Hematopathology / DIGITAL IMAGE ANALYSIS IN HEMATOLOGY
Performance Evaluation of the CellaVision DM96 System
WBC Differentials by Automated Digital Image Analysis Supported
by an Artificial Neural Network
Alexander Kratz, MD, PhD, MPH,1,2 Hans-Inge Bengtsson, MSc,3 Jeanne E. Casey, H(ASCP),1*
Joan M. Keefe, MT(ASCP),1 Gail H. Beatrice, MLT(ASCP),1 Debera Y. Grzybek, H(ASCP),1
Kent B. Lewandrowski, MD,1,2 and Elizabeth M. Van Cott, MD1,2
Key Words: Image analysis; Differentials; Hematology analyzers; Laboratory automation
DOI: 10.1309/XMB9K0J41LHLATAY
Abstract
We evaluated the CellaVision DM96 (CellaVision
AB, Lund, Sweden), an automated digital cell
morphology and informatics system for peripheral
blood smears. Technologists agreed with 82% of the
instrument’s preclassifications. Correlation coefficients
between final results released from the CellaVision and
results obtained by direct microscopy were 0.96 (all
neutrophils), 0.94 (lymphocytes), 0.88 (segmented
neutrophils), 0.73 (eosinophils), 0.69 (bands), and 0.67
(monocytes). After correction for statistically and
clinically insignificant variations, the CellaVision
DM96 had 95% sensitivity and 88% specificity for
immature myeloid cells. It was 100% sensitive and 94%
specific for blasts, and 100% sensitive and 97% specific
for unusual WBCs and nucleated RBCs. Advantages of
the CellaVision DM96 over direct microscopy include
the ability to review slides from a remote location,
consultation and quality control on a cell-by-cell basis,
and potential labor savings.
770 Am J Clin Pathol 2005;124:770-781
770 DOI: 10.1309/XMB9K0J41LHLATAY
Downloaded from https://academic.oup.com/ajcp/article/124/5/770/1759765 by guest on 06 February 2024
The differential counting of leukocytes, introduced almost
115 years ago by Ehrlich,1 has remained a clinically important
and frequently ordered laboratory test.2 Progress in the design
of laboratory instrumentation has replaced the microscope with
automated cell counters as the instrument of choice for the
majority of differential counts (differentials) in most laborato-
ries.3 Modern automated cell counters can provide a reliable
WBC differential count for samples that are normal or that
exhibit only a quantitative abnormality. Qualitatively abnormal
samples, such as those with immature or abnormal cells, still
require the preparation of a slide and microscopic analysis. In
our laboratory, approximately 28% of differentials are
“flagged” by the automated cell counter and require the prepa-
ration of a blood smear for microscopic review.
Difficult differentials with unusual or abnormal cells may
be a relatively rare event in many clinical laboratories; never-
theless, the significant clinical impact of a wrong diagnosis
necessitates the around-the-clock on-site presence of highly
trained personnel for the microscopic review of these blood
smears. This increases labor costs, a major issue in today’s
cost-sensitive health care environment. Other disadvantages of
direct microscopic examination of a stained blood smear
include interobserver and intraobserver variation and the fact
that the individual cells used for the differential count are not
easily retrievable for subsequent review. These issues lead to
difficulties in quality control and maintaining interobserver
and intraobserver consistency in interpretations.
Automated image processing systems have been devel-
oped to address these difficulties. These systems generally
obtain digital images of objects on the blood smear and then
use sophisticated software to identify and preclassify cells; the
images are stored for possible later retrieval. A number of such
© American Society for Clinical Pathology

