, respectively. Measurements are made between these two maxima at a wavelength of 545 nm. ish-blue
nucleus
and
an orangish-red
cytoplasm.
FIG.
1.
Hardware
is shown
in Figure
1. In the
modified
BioPEPR
configuration
a
green phosphor CRT and a 2:1 demagnifying lens is used. Spot size on the film is on the order of 10 to 12 t. This spot size, combined with the xlO magnification onto the film, results in an effective 9-inch
system
random
configuration
resolution
access,
of about
allowing
of BioPEPR.
Downloaded from jhc.sagepub.com by guest on August 14, 2015
1 s. The
instantaneous
device
is capable
positioning
of completely
of the spot
to any
AUTOMATIC point
in the
field.
The
speed
of operation
PRESCREENING
is computer
controlled
OF up
to approximately 5 /zsec (in the cell plane). Two measuring modes are used. The first, the scanning mode, sweeps the spot of light and detects the crossing of a preset darkness threshold as light transmission through the film changes. Using this mode, the widths of objects can be measured extremely rapidly. During
the
sweep
the
is automatically relatively the
low
sweep
the
mode
is placed
second
light
the
by
a circuit
light
level.
into
mode output
of the
film
continually
tracks
the
two
one
buffer
an analog-to-digital conversion, obtained. In practice, the grey on the order of 64 levels. to a PDP 1 1/40 computer with
28K
of
disks
data
storage.
16-bit
This
for
the
on
the
film
photomultiplier
over
a period
terminal, directly
scan
darkness
at
used an
TV
film.
for
The
the
and
output
console
display,
and
latter
operates
on the TV and with
program
interactive
a graphics the
simultaneously of the
are
includes
a TV by
the CRT
of the
and
measuring
photomultiplier.
mode,
is digitized.
RKO5
hardware
a computer
a raster
Additional
transmittance
Two
looking
modulating is
memory. Additional
of
monitor
generating
BioPEPR
637
SMEARS
10 psec, and subsequently performing 256 levels of digital information are level discrimination of the device is The BioPEPR system is attached
consisting
during
memories. while
point
points
of the
recorded
sweep
the
discrete
signal
background
that of
from sweep.
of operation, at
of the
Information
one
of the data from a new
transmittance
integrating
darkness
background
allows the analysis busy collecting data
In the
varying
subtracted
CERVICAL
This
By
output
of
the
photomultipliers
measurement CRT
monitor
is then
of the
CRT
to
correct
used
phosphor. for
slowly
to correct
the
light
output
for any
of the
changes
A signal
is fed
back
varying
light
level
to the
in the
CRT. light
cathode
changes
and
e .
,‘
r
‘
‘I
#{149}‘
6,
,‘
R( ..
,,-
,.
4.Hb:/
.
0
#{149}
.
-,
.
.
“S
‘
V iF:*I..4;e
.
.
.
.
,.
.
4
.
.
-.
A
S
FIG. 2. BioPEPR measurements width ratio (H/W) is ten times the number of nuclei detected
for a) a superficial the actual value; and within the cytoplasmic
epithelial cell and density measurements area.
b) a moderately dysplastic cell. are in nonstandard, BioPEPR
Downloaded from jhc.sagepub.com by guest on August 14, 2015
Area measurements units. Overlap
are in correction
height: indicates
ji2;
of the
638
ZAHNISER
ET
AL.
NA
NA
NA 0
(IC
1 LC
#{149}
EP
UN 0
A2
100%
ON
III.II
0
OR
NA
IC
IC
AT
lIT.
0
0l\
0
Al,
NI)
0
UN ()
NI OF
0
NI
NI
NC
NI
0
0
0
OK 0
NC 0
OK 0
OF
AS
NC
0
0
0/
0
1
OK
0
AS
0
0
AS 0
II
i),’
AS/TOT
0
UN
NC
OK AB/TOT=
UN
1
NO
B
A
NA
UI,
A.,
‘SI
NA
NA
0
Ii:
II
I IN
A.’ ON
IF’ ..)./.
