ORIGINAL ARTICLE INTERNATIONAL JOURNAL OF LA BO RATO RY HEMATOLOGY

Can automated blood film analysis replace the manual

differential? An evaluation of the CellaVision DM96
automated image analysis system
C. BRIGGS*, I. LONGAIR*, M. SLAVIK †, K. THWAITE †, R. MILLS*, V. THAVARAJA*, A. FOSTER ‡,
D. ROMANIN ‡, S. J. MACHIN*

*Department of Haematology, SUMMARY

University College London
Hospital, London, UK
†
Department of Haematology, The
Doctors Laboratory, London, UK
‡
Sysmex UK Ltd., Sysmex House,
Garamonde Drive, Wymbush,
Milton Keynes, UK
Correspondence:
Carol Briggs, Department of Hae-
matology, University College Lon-
don Hospital, 60 Whitfield Street,
London W1T 4EU, UK.
Tel.: +44 207 3809882;
E-mail: carolbriggs@hotmail.co.uk
doi:10.1111/j.1751-553X.2007.01002.x
Received 26 June 2007; accepted
for publication 2 October 2007
Keywords
Leucocyte differential, microscopy,
automated image analysis
Automation of differentials is desirable for economic and time-saving
reasons. Over the last 20 years, automated imaging processes have
started to be introduced where stained blood films are scanned by a
computer-driven microscope and leucocytes classified; however, early
methods were slow and had difficulty in classifying abnormal cells.
More recently the CellaVision DM96 (CellaVision AB, Lund, Swe-
den) has been introduced with added features such as continuous
loading of slides and a faster throughput than previous instruments.
The accuracy of CellaVisionTM DM96 has been evaluated by comparing
results to reference manual differentials. Results from different oper-
ators using the DM96 were compared with their own manual
differential and to a 400-cell reference manual differential. Precision
of the instrument was compared to the manual differential. The
preclassification accuracy of the DM96 was 89.2%. Precision was
similar to that of the 100-cell manual differential. The DM96 was faster
than the manual method, even after reclassification by a laboratory
scientist of any cells wrongly categorized by the instrument. The DM96
accuracy in morphological classification of leucocytes and red blood
cells; depends upon both blood pathology and experience of the
laboratory scientist using the instrument. For some cell types and
operators, DM96 accuracy was better than the individual’s 100 cell
manual differential.