devices have been developed (recently reviewed by Tatsumi
and Pierre2). Capitalizing on advances in digital photography
and data processing and storage, these devices have made it
possible to envision a “virtual blood film” that can be
reviewed, shared with experts in distant locations, stored,
retrieved, and rereviewed at low cost.4,5
The latest such automated image analysis system to
become commercially available is the CellaVision DM96
(CellaVision AB, Lund, Sweden; distributed in the United
States by Sysmex America, Mundelein, IL). The successor
model to the DiffMaster Octavia (CellaVision AB),6 the
CellaVision DM96 is an automated image analysis system for
peripheral blood smears. Barcode-labeled, stained glass slides
are placed into a magazine. The instrument can be loaded with
up to 8 magazines, each containing up to 12 slides, and oper-
ates with a continuous feed that allows magazines to be added
constantly. By scanning the glass slides at low power, the
instrument identifies potential WBCs and then takes digital
images at high magnification. The images are analyzed by an
artificial neural network based on a large database of cells and
preclassified according to WBC class. The WBCs and their
suggested classification are presented to the user on a cus-
tomizable computer display for confirmation or reclassifica-
tion ❚Figure 1❚. The system also provides functionality for
the review of RBC and platelet morphologic features and for
❚Figure 1❚ Computer screen on which WBCs preclassified by
the CellaVision DM96 are presented to a reviewer for
approval or reclassification. The reviewer can enlarge single
cells as needed and can leave individual cells in the
categories suggested by the instrument (accept the
preclassification) or move individual cells into other groups
(reclassify the cells). Digital images and the actions taken by
the reviewer are stored by the instrument and can be
reviewed later for quality control, confirmation of diagnosis,
consultation, or comparison with later findings.
© American Society for Clinical Pathology
Hematopathology / ORIGINAL ARTICLE
estimation of the platelet count. Advantages of this approach
over slide review with a microscope include possible labor
savings owing to the localization and preclassification of the
WBCs by the instrument, more reproducible results, and the
ability to review cases from a remote location or to rereview
cells at a later date.
Direct microscopic review of blood smears is a well-estab-
lished procedure; clinicians and laboratory workers have exten-
sive experience with this method, are comfortable with it, and
have confidence in the results obtained. Introduction of an instru-
ment for automated image analysis, such as the CellaVision
DM96, into the clinical laboratory will necessitate a change in
Downloaded from https://academic.oup.com/ajcp/article/124/5/770/1759765 by guest on 06 February 2024
work habits, retraining of personnel, and increased spending on
hardware and software. For automated image analysis to replace
direct microscopic review, 2 conditions will have to be fulfilled:
the automated method will have to be shown to be at least as reli-
able clinically as the direct microscopic review of slides, and
there will have to be significant labor savings associated with the
new method to justify the additional expenditure associated with
the purchase of the equipment.
In this study, we compare the clinical performance of the
CellaVision DM96 with direct manual microscopy. We also
show the results of workflow timing studies comparing direct
microscopy with the CellaVision DM96.
Materials and Methods
Study Site and Personnel
The study was performed in the Clinical Hematology
Laboratory, Massachusetts General Hospital, Boston, a terti-
ary care academic medical center serving a large inpatient and
outpatient population, including an active hematology-oncol-
ogy service. All studies were performed by hospital laborato-
ry employees who had been trained in the use of the instru-
ment on-site by a CellaVision company representative. A reli-
ability log for the instrument was kept for the duration of the
study.
Sample Selection and Preparation
Routine patient samples on which WBC differentials had
been ordered by the clinician and on which a microscopic
slide review was necessary were used in the study. In our lab-
oratory, the determination that a microscopic slide review is
necessary is made when a sample fails criteria for the direct
release of results from an ADVIA 120 Hematology System
(Bayer HealthCare, Diagnostics Division, Tarrytown, NY).
These criteria have been set to conform with the manufactur-
er’s recommendations and the clinical needs of the patient
population seen at our institution. All samples that were
included in the study had been screened by the ADVIA 120
Am J Clin Pathol 2005;124:770-781 771
771 DOI: 10.1309/XMB9K0J41LHLATAY 771
Kratz et al / DIGITAL IMAGE ANALYSIS IN HEMATOLOGY
Hematology System and flagged for microscopic review. The
use of cases prescreened by an automated cell counter was
intended to mimic the standard workflow in a hematology lab-
oratory, where usually only cases flagged by the automated
cell counter are subject to slide review. Wedge blood smears
were prepared manually with the push-pull method with a
spreader slide or with a semiautomated method, the Miniprep
blood smearing instrument (Geometric Data, Wayne, PA), air
dried, and stained with a Sakura Stainer RSG-61 (Sakura
Fineteck USA, Torrance, CA) with a Wright-Giemsa stain.7
Manual Differential Counts (Reference Method)
Manual differentials using standard microscopic tech-
nique were performed following the laboratory’s standard
operating procedure, which is based on National Committee
for Clinical Laboratory Standards guidelines.8
Analysis With the CellaVision DM96
After analysis by direct microscopy, slides were labeled
with a barcode and loaded into the CellaVision DM96. This
instrument scans slides, identifies potential WBCs, takes digital
images of them, and uses artificial neural network–based soft-
ware to analyze the cells. Digital images of preclassified cells
are presented to the technologist on a computer display (Figure
1). The technologist is asked to review the images and to accept
the preclassification provided by the software or to reclassify
cells. For this study, technologists using the CellaVision DM96
were provided with the same information about the cases as was
available to them during direct microscopic analysis. This
included the results from the automated cell counter on the pres-
ent sample and previous CBC counts, WBC differentials, and
clinical results for the patient, if available. Technologists also
were asked to evaluate whether the areas presented by the
CellaVision DM96 for the evaluation of RBC and platelet mor-
phologic features and platelet count were adequate.
Studies Performed
Three sets of studies were performed: a clinical perfor-
mance evaluation, a manual differential timing study, and a
standard workflow timing study. The clinical performance
evaluation was intended to determine the performance of the
CellaVision DM96 in difficult cases; for this study, cases that
met certain criteria (eg, low WBC count, unusual WBCs) were
obtained during a 2-month period and used for this study. For
the manual differential timing study, 10 cases of average diffi-
culty were selected. Last, for the standard workflow timing
studies, a random sample of cases for which slides were pre-
pared in the laboratory during the study period was used.
Clinical Performance Evaluation
We selected 120 slides with abnormal blood smear find-
ings thought to be difficult cases for evaluation of the clinical
772 Am J Clin Pathol 2005;124:770-781
772 DOI: 10.1309/XMB9K0J41LHLATAY
performance of the CellaVision DM96. These cases included
slides with increased numbers of immature WBCs and the
presence of abnormal WBCs and some cases with very low
WBC counts. All slides had required a full microscopic differ-
ential count.
Three laboratory employees (2 certified laboratory tech-
nologists [J.E.C. and J.M.K.] and 1 certified medical laboratory
technician [G.H.B.]) were asked to participate in this part of the
study; they were selected for representing the range of expertise
in the performance of manual differentials found in the labora-
tory. Expertise of employees was determined by their level of
formal training (certified technologists vs certified technicians),
Downloaded from https://academic.oup.com/ajcp/article/124/5/770/1759765 by guest on 06 February 2024
the length of their work experience (between 1.5 and 35 years),
and whether the employee was training new employees in the
performance of WBC differentials and was a designated
resource person in the laboratory for difficult WBC differential
cases. Each employee was asked to perform manual differen-
tials on 40 slides with the microscope and approximately 1
week later on the same 40 slides with the CellaVision DM96.
Manual Differential Timing Studies
Ten slides that had required full microscopic differential
count were used for a manual differential timing study. Five
technologists and a technician, representing the range of
expertise in the performance of manual differentials found in
the laboratory, were timed while performing manual differen-
tials on these slides with the microscope and then approxi-
mately 1 week later while performing the same task with the
CellaVision DM96. Cases were considered completed on
finalization of the results in the laboratory information system
or on conclusion of analysis with the CellaVision DM96.
Standard Workflow Timing Studies
We used 236 additional randomly selected cases on which
a manual slide review was necessary for a workflow study.
Approximately half of these slides required the performance of
a full microscopic differential count; the other half required
only microscopic review (“scan”) for confirmation of the results
of the differential obtained from the automated cell counter.
These cases first were analyzed by a medical laboratory techni-
cian (G.H.B.) with 1.5 years of experience in reading blood
smears (n = 160 cases) or a very experienced technologist
(J.E.C.) with 35 years of experience (n = 76 cases) with the
direct microscopic method as part of their normal work assign-
ments in the clinical laboratory. The employees then were asked
to reanalyze the same samples with the CellaVision DM96.
These employees also had participated in the other studies.
Statistical Analysis
Statistical analysis was performed using Microsoft Excel
software (Microsoft, Redmond, WA). Two-tailed paired t tests
were used to evaluate differences between the time needed for
© American Society for Clinical Pathology