0#{149}/.
NIl
II” C)
NI
0
OK
NI”
0
I
0/
Ilk I
Up\ =
II I (I
NC 1
AS/TOT
I’ I
I III
NI
05 0
III
III’.
I I).,
TI’
1)
AS 0
‘IL
,.,r,, III!
I
Al’ (I
(I
D
C
FIG. 3. Paths taken through the decision tree by a) a normal cell with a small nucleus, b) an artifact detected by the shape criteria, c) a parabasal cell, and d) a carcinoma in situ cell. NA, nuclear area; QC, quality control; LC, leukocytes; EP, epithelial cells; A2, nuclear area (after re-thresholding); UN, unknown; OK, normal; NO, nuclear overlap (shape); NI, nuclear integrated darkness; NC, nuclear: cytoplasmic
ratio;
AB,
abnormal.
output of the monitor photomultipliers is also divided into the signal of the measuring photomultiplier. This latter correction removes any high frequency deviations due to impurities in the CRT phosphor. Software: The software used for the cell analysis is patterned after the basic morphologic features used by cytologists to discriminate between normal and abnormal cells. Programs have been developed to measure nuclear area, nuclear optical density, nuclear texture, and nuclear:cytoplasmic ratio. Additional programs have been designed to recognize artifacts, overlapping nuclei, and leukocytes. The complete measurements of a typical normal and abnormal cell are shown in Figure 2. However, the measurement of all of these parameters is not necessary to identify most types of normal cells. Because of this and to optimize speed, the programs are invoked through a hierarchical tree strategy whereby only the necessary measurements are made and analyzed. Figure 3 shows the tree in its present
passed
form.
At
to the right
each
of
the
if the result
branch
points
of the program
in
the
is above
tree,
control
a given
is
value
and to the used in the Because widths and
left if the result is below that value. tree are explained in the text below. of the high intrinsic speed of BioPEPR its capability of correcting electronically
variations,
nuclear
test,
at
the
top
area of the
(NA) tree
has
been
structure.
chosen The
The
in detecting object for background
as the search
abbreviations
for
highest
priority
nuclei
and
the
measurement of the nuclear area is done using 2-z sweep intervals. Nuclei smaller than a preset size of 56 z2 (approximately equal to the nuclear area of an intermediate cell) are immediately classified as being normal. A certain percentage (3%) of the cells thus classified as normal
pass
through
they are epithelial (LC). This is done preparation
containing
a quality
an
inadequate
uable cells or too many leukocytes further screening. The leukocyte atives due to the overlapping ofan
Downloaded from jhc.sagepub.com by guest on August 14, 2015
control
check
cells (EP) or whether they by measuring the nuclear number will
be sent
(QC)
to
are in fact : cytoplasmic
see
whether
leukocytes ratio.
of diagnostically to a cytoanalyst
A valfor
check helps to minimize false negabnormal nucleus with a leukocyte.
AUTOMATIC Figure
3a shows
the
path
taken
through
the
tree
PRESCREENING by a typical
OF nuclear
normal
nuclei,
3) a normal
identify tree
having a “nucleus” the tree to ascertain
these
or 4) an
categories
abnormal
cell.
is demonstrated
The
below
process
by going
used
larger
immediately Objects they
checked exceeding
are
a 3002 as
of the
whose
nation
darkly
stained, thresholding
level
above
folded,
makes
tance
measuring
Fig. mate
size
reliable. again
the
the At
profile
where
to
set
of
the
measured,
new this
falling
a new
are
time
under
or large
not
into and cells
height:width
the
the the
using for
known,
this
intermediate
method the
a higher
resolution
cell cutoff
area
is
the smaller,
is designated
are
(OK). that
remain
FIG. 4. Frequency cutoff values used
large
after
distributions in the decision
the
re-thresholding
of the tree.
nuclear
are
integrated
sent
to the
darkness
are
occasional
of 1.25
for
(at
measurement
each of the
Using
size,
and
one
point
transmittance
as well
as average
and
times
criterion. designed
The
cells,
cells)
optical
mode
is measured.