Staff members represent the major expenditure in

INTRODUCTION
most laboratories so in many places where their num-
Over recent years, there has been increasing demand bers have not increased or have even been reduced,
for haematological tests with clearly defined turn- there is now more work for fewer laboratory scien-
around times along with a need for cost containment. tists. Fortunately, there have been advances in blood
 2007 The Authors
48 Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60
 C. BRIGGS ET AL.
cell counting instrumentation and this has resulted in
a reduction in the need for manual blood film review
and white blood cell (WBC) differentials (Lewis &
Rowan, 1991). The examination of blood films by
microscopy remains one of the major labour-intensive
procedures in the laboratory and the challenge is to
reduce the number of films examined without missing
important diagnostic information. Automated blood
cell counters offer a leucocyte count, red cell and
platelet count and five-part (some six-part) leucocyte
differential. In addition, some instruments provide a
nucleated red blood cell (NRBC) count. Haematology
instrument differentials provide only limited informa-
tion on cell morphology using abnormal cell flags and
are often unable to reliably classify abnormal and
immature cells. A blood film is examined for a num-
ber of reasons: to explain an unexpected blood count,
to examine red cell and platelet morphology, to con-
firm an abnormal automated leucocyte count or to
undertake an extended differential including imma-
ture and abnormal cells.
The examination of blood films is not only time
consuming and labour intensive but it also requires
highly trained staff. The impact of a wrong diagnosis
necessitates that experienced staff are present in the
laboratory 24 h a day. However, manual cell classifi-
cation is subjective, with significant inter- and intra-
observer variation (Koepke, Dotson & Shifmann,
1985) and any count is also subject to significant sta-
tistical variance (Rumke, 1985a).
Automated morphological analysis systems have
been introduced in the past, the first being described
in 1966 (Prewitt & Mendlesohn, 1966). These earlier
systems were slow, offering a limited degree of auto-
mation and therefore failed to provide significant
improvements in workflow and counting statistics
(Bentley, 1990).
More recently the CellaVision DM96 (CellaVision
AB, Lund, Sweden) has been introduced. The instru-
ment scans the slides at low power to identify poten-
tial WBCs and then takes digital images at high
magnification. The images are analysed by an artificial
neural network based on a database of cells and pre-
classified according to WBC class. The cells are pre-
sented on a computer screen for conformation or
reclassification. The system also allows review of the
red cell and platelet morphology and estimation of the
platelet count.
 2007 The Authors
Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60
EVALUATION OF THE CELLAVISION DM96 49
In this study, we have evaluated the accuracy and
precision of the DM96 in identifying all classes of
leucocytes including abnormal and immature cells. The
results from five different laboratory scientists’ manual
differentials and those from the DM96 have been com-
pared with a standard reference manual differential
(Clinical and Laboratory Standards Institute H20-A2,
2007) and the time taken for these slides to be analysed
was recorded. An evaluation of the red cell morphology
from the DM96 has also been undertaken.
For any automated image system to be introduced
in the haematology laboratory, it will have to demon-
strate that it is at least as reliable, or preferably more
reliable, in correctly identifying WBC as the routine
manual method and should also provide significant
labour savings and so costs to the laboratory.
MATERIALS AND METHODS
Blood samples
Residual patient blood samples, after all requested hae-
matology tests had been completed, were obtained
from the routine workload of the Department of
Haematology at University College London Hospital.
Samples were collected into K2EDTA as anticoagulant
(Becton Dickinson, Franklin Lakes, NJ, USA). After
collection samples were stored at room temperature,
blood films were prepared and stained using an auto-
mated slide maker and stainer (SP-100, Sysmex, Kobe,
Japan). For the evaluation 136 samples were selected,
these included 45 normal samples, all parameters
within our laboratory reference ranges, and the rest
consisting of various clinical conditions such as infec-
tions, chronic and acute leukaemais, lymphomas, hea-
moglobinopathies, iron deficiency, renal disease, liver
disease, HIV and chemotherapy. The WBC in these
samples ranged from 0.51 · 109/l to 73.89 · 109/l. All
samples included in this study had been analysed on
the Sysmex XE-2100 to ascertain whether the results
were normal or abnormal and to collect the WBC in
order to calculate the absolute values for the leuco-
cytes from the DM96 percentage results. For the preci-
sion study, 10 samples were selected with varying
WBC counts, both normal and abnormal samples were
included. For the comparison of results between oper-
ators 30 samples were selected, again normal and
abnormal samples were included.
50 C. BRIGGS ET AL. EVALUATION OF THE CELLAVISION DM96
Reference method
The reference method was the manual differential
count of 200 cells performed by two experienced labo-
ratory scientists (Clinical and Laboratory Standards
Institute H20-A2, 2007) using light microscopy, if
there were discrepancies between the two results a
third 200-cell differential was performed by a different
laboratory scientist.
Analysis with CellaVision DM96
The DM96 consists of a slide scanning unit and a com-
puter with the CellaVision Blood (CellaVision, AB,
Lund, Sweden) differential Software (version 1.5). The
scanning unit consists of a motorized microscope (·10,
·50 and ·100 objective), a digital CCD camera, an
automatic immersion oil unit, slide feeder unit with
barcode reader, rack feeder unit and a control unit
which controls motors, sensors, oil applying and illu-