fully manual differentials with the CellaVision DM96 and
with direct microscopy. Clinical sensitivity and specificity of
the CellaVision DM96 were defined as its ability to obtain
positive and negative results concordant with the results
obtained by direct microscopy.
Results
System Reliability
The CellaVision DM96 was in use in our laboratory for 2
weeks. During this time, a “critical oil error” required 5 min-
utes for resolution by priming the immersion oil pump to
remove air bubbles. There were 4 rack jams; each of these
incidents required approximately 1 or 2 minutes to resolve; 2
of the jams were caused by trying to load slides not appropri-
ate for the instrument. There were 3 cases of software failure
that required the viewing program to be restarted, leading to
the loss of approximately 1 minute per incident. In total,
approximately 12 to 20 minutes were lost in these 8 downtime
incidents during 2 weeks.
Ability of the CellaVision DM96 to Present to the
Reviewer WBCs That Are Qualitatively and
Quantitatively Adequate for a Differential Count
We used 120 cases that required a full microscopic differ-
ential to evaluate the clinical performance of the CellaVision
DM96. A laboratory employee first performed a differential on
these cases using the microscope; the same person then per-
formed a differential on the same slides using the CellaVision
DM96. The slide readers were asked to note when they thought
that the images presented by the CellaVision DM96 were of
insufficient quality to allow them to perform a differential.
Such a notation was made in 11 (9.2%) of the cases. For 10 of
these 11 cases, the same technologists had been able to do a
full 100-cell differential with the microscope; in the remaining
case, the microscopic differential also was inadequate (it was
based on 5 WBCs). Of the 11 slides, 10 had been prepared with
the manual push-pull method with a spreader slide; 1 had been
prepared with a semiautomated method, the Miniprep blood
smearing instrument ❚Table 1❚. These 11 cases in which no dif-
ferential was available from the CellaVision DM96 were
excluded from subsequent correlation analysis.
The precision of a WBC differential is affected negatively
if only a few cells are counted.9,10 We compared the total num-
ber of cells found by the technologists when performing WBC
differentials using the microscope with the number of adequate
WBCs presented for review for each case by the CellaVision
DM96 ❚Figure 2❚. The ability of both methods to find identifi-
able cells on a blood smear was similar, with a slightly higher
number of cases with very few (< 20) WBCs encountered with
© American Society for Clinical Pathology
Hematopathology / ORIGINAL ARTICLE
❚Table 1❚
Adequacy of Images From the CellaVision DM96 for WBC
and RBC Analysis According to Slide Preparation Method*
Manual Method Semiautomated
Images Inadequate for (n = 90) Method (n = 30)
WBC analysis 10 (11) 1 (3)
RBC analysis 18 (20) 10 (33)
* The manual method refers to slides prepared with the push-pull method with a
spreader slide; the semiautomated method refers to slides prepared with the
Miniprep slide preparation instrument. A total of 120 slides were analyzed.
Numbers represent the total number of slides. Data are given as number
(percentage).
Downloaded from https://academic.oup.com/ajcp/article/124/5/770/1759765 by guest on 06 February 2024
the microscopic method (Figure 2). There were 11 cases
(9.2%) with fewer than 20 cells counted when using the man-
ual method; the CellaVision DM96 differential was based on
fewer than 20 cells in 7 cases (5.8%). Among the 11 cases in
which the manual differential was based on fewer than 20 cells,
in 5 cases the CellaVision DM96 was able to find more than 20
cells per slide. There were 2 cases in which the CellaVision
DM96 differential had to be based on fewer than 20 cells per
slide, whereas the manual differential was based on more than
20 cells per slide. Because of the limited reproducibility of dif-
ferential results based on such low counts, the 13 cases in
which the manual differential, the CellaVision DM96 differen-
tial, or both were based on fewer than 20 cells per slide were
excluded from subsequent analysis.
Ability of the CellaVision DM96 to Provide Images
Adequate for Evaluation of RBC and Platelet
Morphologic Features and the Platelet Count
The CellaVision DM96 allows the user to evaluate RBC
and platelet morphologic features and to estimate the platelet
70
60
50
No. of Samples
40
30
20
10
0
<10 11-20 21-30 31-40 41-50 51-60 61-70 71-80 81-90 91-100 >100
Counted Cells
❚Figure 2❚ Total number of cells used by technologists for WBC
differential counts when using the microscope (white bars) and
the CellaVision DM96 (black bars). Only objects classified as
WBCs by the technologist were included in the analysis.
Am J Clin Pathol 2005;124:770-781 773
773 DOI: 10.1309/XMB9K0J41LHLATAY 773
Kratz et al / DIGITAL IMAGE ANALYSIS IN HEMATOLOGY
count by producing an overview image consisting of 35
patched areas of the slide. We asked technologists to indicate
whether they thought that these images were adequate for the
evaluation of RBCs and platelets. In 28 cases (23%), the tech-
nologists thought the images did not allow them to adequate-
ly evaluate RBC and platelet morphologic features. The use of
a semiautomated slide preparation method was associated
with a higher percentage of inadequate slides (33%) than the
manual push-pull method (20%) (Table 1). When reviewing
the same slides with the microscope, the technologists thought
in 100% of cases that they were able to find areas allowing
them to adequately address RBC and platelet morphologic
features and to estimate the platelet count.
Correlation of the CellaVision DM96 Preliminary
Classifications With Final Classifications by Medical
Technologists and Technicians
The CellaVision DM96 presents, on a computer display,
WBCs preclassified according to cell type. The technologist
has to review every cell and can move a cell to a different cell
class (reclassify the cell) or can leave the cell in the category
suggested by the instrument by taking no additional action
regarding the cell. ❚Table 2❚ shows the percentages of the var-
ious cell types preclassified correctly by the CellaVision
DM96. Overall, the instrument preclassified 82% of all cells
correctly (86% if band forms were not differentiated from seg-
mented neutrophils). The percentages of correctly preclassi-
fied cells were highest for mature cells and lowest for imma-
ture and abnormal cells.
Correlation of Direct Microscopic WBC Differential
Results With Final Results From the CellaVision DM96
After review and, if necessary, reclassification of the cells
presented by the CellaVision DM96, a reviewer releases the
final differential results to the laboratory information system.
Differential results obtained by direct microscopy were com-
pared with the final results released by the same technologist
for the same slide from the CellaVision DM96. The correla-
tion graphs for neutrophils (including segmented neutrophils
and band forms), lymphocytes, monocytes, and eosinophils
are shown in ❚Figure 3❚. The correlation coefficients were
highest for total neutrophils (0.96), lymphocytes (0.94), and
segmented neutrophils (0.88). For eosinophils, the correlation
coefficient was 0.73. The lowest correlations were observed
for band forms (0.69) and monocytes (0.67).
Ability of Digital Image Analysis by CellaVision DM96 to
Identify Clinically Important Abnormalities
In clinical practice, the ability to identify the presence of
an abnormality on a blood smear, for example, the presence of
blasts, is an important determinant of the reliability of a differ-
ential counting method. We compared the ability of the
774 Am J Clin Pathol 2005;124:770-781
774 DOI: 10.1309/XMB9K0J41LHLATAY
❚Table 2❚
Percentages of Cells Correctly Preclassified by the CellaVision
DM96*
Correct All Verified Results
Suggestions by Suggested Correctly
Cell Class CellaVision (%) by CellaVision (%)
Segmented neutrophils 92.5 82.8
(n = 3,510)
Band neutrophils (n = 868) 57.1 54.2
Lymphocytes (n = 2,585) 96.4 95.3
Monocytes (n = 763) 81.4 74.8
Eosinophils (n = 231) 63.2 93.6
Basophils (n = 50) 80.0 85.1
Blasts (n = 395) 65.1 84.8
Downloaded from https://academic.oup.com/ajcp/article/124/5/770/1759765 by guest on 06 February 2024
Immature myeloid cells (n = 627) 53.2 63.8
Nucleated RBCs (n = 165) 86.7 79.9
* The percentage of correct suggestions is the percentage of the cells in the various
groups, as preclassified by the CellaVision DM96, in which the technologist agreed
with the instrument’s preclassification (ie, the percentages of original suggestions
that were correct); the percentage of all verified results suggested correctly is the
percentage of cells in each group released by the technologist that were
preclassified correctly by the CellaVision (ie, percentages of final results
preclassified correctly by the CellaVision). A total of 9,194 cells were analyzed.
CellaVision DM96 to identify the presence of several clinical-
ly important abnormalities with the results of direct
microscopy. These abnormalities included the presence of
immature and abnormal WBCs and of nucleated RBCs. As
shown in ❚Table 3❚, a significant number of cases showed an
abnormality on review with the CellaVision DM96 that were
not reported on direct microscopy; some cases did not show an
abnormality on review with the CellaVision DM96 that were
reported on direct microscopic review. When direct
microscopy was used as the reference method (“gold stan-
dard”) against which the results of the CellaVision DM96
were compared and when all discrepant results were included
in the analysis, the specificity of the CellaVision was 82% to
93%; sensitivity was 25% to 91% ❚Table 4❚.
To resolve the significant percentage of false-positive and
false-negative findings of the CellaVision DM96 compared
with direct microscopy, digital images from the CellaVision
DM96 and microscopic differentials were rereviewed by the
medical director of the laboratory (A.K.) for all cases with dis-
crepant results. As shown in ❚Table 5❚, ❚Table 6❚, and ❚Table 7❚,
most of the discrepant cases were due to small variations with-
in the 95% confidence interval of the cell count (eg, most
cases with a discrepancy involving nucleated RBCs showed 1
nucleated RBC/100 WBCs with one method and none with
the other method).
Some other discrepancies were due to a slight difference in
nomenclature without clinical relevance (eg, the same technolo-
gist would call cells “blasts” with one method and use the
expression “other cells; immature myeloid cells, query blasts”
with the other method). In 15 of the discrepant cases, rereview of
the digital images showed that the discrepancy was due to
© American Society for Clinical Pathology
 Hematopathology / ORIGINAL ARTICLE