dotted
higher
darkness
lines
cutoff
times
and
density
measuring density
half
to
(parabasal
metaplastic
The
one
might through
of the integrated
cells. tree.
two
that path
are
measures
distributions in the
half
and
the
darkness
tree
nuclei
(NI)
abnormal
used
shape
the
large
columnar,
two frequency values
by this
program
integrated and
the
in
with
objects
3b shows
as the
represent
corresponds
normal,
and
the
normal.
final program : cytoplasmic
amplification low contrast
is highly
nuclear
cutoff
cells
first
of normal
to one
scans
approxi-
is fitted
the
cigar-shaped
shown
cervical
nucleus.
integrated
A
nuclear the
two
nuclear
(see
to discard
normal
The
4 shows
lower
transmitprofile
the
an
a parabola
during
programs
intermediate,
a sample
to
algo-
optimal and
the
darkness
point
in the
position
sim-
the
a fixed the
for
in which
obtained
in order
between
a 1- step
Figure
have
data
remaining
cells.
within with
in the
a rethresholding
threshold
using
normal
always
some
abnormal
discrimi-
confusing
points
and
further.
cytoplasm
almost
is determined,
scanning
clusters; are
ratio
the tree The two discriminate
is
(NO)
to the
pass the parabola check. Figure for clusters and artifacts detected
otherwise
nucleus.
tree
uses
A2 cell
nuclei
to improper
which
the
already
the
a single analyzed
inflection
threshold
be
cytoplasms;
Because
nucleus
the
normal
are
Such
of the the
for
artifacts
due
branch-point
and
to
nucleus
check
first
discrimination.
A cell now Nuclei
large
area.
mode,
After
as normal
deeper
or overlapping of
2) indicate
and
passing
nuclear
a darkness
cytoplasmic
usually
artificially
between
background.
rithm
are
cutoff large
(UN)
now
of
size
if it is too
cutoff
appear
(thresholding)
plistic
initial
unknown
“nuclei”
nuclei
measurement
first
the
to see
classified
Many
cells
than
program
cell)
639
Objects not having a round or oval shape are designated and are not analyzed further. The program also checks
unknown
to
through
of the
nucleus.
structure. A “nucleus”
fL.
cell,
with a suspiciously large area is sent if it is 1) an artifact, 2) a cluster of
SMEARS
overlap
side
cell.
An object deeper into
CERVICAL
easily
(NC) locates the cytoplasm and calculates the ratio. Because of the large size of the cytoplasm, made in steps of 2 i. Here the computer boosts the analog of the detection system to make the detection of the cytoplasm
easier.
superficial
epithelial
darker
cytoplasms
detected.
very
high
cells
falling
between
253
normal
cells
Downloaded from jhc.sagepub.com by guest on August 14, 2015
cutoff
A
integrated
and
127
two
lightly may
of the of
darkness, the
The cell 15% and
cytoplasm
diagnostically is used
a cutoff
integrated
abnormal
stained
be incorrectly for
darkness
cells.
Dotted
but
borderline
cells
those
of 24%
of an
measured, cells is used
having for
cutoffs.
The
lines
indicate
a
those former
the
640
ZAHNISER
is set
low
this
because
is most
cytoplasm of
integrated
an
two
and
cells
located
nuclei
to
ratio,
or an
lie within
is then
provide
up
structure.
cells.
Figure
cytoplasmic
Figure 3d shows
that of
S ingle
the
nuclear:
due
a normal
cell.
Using
Normal
how
many
Abnormal
the
the
typically
Total
nuclear:cytoplasmic
1270
TABLE
cells
going
deep
often
followed
taken
by a carcinoma
into
the
Smear
Sampling
strategy: and
false
identify
possible
To
estimate
negative
rates
shortcoming
a lower
of
the
Results
BioPEPR
classification
by parabasal
in the
limit
BioPEPR
programs
Analyst:
in situ
for
the
false
system,
and
Total
or decision
logic,
Normal
Suspect
rate
according
from
photonegatives
normal and two abnormal In the test measurement that BioPEPR measured types.