mination. Slides are loaded onto the system and pro-
cessed immediately and continuously. The instrument
scans the slides, identifies potential WBCs, takes digital
images of them and uses artificial neural network-
based software to analyse the cells. Digital images of
preclassified cells are presented to the operator on a
Figure 1. Preclassified white blood cells presented on the CellaVision DM96 computer screen.
Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60
user definable computer display (Figure 1). The opera-
tor can enlarge single cells for a more detailed view,
leave cells in the category suggested, accept the pre-
classification or move individual cells to a different
category, reclassify, by clicking on the cell and using
the drag and drop function. Results from the DM96
are not complete and will not be released until all
preclassified cell categories have been reviewed. To
conform with the way our laboratory reports WBC
differentials the band cell count was added to the
segmented neutrophil count (band cells are not
counted separately) and the variant lymphocyte count
and plasma cell count were added to the total lympho-
cyte count, atypical lymphocytes and plasma cells are
commented on, but not quantitated in the routine
manual differential. For the evaluation of the red cell
and platelet morphology, the DM96 provides an image
which corresponds to eight areas of the smear viewed
at ·100 objective. The red cell and platelet images can
be enhanced by clicking on the screen to magnify the
cells. Red blood cell morphology is partially classified
by the DM96; polychromasia, hypochromia, micro-
cytosis, macrocytosis, anisocytosis and poikilocytosis
are reported using a scale 0–3. Abnormalities in red
cell shape, such as fragmented red cells or sickle cells
are not automatically classified by the DM96 but can
 2007 The Authors
 C. BRIGGS ET AL.
be added by the user after examining the image pre-
sented by the instrument. The default settings for the
number of cells exhibiting an abnormality and being
classified as normal or 1–3 severity can be adjusted by
the user. An estimate of platelet count can be made by
the DM96 if this feature is selected. The image is then
subdivided into sub-images and the user is then
required to count the number of platelets in each field
(or sub-image), a calculated estimate of the platelet
concentration is then made based on the counts per
field and a platelet estimate factor. This feature was
not selected or evaluated in this study.
Accuracy
One hundred and thirty-six samples were analysed on
the DM96 and by a 100-cell differential, which is our
established routine practice, by an experienced labora-
tory scientist. WBC counts ranged from 0.51 · 109/l
to 73.89 · 109/l. The operator was not provided with
any previous or clinical information on the patient or
results from the haematology analyser. The DM96
was set to count 105 cells; this was to allow for the
possible inclusion of artefacts in the WBC count.
However, in samples with low WBC counts 100-cell
counts were not always achieved on the DM96. The
accuracy of red blood cell analysis on the DM96 was
also assessed by comparing morphological changes
with those reported by the microscopic method.
Accuracy of results when the DM96 is operated by
different laboratory scientists
Laboratory scientists were selected with different lev-
els of experience in blood cell morphology. Two were
very experienced, one moderately experienced and
two newly qualified members of staff who perform
differentials routinely in the laboratory, but are still
expected to need some help with difficult blood films.
For this study, the operators were not provided with
any previous or clinical information on the patient or
results from the haematology analyser. Thirty blood
films were first analysed on the DM96 and then by a
100-cell microscopic differential. The samples con-
sisted of 10 from normal patients and the remaining
with abnormal WBCs and red blood cells present,
WBC count range 0.51–51.63 · 109/l. The 100-cell
manual differentials for each individual were com-
 2007 The Authors
Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60
EVALUATION OF THE CELLAVISION DM96 51
pared with the results for the preclassified and then
their reclassified cells from the DM96, additionally all
results were compared with the reference 400-cell dif-
ferential. The operators were also asked to evaluate
whether the areas presented by the DM96 for red
blood cell and platelet morphology were adequate.
Timing studies
The same 30 blood films and five laboratory scientists
were used for the timing study as for the accuracy
study above. The time taken to perform the total 30
100-cell differentials manually and on the DM96 from
loading the slides onto the instrument to the end of
reclassifying the cells was noted for each individual.
Precision
Ten samples with varying WBC counts, normal sam-
ples and samples containing abnormal cells were
selected. WBC counts ranged from 1.62 · 109/l to
32.87 · 109/l. Each blood film was analysed on the
DM96 five times and by the same five laboratory sci-
entists used for the above accuracy study.
Statistical analysis
Statistical analysis was performed using Microsoft Excel
(Microsoft Ltd., Reading, UK) and MINITABTM STATISTICAL
SOFTWARE (Minitab Ltd., Coventry, UK) (www.minitab.
com). Multiple comparisons between the different
WBC differential methods and by the different opera-
tors performing them were assessed using Friedman’s
ANOVA test. Specific differences were shown using
Mann–Whitney test for paired data with a P-value
<0.05 considered statistically significant.
RESULTS
Accuracy
The DM96 monitor shows leucocytes preclassified
according to cell type. The operator has to review
each cell and can move a cell to a different cell class
(reclassify the cell) or leave the cell in the category
suggested by the instrument. Table 1 shows the
percentage agreement of the various cell classes
preclassified correctly by the DM96. The percentage of
52 C. BRIGGS ET AL. EVALUATION OF THE CELLAVISION DM96