A B

100% 100%
90% 90%
80% 80%
70% 70%
60% 60%
50% 50%
40% 40%
30% 30%
20% 20%
10% 10%
0% 0%

Downloaded from https://academic.oup.com/ajcp/article/124/5/770/1759765 by guest on 06 February 2024

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

C D

70% 100%
90%
60%
80%
50% 70%
40% 60%
50%
30% 40%
20% 30%
20%
10% 10%
0% 0%
0% 10% 20% 30% 40% 50% 60% 70% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

E F

70% 25%
60%
20%
50%

40% 15%

30% 10%
20%
5%
10%
0% 0%
0% 10% 20% 30% 40% 50% 60% 70% 0% 5% 10% 15% 20% 25%

❚Figure 3❚ Correlation of neutrophil (A), segmented neutrophil (B), band form (C), lymphocyte (D), monocyte (E), and eosinophil
(F) counts on direct microscopy and on the CellaVision DM96. A, y = 0.9451x + 0.0091; R2 = 0.9563. B, y = 0.8658x + 0.0067;
R2 = 0.8771. C, y = 1.188x + 0.0247; R2 = 0.6852. D, y = 0.9597x + 0.013; R2 = 0.9393. E, y = 0.7701x + 0.0257; R2 = 0.6658.
F, y = 0.7999x + 0.002; R2 = 0.73.

misidentification of cells by the initial reviewer. There were only
3 cases in which the discrepancy could not be resolved by statis-
tical variation, differences in nomenclature, or misclassification
of digital images presented by the CellaVision DM96; in these
cases, there were clearly cells consistent with immature myeloid
cells (promyelocytes, metamyelocytes, and/or myelocytes) pres-
ent on the CellaVision DM96 images that had not been reported
on direct microscopic review. A repeated standard microscopic
differential also did not show these immature myeloid cells on
the slides. When a third direct microscopic differential based on
up to 300 cells was performed, populations of immature myeloid
cells (between 2.3% and 8%) were identified.
As noted in the preceding paragraph, most of the discrepan-
cies were insignificant statistically (within the 95% confidence

© American Society for Clinical Pathology Am J Clin Pathol 2005;124:770-781 775

775 DOI: 10.1309/XMB9K0J41LHLATAY 775
Kratz et al / DIGITAL IMAGE ANALYSIS IN HEMATOLOGY

interval of the cell count) or clinically (slightly different ❚Table 3❚

nomenclature without clinical relevance). We therefore reana- Identification of the Presence of Abnormal Findings on a
Blood Smear*
lyzed the data and omitted these cases. The results of the
reanalysis are given in ❚Table 8❚. Sensitivities for the various CellaVision DM96
abnormalities now were between 95% and 100%; specificities Manual Microscopy Positive Negative
varied between 88% and 97%.
>3% Promyelocytes, myelocytes, and/or metamyelocytes
Positive 21 2
Timing Studies on Fully Manual Differentials Negative 12 61
An important issue in the evaluation of a new technology Blasts
Positive 12 2
is its cost-effectiveness. For an instrument like the CellaVision Negative 7 75
DM96, a major determinant of cost-effectiveness is the Other unusual WBCs, eg, promonocytes, plasma cells,
prolymphocytes
amount of time it takes for a technologist to perform a manu-

Downloaded from https://academic.oup.com/ajcp/article/124/5/770/1759765 by guest on 06 February 2024

al differential. To compare the time needed to perform fully
manual differentials on the CellaVision DM96 with direct
microscopy, 10 slides that required a full manual differential
(as opposed to a review of the slide and release of the automat-
ed differential from the cell counter) were analyzed by 6 tech-
nologists with a microscope and with the CellaVision DM96.
It took the technologists, on average, 1.3 minutes longer
per manual differential when they were using the CellaVision
DM96 than when using the standard microscopic method (P <
.01) ❚Table 9❚.
Standard Workflow Timing Studies
The timing studies using fully manual differentials do not
adequately reflect the usual workload of the differential work-
station of a clinical hematology laboratory. At our institution,
only approximately 50% of specimens that are flagged by the
automated cell counter and have a slide prepared require a full
manual differential. The other 50% of the slides are reviewed
(scanned) by the technologist, and the WBC differential results
of the automated cell counter are released, without need for a
full manual differential. To capture this aspect of the normal lab-
oratory workflow, 2 technologists (J.E.C. and G.H.B.) were
timed on randomly selected slides. These slides consisted of
cases that required a full manual differential and of scanned
slides that were reviewed only by the technologist. The 2 tech-
nologists selected for this study were a senior technologist with
a special interest in manual differentials who is the designated
trainer for the differential workstation for new employees and
an employee who had joined the laboratory more recently and
who had no special expertise in manual differentials. By this
Positive 1 3
Negative 6 86
Nucleated RBCs
Positive 9 5
Negative 15 67
* Based on the analysis of 96 blood smears. Data are given as the number of blood
smears positive or negative by the respective method.
time, these 2 reviewers had received more extensive training in
the use of the CellaVision DM96 than the technologists partici-
pating in the timing study involving only manual differentials.
All slides used for workflow timing were prepared with the
Miniprep blood smearing instrument.
As shown in ❚Table 10❚, there was no significant differ-
ence in the average time per slide for the very experienced
technologist. The technician with no special expertise in WBC
differentials, in contrast, was 25% faster using the CellaVision
DM96 than using the manual method. The average time per
slide for the first 77 slides that the technician analyzed with
the CellaVision DM96 was 3.1 minutes; for the last 83 slides,
her average time per slide using the CellaVision DM96 was
2.8 minutes. The reviewers thought that in all 236 cases, the
images presented by the CellaVision DM96 for the WBC dif-
ferential were adequate.
Discussion
The CellaVision DM96 is an automated digital image
analysis system. Routinely prepared peripheral blood smears
are scanned at low power, WBCs are identified, and digital