The
normal
parabasal, (78)
(1276)
and
mildly
dysplastic,
and
carcinoma
in situ
of nuclear
area,
has
been
is passed
left
maximum tively
shape, recorded
diagnosis
or
made
two node,
by to
cells
severely
the
dys-
measured
density,
disk
of the
a preset
and
node
cutoff
along
the
chosen This
relevant
cells
To
tree
obtain
have
classified
normal.
19) the
The
missed. ±
this
procedure
proved
efficiency
to be necessary
these
affect
program smears
most of
specimens cells,
efficiency
has begun. are presented.
32 abnormal the
smears outpatient
from patients of the abnormal
were patients
were
used
false
negative
rates,
an
iterations
of
optimal
this
cell level
rates
contain
an
overlapping
impression
full-smear
of how
measurement
the results of the first set of 192 from 160 normal smears and The
normal
according
BioPEPR.
smears to the
The
were
The
a carcinoma
remaining
smears
in
cancer.
a first
estimate
on
smear
the
or invasive
situ
The
abnormal
results
were
of the level
present
under
smears
were
cancer
from realistic
and
these
192
false
positive
shown
matrix
in Table
as normal superficial cervical
I. Of the
and cell,
for the
one
smears and
1276
normal
6 were incorrectly 3 intermediate
columnar
cell).
cells,
1270
were
results
in a false
cells, classified
cells
(3 out
of
negative
rate
for
this
sample
is
minus
and false an average
for each of 160 normal II). A fraction representing false
(truly
negatives)
optimal smears. classified were
characterized
parakeritosis,
and
the
abnormal
two one
by the
senting metrium.
plus
total
number
leukocytes and arA cutoff value of 1%
metaplasia.
The
smears dysplasia
folded heavy
remainor rolled cytolysis.
were incorrectly and the other
classirepre-
a highly differentiated adenocarcinoma of the The former was characterized by an extremely ratio,
few abnormal The analysis upon the number structure.
the
latter
was
endolow
characterized
by
cells to pass the 1% cutoff. time for each smear is of course dependent of cells which must go deeper into the tree
Typically,
probably
and
of
false
between the normal and smears, 18% were thus inOf these 28 false positive by slight abnormalities: trichi-
squamous
of the 32 abnormal representing a mild
32 abnormal number
cells,
divided
discrimination Of the normal as suspect.
and
negative of 6000
20
to
containing be necessary
photonegatives)
per
40
sec
up to scan
woman,
are
to 4000
needed
cells.
about thus
to
15,000
taking
analyze
one
In practice, cells
it will
(four
to six
1 to 2 mm.
DISCUSSION cells
is
classified
classified as abnormal cells, one parabasal cell,
This
dysplastic
ing 11 contained large clusters of leukocytes, cytoplasms, darkly stained cytoplasms, or Only fled;
calculated
incorrectly
level false positive “dirty” samples,
conditions.
on single
being
78 abnormal
cells (3 out of 36). None of carcinoma in situ cells were
of diagnostically valuable cells (excluding tifacts) was calculated for each smear.
The
measurements
32
error
6 were
3 mildly
objects
photonegative
RESULTS A confusion
smear realistic
too
above-described
abnormal
and
false
nuclear:cytoplasmic
collected
earlier diagnosed as having abnormalities. smears were dysplastic, and of those one-
endometrial
to make
they get
extensive
prepared by
dysplastic. with
To
analyzed.
measured
mildly with
artifacts.
clinic,
obtained One-half
patient
few
at the
are not clean;
and
were
and
from
to arrive
In this report, Photonegatives
method,
half
a
rates.