100-cells on 16 (11.8%) of samples, this was usually

Table 1. Percentage of cells from 286 blood films
correctly preclassified by the CellaVision DM96 when the total WBC count was <1.5 · 109/l but also
occurred if there was more than the usual amount of
Preclassifying artefact on the slide. All manual differentials counted
Cell class agreement (%) 100 cells. Pre- and postclassification results for the red
blood cells are presented in Table 2. Preclassification
Neutrophil (Neut) 99.5
Lymphocyte (Lymph) 94.9 agreement with the microscopic morphology for aniso-
Monocyte (Mono) 87.6 cytosis and macrocytosis is low; this is because of a
Eosinophil (Eos) 79.9 high number of false positives reported by the DM96.
Basophil (Baso) 54.1 Classification of the red cells by the DM96 failed for
Metamyelocyte 32.6 one sample and for seven samples (5%) the operator
Myelocyte 37.7
reported the cells presented by the DM96 as inade-
Promyelocyte 77.6
Blast 76.6 quate for morphological comment. All blood films
Nucleated red blood cell 89.6 were adequate when viewed under the microscope.
Neut, Lymph and Mono 97.3
Neut, Lymphs, Mono, Eos and Baso 87.2
All cell classes 89.2 Accuracy of results when the DM96 is operated by five
Abnormal cells called normal 0.9 different laboratory scientists
Normal cells misclassified as 9.1
other normal cells The correlation coefficients of determination (r2) from
Normal cells called abnormal cells 1.8 30 blood films analysed on the DM96 by five labora-
tory scientists are presented in Table 3. Each individ-
ual’s manual differential result as well as the results
correctly classified cells was highest for the mature for the preclassified and reclassified cells from the
cells and lowest for the immature granulocytes. NRBC DM96 are compared with the reference 400-cell
preclassification was good. Reclassification of the cells differential.
was performed by the same person who performed Correlation to the reference method is similar for
the 100-cell microscopic differential counts. an operator whether the differential is performed
Assessment of the accuracy of results was carried manually or by the DM96 with reclassification of cells.
out by determining the correlation coefficient (r2) Operators 1 and 2 were experienced morphologists,
between one experienced laboratory scientist’s manual operator 3 had moderate experience and 4 and 5 were
differential and the pre- and reclassified results from newly qualified. For some cell types and for some
the DM96 for all cell classes (Figure 2). Because of the operators, the accuracy of the DM96 compared with
low number and subjectivity of classification of met- the reference method was better than that of the indi-
amyelocytes, myelocytes and promyelocytes the results vidual’s 100-cell manual differential. Correlation with
for these cells are also demonstrated as a total imma- the reference method for the lymphocyte count from
ture granulocyte count. The correlation coefficients the reclassified results from the DM96 is very poor for
were higher after the cells had been reclassified. R2 operator 4. One sample contained 90% blasts which
values were greater than 0.9 for neutrophils (including the DM96 preclassified as lymphocytes, operator 4
band cells), lymphocytes, total immature granulocytes, failed to reclassify these cells while all other operators
blasts, NRBC and metamyelocytes. Correlation for the correctly reclassified the cells. If the results from this
reclassified monocytes was 0.81 and eosinophils slide are removed, correlation for lymphocytes by
0.67.There is one significant outlier on the preclassi- operator 4 improves to a r2 of 0.61 Correlation for the
fied vs. manual differential monocyte graph, the DM96 eosinophils on the DM96 are not as good as the man-
preclassified myeloblasts in this samples as monocytes. ual method even after reclassification.
By removing this data point, correlation is then Statistically the preclassification results for lympho-
improved to 0.77. Because of the low number of cells cytes, immature granulocytes, myelocytes and to a les-
counted, correlation for basophils, myelocytes and ser extent basophils had significantly higher values
promyelocytes was poorer. The DM96 failed to count than those by the reference method, reclassified results
 2007 The Authors
Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60
 C. BRIGGS ET AL. EVALUATION OF THE CELLAVISION DM96 53