❚Table 4❚
Clinical Performance of the CellaVision DM96 Compared With Direct Microscopy as the Reference Method, Including All
Discrepant Cases

Cells Identified on a Slide Specificity (%) Sensitivity (%) False-Positive (%) False-Negative (%)

>3% Promyelocytes, myelocytes, 84 91 16 9

and/or metamyelocytes
Blasts 91 86 9 14
Other unusual WBCs 93 25 7 75
Nucleated RBCs 82 64 18 36

776 Am J Clin Pathol 2005;124:770-781 © American Society for Clinical Pathology

776 DOI: 10.1309/XMB9K0J41LHLATAY
 Hematopathology / ORIGINAL ARTICLE

❚Table 5❚
Rereview of Cases With Positive Findings by the CellaVision DM96 and Negative Findings by Direct Microscopy (False-Positives
on the CellaVision DM96)*

Case No. Resolution Category

Presence of immature myeloid cells (>3% promyelocytes, myelocytes, and/or metamyelocytes) on CellaVision DM96 differential and
not on manual differential
004 Misclassification of CellaVision images by technologist 3
016 4% immature myeloid cells present on CellaVision differential; 2% immature myeloid cells present on microscopic 1
differential
019 Original review and rereview of CellaVision images showed presence of 17% immature myeloid cells; original 4
microscopic review and first rereview showed no immature myeloid cells; third microscopic differential based
on 200 cells showed 6.5% immature myeloid cells
039 Misclassification of CellaVision images by technologist 3
056 Misclassification of CellaVision images by technologist 3

Downloaded from https://academic.oup.com/ajcp/article/124/5/770/1759765 by guest on 06 February 2024

071 Misclassification of CellaVision images by technologist 3
076 Original review and rereview of CellaVision images showed presence of 4% immature myeloid cells; original 1
microscopic review showed 0% immature myeloid cells; microscopic rereview of the slide showed 2%
immature myeloid cells
098 Misclassification of CellaVision images by technologist 3
104 Original review and rereview of CellaVision images showed presence of 13% immature myeloid cells; original 4
microscopic review and first rereview showed no immature myeloid cells; third microscopic differential showed
8% immature myeloid cells
118 Cells classified as promyelocytes by CellaVision classified as “Others, immature myeloid cells” on the microscopic 2
differential
119 Original review and rereview of CellaVision images showed presence of 4% immature myeloid cells; original 1
microscopic review and rereview showed no immature myeloid cells
120 Original review and rereview of CellaVision images showed presence of 8% immature myeloid cells; original 4
microscopic review and first rereview showed no immature myeloid cells; third microscopic differential based
on 300 cells showed 2.3% immature myeloid cells
Presence of blasts on CellaVision DM96 differential and not on manual differential
013 Misclassification of CellaVision images by technologist 3
047 Misclassification of CellaVision images by technologist 3
051 Misclassification of CellaVision images by technologist 3
070 Original review and rereview of CellaVision images showed presence of 1% blasts; microscopic review and 1
rereview showed no blasts
083 Misclassification of CellaVision images by technologist 3
090 Misclassification of CellaVision images by technologist 3
108 Blasts had been classified as “Other, immature myeloid cells” on manual differential 2
Presence of unusual cells on CellaVision DM96 differential and not on manual differential
046 Misclassification of CellaVision images by technologist 3
085 Original review and rereview of CellaVision DM96 images showed presence of 1% plasma cells; microscopic 1
review and rereview showed no plasma cells
094 Original review and rereview of CellaVision images showed 1% prolymphocytes; microscopic review and rereview 1
showed no prolymphocytes
101 Misclassification of CellaVision images by technologist 3
109 Cells classified as “Other, immature myeloid cells” on CellaVision classified as blasts on microscopic review 2
110 Misclassification of CellaVision images by technologist 3
Presence of nucleated RBCs (nRBCs) on CellaVision DM96 differential and not on manual differential
019 1 nRBC/100 WBCs on CellaVision; no nRBCs on manual differential 1
027 Misclassification of CellaVision images by technologist 3
033 Misclassification of CellaVision images by technologist 3
051 2 nRBCs/100 WBCs on CellaVision; no nRBCs on manual differential 1
052 4 nRBCs/100 WBCs on CellaVision; no nRBCs on manual differential 1
062 1 nRBC/100 WBCs on CellaVision; no nRBCs on manual differential 1
065 1 nRBC/100 WBCs on CellaVision; no nRBCs on manual differential 1
072 1 nRBC/100 WBCs on CellaVision; no nRBCs on manual differential 1
078 1 nRBC/100 WBCs on CellaVision; no nRBCs on manual differential 1
083 3 nRBCs/100 WBCs on CellaVision; no nRBCs on manual differential 1
098 2 nRBCs/100 WBCs on CellaVision; no nRBCs on manual differential 1
099 4 nRBCs/100 WBCs on CellaVision; no nRBCs on manual differential 1
104 1 nRBC/100 WBCs on CellaVision; no nRBCs on manual differential 1
118 1 nRBC/100 WBCs on CellaVision; no nRBCs on manual differential 1
120 2 nRBCs/100 WBCs on CellaVision; no nRBCs on manual differential 1
* Category 1, the discrepancy could be explained by random variation (ie, results of both methods were within the 95% confidence interval of the number of cells counted);
category 2, cases with essentially identical diagnoses (eg, the technologist used the expression “blasts” with one method and language such as “other cells; immature myeloid
cells, query blasts” with the other); category 3, rereview of the digital images obtained by the CellaVision indicated a misclassification by the technologist of cells shown on the
images; category 4, an abnormality identified by one method was not identified with the other and the discrepancy could not be explained by random variation within the 95%
confidence interval, differences in nomenclature, or misclassification of cells on digital images. Analysis was performed twice for all cases listed (once as part of the original
analysis and then as rereview of cases with discrepant findings between direct microscopy and the CellaVision). As described in the “Results” section, a third microscopic
review then was performed on cases 019, 104, and 120 to resolve the remaining discrepancy.