In practice, nuclei, cluster
from
Only
160
Of the
dysplastic or the
overall
cells were analyzed smears (see Table
monas,
obtained.
the
0. 19%,
6 included
estimate for more
false
were
28 30
3.0%.
gave the abnormal correctly smears, 17
rates
±
distribution.
as abnormal
These
point in the tree and the other representing the abnormal cells. By comparing the two distributions, and in particular their area of overlap, an “optimal” cutoff value was chosen at each node. This was continued throughout the tree and fmal cell level false positive and negative
of 0.47
and 3 moderately severely dysplastic
positives, at
reached
sample
to a binomial
abnormal-classified
interacprogram
parameter
that
as
To rates
nuclear:
in the
value.
cutoff value was graphics package. normal
for the
7.7
values
computer
at each
of
the
abnormal
a cytoanalyst.
distributions
representing
cell
control
according
frequency one
the
previously,
right
each optical
onto
four
intermediate,
The
dysplastic,
nuclear
system efficiency, each the help of a simple
with
displays
moderately
were explained
superficial,
cells.
For
from
care was taken to ensure of a wide variety of cell
included
cells.
fields
72 were
metaplastic
nuclear
ratio
the cell type
As
each
cells
included
cytoplasmic
with
smears. of these cells, enough of each
endocervical,
plastic,
containing
2
a set
cells
obtained
132
Abnormal
to
of measurements on single cells has begun. To date, 1354 preclassified normal and abnormal cells have been measured by BioPEPR. The were
78
II
Classification
Normal
positive
1276
72
ratio
cell.
one
6 6
present.
path
Suspect
Normal
measures
for
Results classification
Analyst:
to clusters
touching
paths
I
Classification BioPEPR
program The
Cell
overlapping
for
negatives
TABLE
the
the
3c shows path
estimate
are
typical
of the
AL.
cell
nuclei
3d show
indicated feature
detection
on false
area.
if additional
3c and
tree
down
already
additional
a better
cut
as a mask,
the
scaled
Figures
the
abnormal
cytoplasm
An
allows
thus
has
cell.
routine
cytoplasms
of abnormal
darkness
abnormal
measuring
cytoplasmic
the
the
likely
ET
(1 and
positive
results
representative smears; they and sion smears,
of the
false negative is that if the
measurements
ofthe results thus provide
desired
Downloaded from jhc.sagepub.com by guest on August 14, 2015
rates BioPEPR smear
done
on
that can be obtained a lower limit for the
on the cell level. is presented level
false
single from false
cells
are
“clean” positive
The primary concluwith well-prepared
positive
(10%)
and
false
AUTOMATIC
negative
(1%)
rates
apparatus and In principle, preted
level
types
assortment
of
cells,
parabasal
positive
rate
of 0.47%
with false
a much positive
rate, cells
it is important to note that and 8.3% of the moderately none
severely
us
begin
presence
From the
to of
data,
the
rate
the
of about
smear system. desired. that
derive
suggest
preparation and analysis This is approximately The experience from false
of clusters and overlaps should false positive rate. In addition, of
give
an improvement
weighting
certainty
the factor
analysis is being
metrial more sional
same
as our
test
of
sample,
For
a
a smear
cells, false
the
the
abnormality percentage
a higher negative
or carcinoma
in
screening
level, there
a slightly or moder-
false negative are missed,
rate of 8.3 to and therefore,
due
nuclei,
clusters
some
preliminary
that
a smear
level
with
present
the
false
rate
can
on
objects quality
it should
We would
like to thank
in the
efficiency of each
To date no in-depth study has been of the BioPEPR system in detecting endometrium. The system has been
rates; suspect
development
of a
measuring feasibility.
occain a
system apThe system
to screen
2 or 3 smears
per
Riet
Harbers-Hendriks
of smear
ofsmears;
the
for her expert
preparation
staffof
the
techniques
Laboratory
their
especially neering
and
of
work
on
hardware
support,
and superb
in developing
Husain for
of the science construction;
and
due group
the BioPEPR
GJ, Todd gynecologic
P, Sattilaro specimens:
25:513,
automated
system.
RF: The Chemical
systems for Cytology.
dispersal agents.
of cells from J Histochem
JA: A sample
preparation
1977
Page-Roberts
OAN,
the engifor their
CITED
PH, Wied GL: High resolution processing cancer. The Automation of Uterine Cancer of Cytology, Chicago, 1976. p 144
Tutorials 2. Escobar human
the
of Nijmegen cell samples.