Man v DM96 reclassified neutrophils Man v DM96 neutrophils

60 60
y = 1.0079x + 0.2756

DM 96 neutrophils x 109/l
50 y = 0.9923x + 0.3026 50 R 2 = 0.9507
neutrophils x 109/l
R 2 = 0.9859
DM96 reclassified

40 40

30 30

20 20

10 10

0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Manual neutrophils x 109/l Manual neutrophils x 109/l

Man v DM96 reclassified lymphocytes 40 Man v DM96 lymphocytes

35
35 y = 1.0597x + 0.8266

DM96 lymphocytes x 109/L

y = 0.9706x - 0.1226
30 R 2 = 0.7267
R 2 = 0.9591
lymphocytes x 109/l

30
DM96 reclassified

25
25
20
20
15 15
10 10
5 5
0 0
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35 40
Manual lymphocytes x 109/l Manual lymphocytes x 109/l

Man v DM96 reclassified monocytes 45 Man v DM96 monocytes

14
40
12
DM96 monocytes x 109/l

y = 0.8529x - 0.012 35
monocytes x 109/l
DM96 reclassified

10 R 2 = 0.805 30
y = 0.8393x + 0.363
8 25
R 2 = 0.1164
6 20
15
4
10
2
5
0 0
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
Manual monocytes x 109/l Manual monocytes x 109/l

Figure 2. Correlation of cell classification on the CellaVision DM96 with the 100-cell manual microscopic method.
Man vs. DM96: manual microscopic method vs. the preclassified results from the DM96; Man vs. DM96 reclassified:
manual microscopic method vs. the results from the DM96 after reclassification of cells; Imm Gran, immature gran-
ulocytes, the sum of metamyelocytes, myelocytes and promyelocytes.

or each operator’s manual differential. The DM96 has a
tendency to wrongly classify other cells into these cell
classes.
 2007 The Authors
Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60
For reclassified data, far fewer statistically signifi-
cant differences were identified between the specific
users’ manual and DM96 WBC differentials when
54 C. BRIGGS ET AL. EVALUATION OF THE CELLAVISION DM96

Man v DM96 reclassified eosinophils Man v DM96 eosinophils

3.0 3.0
y = 0.8044x + 0.1154

DM96 eosinophils x 109/l

eosinophils x 109/l 2.5 y = 0.997x + 0.0415 2.5 R 2= 0.4973
DM96 reclassified

R 2 = 0.672
2.0 2.0

1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Manual eosinophils x 109/l Manual eosinophils x 109/l

Man v DM96 reclassified basophils Man v DM96 basophils

1.2 4.0
y = 0.3346x + 0.0379 3.5
1.0 y = 0.0379x + 0.1292
R 2 = 0.0534 DM96 basophils x 109/l
DM96 reclassified

3.0
basophils x 109/l

R 2 = 0.00
0.8
2.5
0.6 2.0

0.4 1.5
1.0
0.2
0.5
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 0.0 0.2 0.4 0.6 0.8
Manual basophils x 109/l Manual basophils x 109/l

Man v DM96 reclassified Imm Gran Man v DM96 Imm Gran

12.0 y = 1.5927x + 0.1427
y = 1.2512x + 0.0363 14
R 2 = 0.9129
10.0 R 2= 0.9514
DM96 Imm Gran x 109/l

12
DM96 reclassified
Imm Gran x 109/l

8.0 10
8
6.0
6
4.0
4
2.0
2
0.0 0
0.0 2.0 4.0 6.0 8.0 10.0 12.0 0 2 4 6 8 10 12 14
Manual Imm Gran x 109/l Manual Imm Gran x 109/l

Figure 2. Continued.