© American Society for Clinical Pathology Am J Clin Pathol 2005;124:770-781 777

777 DOI: 10.1309/XMB9K0J41LHLATAY 777
Kratz et al / DIGITAL IMAGE ANALYSIS IN HEMATOLOGY

❚Table 6❚
Rereview of Cases With Negative Findings by the CellaVision DM96 and Positive Findings by Direct Microscopy (False-Negatives
on the CellaVision DM96)*

Case No. Resolution Category

Presence of immature myeloid cells (>3% promyelocytes, myelocytes, and/or metamyelocytes) on microscopic differential and
not on CellaVision DM96 differential
046 3% immature myeloid cells present on CellaVision; 4% immature myeloid cells present on microscopic 1
differential
052 Misclassification of CellaVision images by technologist 3
Presence of blasts on microscopic differential and not on CellaVision DM96 differential
052 No blasts seen on CellaVision; 2% blasts seen on microscopic differential 1
109 Blasts classified as “Others (immature myeloid cells)” on microscopic differential 2
Presence of unusual cells on microscopic differential and not on CellaVision DM96 differential
089 “Others (immature myeloid cells)” called blasts on the CellaVision 2

Downloaded from https://academic.oup.com/ajcp/article/124/5/770/1759765 by guest on 06 February 2024

090 Review and rereview of CellaVision differential showed no unusual cells; 1% prolymphocytes seen on 1
microscopic differential
108 “Others (immature myeloid cells)” called blasts on the CellaVision 2
Presence of nucleated RBCs (nRBCs) on microscopic differential but not on CellaVision DM96 differential
094 No nRBCs seen on CellaVision, 1 nRBC/100 WBCs seen on microscopic differential 1
095 No nRBCs seen on CellaVision, 1 nRBC/100 WBCs seen on microscopic differential 1
106 No nRBCs seen on CellaVision, 2 nRBCs/100 WBCs seen on microscopic differential 1
109 No nRBCs seen on CellaVision, 1 nRBC/100 WBCs seen on microscopic differential 1
112 No nRBCs seen on CellaVision, 1 nRBC/100 WBCs seen on microscopic differential 1
* See Table 5 for an explanation of the categories.

images of these cells are taken at high magnification. An ❚Table 7❚

Summary of Rereview of Cases With Differing Results on
artificial neural network analyzes the pictures and preclassi- Manual Microscopy vs CellaVision DM96*
fies them according to WBC class. Preclassified WBCs are
presented to a reviewer for confirmation or reclassification Category

(Figure 1). 1 2 3 4
We compared the clinical performance of this system
CellaVision false-positive cases
with the standard direct microscopic method. In 11 (9.2%) of Immature myeloid cells 3 1 5 3
120 cases, the technologists thought the cells presented by the Blasts 1 1 5 0
Unusual cells 2 1 3 0
CellaVision DM96 were inadequate for a reliable WBC differ- Nucleated RBCs 13 0 2 0
ential. Such a high rejection rate would seriously interfere CellaVision false-negative cases
with the ability of a clinical laboratory to report differentials Immature myeloid cells 1 0 1 0
Blasts 1 1 0 0
in a timely manner; however, 10 of these 11 cases had been Unusual cells 1 2 0 0
prepared with the manual push-pull method with a spreader Nucleated RBCs 5 0 0 0
slide; only 1 of the cases had been prepared with a semiauto- * Data are given as number of cases. See Table 5 for an explanation of the categories.
mated slide maker. In the standard workflow timing studies,
the technologists were asked to use the CellaVision DM96 to
review 236 slides prepared with the semiautomated method. decreases the number of cases with inadequate digital images
They thought that for all cases the images presented for the presented by the CellaVision DM96 and allows the perfor-
WBC differential were adequate. This indicates that the use of mance of the overwhelming majority of manual differentials
an automated or semiautomated slide maker significantly using the CellaVision DM96.

❚Table 8❚
Clinical Performance of the CellaVision DM96 Compared With Direct Microscopy as the Reference Method, Excluding
Discrepancies Due to Statistically Insignificant Variations (Within the 95% Confidence Interval of the Cell Count) and
to Clinically Insignificant Variations in Nomenclature

Cells Identified on a Slide Specificity (%) Sensitivity (%) False-Positive (%) False-Negative (%)

>3% Promyelocytes, myelocytes, 88 95 12 5

and/or metamyelocytes
Blasts 94 100 6 0
Other unusual WBCs 97 100 3 0
Nucleated RBCs 97 100 3 0