Special thanks are High Energy Physics
work
in
for Cytopathology
development
the camera system. technical staffof the
work
and
cervical
BA, Millet cancer
screening.
Acta
Cytol
22:15,
1978
4. Mullaney PF, Steinkamp JA, Crissman HA, Crowell JM, Martin JC, Salzman GC, Cram 5: An introduction to flow systems for cell analysis and sorting. The Automation of Uterine Cancer as
Cytology,
5. Pless
is
Tutorials
IA: PEPR.
High
Energy
6. Poulsen
the
7.
a 8.
made of the efficiency abnormalities of the optimized for use in a
of Cytology,
1976.
p 90
of the International Conference on Dubna, 1964, p II 409 LH, Cahn RL, Louis C, Toussaint G: High
Oliver
analysis
of cervical
cells-A
progress
report.
J Histo-
chem Cytochem 25:689, 1977 Tanaka N, Ikeda H, Ueno T, Watanabe S, Imasato R: Fundamental study on automatic cyto-screening
Y, Kashida for uterine
cancer. utilizing
(CYBEST) 21:85, 1977
Van
III. the Duijn
New pattern
system of recognition
P: A histochemical method positive substances.
its use in a bi-color acid 9.
Chicago,
Proceedings
Physics,
RS,
resolution
by
be possible
development
for
a method cell
the
faculty
3.
also lead to a reduction in the a modification ofthe 1% criteria
results studied.
the
The
to endo-
detection of the is represented
and the Departments of Gynecology of the Universities and Utrecht for their help in obtaining and classifying We would also like to thank the Technical Services
of
a reliable method in suspension onto in the recognition
scrape.
method
entirely limited by the speed of the computer. computer power and/or more efficiently writ-
Cytochem
positive
be expected
warrant
LITERATURE
techniques of the BioPEPR a factor of two higher than these measurements indicates positive
not
a cervical
sampling
speed of the for commercial
ten programs, min.
1. Bartels cervical
etc.
limits
may
scrape. The that desired is now increased
ideas,
to the
of cells,
scrape
analysis that guarantees where endometrial cancer
classification
is not a large relatively few
arise
using
cervical
abnormalities complex cases
speed With
with
this are
that
program
of the
cervical proaches
641
SMEARS
ACKNOWLEDGMENTS
cells
situ
and false negative rates. criteria of 1% suspicious
be achieved
in the
inter-
making
on the
achieved.
disaggregation methods improve and found for randomly depositing the cells glass slide. Furthr software improvements
should
types,
complications
can
18% can
scale
smear
entire abnormal smear is unlikely. the 192 “complete” smears allows
positive decision
results
a reduction
level cells
overlapping
one
CERVICAL
large
15.8% of the mildly dysplastic dysplastic cells were missed,
smear because
A cell a few
examine
artifacts,
these
the
of an from
smear level false Using the sample
in a smear,
etc.
be Each
that BioPEPR will have containing only mildly
cells. On however,
misclassification The data collected to
cell
be
dysplastic
abnormal cells present. 15.8% means that only the
cells, would
missed. We can infer efficiency for smears
ately dysplastic number of cells,
scanning
should
sample.
higher percentage of parabasal level may be expected. As for
of the
existing
rates
in the
of a test statistic for deciding difficult. For a smear with the
intermediate
were lower
the
efficiency
cell
false
but
using
OF
insensitivity
of the
different
a
development of a smear
be obtained
software. the cell
in terms
contains
can
PRESCREENING
schiff
Wheeless LL, Hardy system for automated
Downloaded from jhc.sagepub.com by guest on August 14, 2015
automated method.
apparatus Acta Cytol
specific thionine-SO2 for deoxyribonucleic acid J Histochem Cytochem
JA, Balasubramanian cytopathology. Acts
reagent and and periodic 4:55, 1956 Slit scan flow
N: Cytol 19:45,
1975