778 Am J Clin Pathol 2005;124:770-781 © American Society for Clinical Pathology

778 DOI: 10.1309/XMB9K0J41LHLATAY

❚Table 9❚
Average Time per Slide for 6 Technologists Performing 10
Fully Manual Differential Counts With the Microscope or the
CellaVision DM96*
Technologist Microscope CellaVision
A 3.6 4.5
B 4.5 5.0
C 4.7 5.8
D 4.9 6.6
E 5.9 7.5
F 7.2 9.0
Average for all technologists 5.1 6.4
* Times are given in minutes.
It is possible that in laboratories that decide to maintain the
manual push-pull method for slide preparation, feedback provid-
ed to technologists will improve the quality of smear preparation.
The importance of the enforcement of rigorous quality standards
for slide preparation was recently stressed by Sandhaus and
coworkers.11 Among cases with an appropriate smear quality, the
ability of the CellaVision DM96 to find a sufficient number of
WBCs suitable for a reliable differential was at least equivalent if
not better than the direct microscopic method (Figure 2).
The number of cases in which the instrument provided
images inadequate for RBC and platelet analysis was high
(23%) and did not improve with the use of a semiautomated
slide maker. By using the microscope, the reviewers were able
to find areas appropriate for the evaluation of RBC and platelet
morphologic features on 100% of these slides. It is possible
that the percentage of slides with inadequate RBC areas pre-
sented by the CellaVision DM96 can be decreased by better
slide preparation, better training of technologists in the use of
the CellaVision DM96 for this purpose, and/or adjustments in
the CellaVision DM96. Otherwise, laboratories using the
CellaVision DM96 may have to decide not to provide informa-
tion on RBC and platelet morphologic features as part of their
standard manual differentials, unless there is reason to suspect
an RBC or a platelet abnormality (ie, an RBC- or a platelet-
specific flag from the automated cell counter). Such a develop-
ment would be in line with the fact that already, neither RBC
nor platelet morphologic features are reviewed microscopical-
ly for differential samples that are reported without preparation
of a slide directly from the automated cell counter.
❚Table 10❚
Average Time per Slide Using the Standard Workload of the Differential Workstation (Approximately 50% Fully Manual
Differential Counts and 50% Scanned Slides With Release of the Automated Differential Count)
Staff Member/No. of Slides
Very experienced technologist with special expertise in WBC differentials (n = 76)
Technician with no special expertise in WBC differentials (n = 160)
© American Society for Clinical Pathology
Hematopathology / ORIGINAL ARTICLE
The CellaVision DM96 presents preclassified WBCs to a
reviewer for approval or reclassification. We found that the
instrument overall correctly preclassified 82% of all cells.
Preclassification of WBCs was most accurate for segmented
neutrophils (92.5% correctly classified), lymphocytes (96.4%),
and monocytes (81.4%); it was less reproducible for basophils
(80.0%) and for eosinophils (63.2%) (Table 2). A study based
on images of WBCs sent to 2,400 laboratories in surveys of
the College of American Pathologists and classified by human
observers was published in 197712; 99% of segmented neu-
trophils, 96% of lymphocytes, 87% of monocytes, 95% of
basophils, and 96% of eosinophils were identified correctly by
Downloaded from https://academic.oup.com/ajcp/article/124/5/770/1759765 by guest on 06 February 2024
the participants. This indicates that the ability of the CellaVision
DM96 to preclassify segmented neutrophils, lymphocytes,
and monocytes is similar to that of the human observer,
whereas improvements to the neural network are needed for
eosinophils and basophils.
After the WBCs presented by the instrument are reclassi-
fied or confirmed by the user, final results are released. We
compared the final results obtained with digital image analy-
sis to differentials from direct microscopy. The correlation
coefficients for differential counts for the WBC classes found
in blood samples with a qualitatively normal differential were
between 0.67 and 0.96 (Figure 3). Koepke and colleagues13
evaluated the performance of differential WBC counting by
73 technologists and technicians in 5 different laboratories in
a large medical center by comparing them with differentials
performed by referees. Correlation coefficients were lowest
for basophils (0.32) and monocytes (0.41) and highest for
lymphocytes (0.73), eosinophils (0.83), and segmented neu-
trophils (0.87). The correlation coefficients for the cell counts
from the CellaVision DM96 compare quite favorably with
these results. Correlation for the various cell classes with the
direct microscopic differential also was very similar to the
findings reported by Swolin and colleagues6 for the
DiffMaster Octavia, the predecessor of the CellaVision
DM96. The low correlation for band forms might be due to the
known high variability in band counts.14
There was a significant percentage of cases that were
false-positive or false-negative for clinically significant abnor-
malities on the CellaVision DM96 compared with the results
of direct microscopy. Similar findings were reported by
Koepke and coworkers,13 who found that the sensitivity of
Microscope CellaVision DM96
1.4 1.5
4.1 3.0
Am J Clin Pathol 2005;124:770-781 779
779 DOI: 10.1309/XMB9K0J41LHLATAY 779
Kratz et al / DIGITAL IMAGE ANALYSIS IN HEMATOLOGY
slide review by technologists ranged from 34% to 100%,
depending on the abnormality. Review of the discrepant cases
in our study showed that the majority of discrepant cases were
due to variation that was not statistically significant (within
the 95% confidence interval of the number of cells on which
the differential was based) or to clinically insignificant varia-
tions in nomenclature (eg, blasts vs other cells, query blasts).
When we corrected for these statistically or clinically insignif-
icant discrepancies, the sensitivity of the CellaVision DM96
was between 95% and 100% and the specificity between 88%
and 97%, depending on the abnormality.
The remaining discrepancies were mostly (16/19 dis-
crepant cases) due to misidentification of cells on digital
images by the technologists. In the remaining 3 discrepant
cases, cells that were clearly present on the digital images
obtained with the CellaVision DM96 had not been reported on
direct microscopy. A rereview of these slides by experienced
reviewers at first did not allow the visualization of these cells
with the microscope. When an additional microscopic differ-
ential based on up to 300 cells was performed on these slides,
the immature myeloid cells seen with the CellaVision DM96
were seen with the microscope. Therefore, these discrepant
cases were most likely due to random variation outside the
95% confidence interval of a 100-cell differential count.
As described in the preceding paragraphs, when we found
discrepancies between the findings with direct microscopy
and with the CellaVision DM96, we rereviewed the digital
images provided by the CellaVision DM96 and determined
that in a number of cases, the initial reviewers had misidenti-
fied cells. This episode underlines a strength of digital image
analysis compared with direct microscopy: we were able to do
quality control on a cell-by-cell basis, identify errors, correct
them, and perform corrective action by educating the technol-
ogists who had made the misidentification. Such interventions
are much more difficult or impossible using direct
microscopy. Even after correction for misclassified cells, there
were more positive findings on the CellaVision DM96 than on
direct microscopy. Most of these false-positive findings
involved very small populations of cells (<5%). Since the
presence of abnormal WBCs will only be recognized with
greater than 95% certainty in a 100-cell differential if the per-
centage of the abnormal cell population is at least 5%, random
variation is the most likely explanation for these “false posi-
tives.”15 Another possibility is that some cells were misclassi-
fied or missed on direct microscopy. Finally, it is very plausi-
ble that some of the positive findings identified with the
CellaVision DM96 and not reported on direct microscopy
were due to the fact that it is not possible for the technologist
to “skip” cells on the CellaVision DM96. Every cell has to be
classified (and this classification remains subject to rereview),
leading to a higher likelihood of reporting rare or unusual cells
when using the CellaVision DM96.
780 Am J Clin Pathol 2005;124:770-781
780 DOI: 10.1309/XMB9K0J41LHLATAY
To evaluate the impact of the CellaVision DM96 on labor
costs, we performed 2 studies. First, we timed 6 technologists
performing 10 fully manual differentials with the microscope
and with the CellaVision DM96. It took the technologists, on
average, 5.1 minutes per slide to perform a manual differential;
the time required with the CellaVision DM96 was longer, 6.4
minutes. The limitations of this study are that the technologists
had years of experience in performing differentials with a
microscope; in contrast, the CellaVision DM96 was new to
them, and they lacked any prolonged experience on this instru-
ment. It therefore is possible that with increased exposure to the
CellaVision DM96, the average time per manual differential
Downloaded from https://academic.oup.com/ajcp/article/124/5/770/1759765 by guest on 06 February 2024
would have decreased. The second limitation of this study was
that it addressed only fully manual differentials; the usual work-
flow at the differential workstation in our laboratory consists of
approximately 50% manual differentials and 50% slides that
only need to be reviewed for release of the differential results
from the automated cell counter (scanned slides). It is possible
that the major labor saving that can be derived from the
CellaVision DM96 will be from these scanned slides, for which
the technologist is conveniently presented with a large number
of WBCs on a computer screen that can be reviewed quickly for
the presence of any abnormality, with subsequent release of the
differential results from the automated cell counter.
To address these possibilities, we performed a second tim-
ing study, in which 2 employees were trained more extensively
in the use of the CellaVision DM96. They were asked to result
randomly selected cases with the microscope and to process the
same cases with the CellaVision DM96. These cases consisted
of both fully manual differentials and of slides that required
only a scan. As shown in Table 10, it took a very experienced
technologist approximately the same time per slide to result dif-
ferentials from the CellaVision DM96 (1.5 minutes per slide) as
from the microscope (1.4 minutes per slide). However, a less
experienced employee was faster using the CellaVision DM96
(3.0 minutes per case) than using the microscope (4.1 minutes
per case), indicating that after training on the CellaVision
DM96 and with a caseload similar to the normal workflow in
the laboratory, an average technologist is likely to be more pro-
ductive using the CellaVision DM96 than using the microscope.
The possibility of a learning curve was also indicated by the fact
that the less experienced employee was 10% faster using the
CellaVision DM96 when analyzing the second half of her
assigned slides than when analyzing the first half.
The cost-effectiveness of the CellaVision DM96 will be
determined by a variety of factors. Our data indicate that the
instrument may allow some labor savings by decreasing the
time for technologists trained in the use of the instrument to
release a differential. Increased efficiencies also could be
achieved by centralizing the expertise for morphologic diag-
nosis and by making it less time-consuming to obtain confir-
mation of abnormal findings by referees. These cost savings
© American Society for Clinical Pathology

might be offset partly by the costs of the instrument, its asso-
ciated software, and necessary updates in information systems
infrastructure. Our prediction is that the CellaVision DM96
will be cost-effective in high-volume laboratories, where
small gains in the time spent per differential can translate into
major labor savings. It also may lead to significant labor sav-
ings in small laboratories that presently have to maintain
expertise in microscopy and that would be able to rely on
experts in remote locations by using the CellaVision DM96.
Our study did not address a number of advantages of the
CellaVision DM96 that do not lend themselves easily to quan-
titation. Use of the CellaVision DM96 will allow laboratories
to store and easily access digital images of WBCs from previ-
ous differentials, allowing comparison of present findings
with historic data for a patient. The ability to determine on a
cell-by-cell basis how a technologist classified a cell should
make training, competency assessment, quality control, and
corrective action easier and more effective. Digital images of
cells can be reviewed remotely, allowing the concentration of
expertise in one location (which can be as close as in a differ-
ent room or building of the same medical center or thousands
of miles away) and easy consultation with experts in the field.
These developments should improve patient care in ways that
are difficult to quantify but no less important. Clinicians, who
may like to review their patients’ blood smears, could do so
more easily from their own computer workstations, without
the burden of preparing additional blood smears and the
repeated search for rare diagnostic abnormalities. Further
studies in a setting in which the CellaVision DM96 is in clin-
ical use will be necessary to explore such impacts of this new
technology. Such studies also will allow determination of
intraobserver and interobserver variability in WBC differen-
tials performed by technologists and pathologists.
Automated analysis of slides is already in daily clinical use
in cervical cytology, providing significant savings in labor costs
and more reproducible clinical results.16,17 This precedent shows
that such systems can have an important role in a modern clini-
cal laboratory. Our evaluation of the CellaVision DM96, a new
automated digital image analysis system for the clinical hema-
tology laboratory, showed it to have performance at least equal
to that of direct clinical microscopy. Further studies in a live clin-
ical setting will be needed to determine the impact of this instru-
ment on the workflow in the clinical laboratory, on cost, on the
overall quality of differential results and clinical care, and on the
education of laboratory employees and clinicians.
From the 1Department of Pathology, Division of Laboratory
Medicine, Massachusetts General Hospital and 2Harvard Medical
School, Boston; and 3CellaVision AB, Ideon Research Park, Lund,
Sweden.
Address reprint requests to Dr Kratz: Clinical Hematology
Laboratory, Division of Laboratory Medicine, Massachusetts
General Hospital, 55 Fruit St, GRJ 235, Boston, MA 02114.
© American Society for Clinical Pathology
Hematopathology / ORIGINAL ARTICLE
* Ms. Casey received an honorarium from Sysmex for a
scientific presentation.
Acknowledgments: We thank Amy O’Carroll, Nancy Zhang,
and Barbara Pereira for participating in the timing study; Svetal
Patel, Hemali Patel, and the other staff members of the
Massachusetts General Hospital Clinical Hematology Laboratory
for their help and advice. We also acknowledge Lisa Palm, Pat
LaMothe, and other staff members of CellaVision AB, and Debora
Low, Sysmex America, for help in data collection and analysis.
References
1. Ehrlich P. Histologie und Klinik des Blutes. Berlin, Germany:
Downloaded from https://academic.oup.com/ajcp/article/124/5/770/1759765 by guest on 06 February 2024
August Hirschwald; 1891.
2. Tatsumi N, Pierre RV. Automated image processing: past,
present, and future of blood cell morphology identification.
Clin Lab Med. 2002;22:299-315, viii.
3. Simson E, Groner W. The state of the art for the automated
WBC differential. Lab Hematol. 1995;1:13-22.
4. Riley RS, Ben-Ezra JM, Massey D, et al. The virtual blood
film. Clin Lab Med. 2002;22:317-345.
5. Beksac M, Beksac MS, Tipi VB, et al. An artificial intelligent
diagnostic system on differential recognition of hematopoietic
cells from microscopic images. Cytometry. 1997;30:145-150.
6. Swolin B, Simonsson P, Backman S, et al. Differential
counting of blood leukocytes using automated microscopy and
a decision support system based on artificial neural networks:
evaluation of DiffMaster Octavia. Clin Lab Haematol.
2003;25:139-147.
7. Brown BA. Routine hematology procedures. In: Hematology:
Principles and Procedures. Philadelphia, PA: Lippincott;
1993:83-126.
8. National Committee for Clinical Laboratory Standards.
Reference leukocyte differential count (proportional) and
evaluation of instrumental methods; Approved document
H20-A. Villanova, PA: NCCLS; 1992.
9. Bain BJ. Performing a blood count. In: Blood Cells. A Practical
Guide. 3rd ed. Oxford, England: Blackwell Science; 2002:16-51.
10. Rumke CL. Imprecision of ratio-derived differential leukocyte
counts. Blood Cells. 1985;11:311-314, 315.
11. Sandhaus L, Dillman CA, Clement R, et al. Errors in the
hematology laboratory: why do they occur and what can we do
to reduce them? Lab Hematol. 2004;10:197-199.
12. Koepke JA. A delineation of performance criteria for the
differentiation of leukocytes. Am J Clin Pathol. 1977;68:202-
206.
13. Koepke JA, Dotson MA, Shifman MA. A critical evaluation
of the manual/visual differential leukocyte counting method.
Blood Cells. 1985;11:173-186.
14. Cornbleet PJ. Clinical utility of the band count. Clin Lab Med.
2002;22:101-136.
15. Rumke CL. Statistical reflections on finding atypical cells.
Blood Cells. 1985;11:141-144.
16. Parker EM, Foti JA, Wilbur DC. FocalPoint slide classification
algorithms show robust performance in classification of high-
grade lesions on SurePath liquid-based cervical cytology slides.
Diagn Cytopathol. 2004;30:107-110.
17. Wilbur DC, Prey MU, Miller WM, et al. AutoPap system
detection of infections and benign cellular changes: results
from primary screener clinical trials. Diagn Cytopathol.
1999;21:355-358.
Am J Clin Pathol 2005;124:770-781 781
781 DOI: 10.1309/XMB9K0J41LHLATAY 781