compared with the reference method. The greatest counted. For lymphocytes, the median range of
number of statistically significant differences occurred reclassified cells on the DM96 is 1.20–1.42 · 109/L,
again with lymphocytes and basophils, for basophils for the manual differentials 1.07–1.52 · 109/l and for
this may be influenced by the low number of cells the reference method the count is 1.42 · 109/l. The
 2007 The Authors
Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60
 C. BRIGGS ET AL. EVALUATION OF THE CELLAVISION DM96 55

Man v DM96 reclassified blasts Man v DM96 blast

80 25
y = 0.2912x + 0.2572
70
R 2 = 0.6362
y = 0.9937x + 0.001 20

DM96 blasts x 109/l

60
DM96 reclassified

R 2 = 0.9953
blasts x 109/l

50
15
40
30 10

20
5
10
0 0
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
Manual blasts x 109/l Manual blasts x 109/l

Man v DM96 reclassified NRBC Man v DM96 NRBC

300 300

250 y = 1.0895x - 2.8732 250 y = 1.0142x - 2.9026

DM96 NRBC/100WBC
R 2= 0.973 R 2 = 0.9636
DM96 reclassified
NR BC/100WBC

200 200

150 150

100 100

50 50

0 0
0 50 100 150 200 250 300 0 50 100 150 200 250 300
Manual NRBC/100WBC
Manual NRBC/100WBC

Man v DM96 reclassified metamyelocyts Man v DM96 metamyelocytes

8.0 8.0
y = 0.9602x + 0.1206
7.0 y = 1.1202x + 0.005 7.0
DM metamyelocytess x 109/l

R 2 = 0.9331 R 2 = 0.7809
metam yelocytes x 109/l

6.0 6.0
DM96 reclassified

5.0 5.0

4.0 4.0

3.0 3.0

2.0 2.0

1.0 1.0

0.0 0.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0
Manual metamyelocytes x 109/l Manual metamyelocytes x 109/l

Figure 2. Continued.

two highly experienced operators showed better All operators found the quality of the WBC images
statistical agreement between the manual and DM96 presented to be excellent and the use of the instru-
differentials. ment to be simple.
 2007 The Authors
Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60
56 C. BRIGGS ET AL. EVALUATION OF THE CELLAVISION DM96

Man v DM96 reclassified myelocytes Man v DM96 myelocytes

DM96 reclasified myelocytes 3.5 7.0

3.0 6.0

DM96 myelocytes x 109/l

y = 0.7989x + 0.0884
2.5 R 2 = 0.3709 5.0 y = 1.3028x + 0.1861
R 2 = 0.311
2.0 4.0
x 109/l

1.5 3.0

1.0 2.0

0.5 1.0

0.0 0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Manual myelocytes x 109/l Manual myelocytes x 109/l

Man v DM96 reclassified promyelocytes Man v DM96 promyelocytes

2.5 3.0

DM96 promyelocytes x 109/l

2.0 2.5
promyelocytes x 109/l
DM96 reclasified

y = 0.4561x + 0.0273
2.0 R 2 = 0.248
1.5 y = 0.6811x + 0.0392
R 2 = 0.4175 1.5
1.0
1.0

0.5
0.5

0.0 0.0
0.0 0.5 1.0 1.5 2.0 2.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Manual promyelocytes x 109/l Manual promyelocytes x 109/l

Figure 2. Continued.

All but one area presented by the DM96 for red of samples with low WBC counts and highly abnor-
blood cell and platelet morphology was judged to be mal cells.
adequate.

Precision
Timing studies
There was no significant difference identified for preci-
Results for the timing study are presented in Table 4. sion on the DM96 preclassified, reclassified or the man-
The total time for the analysis of the 30 blood films ual differential for most cell classes except for basophils,
and reclassification of cells is similar for all operators blasts and NRBC. These differences in precision were
on the DM96, despite the variability of experience in found to be small and the differences between median
the operators (average time 1 h 20 min, range 1 h values ranged only from 0.07 to 0.2 · 109/l.
5 min–1 h 40 min). The time to perform the manual
differential is very different between operators (aver-
DISCUSSION
age time 2 h 54 min, range 1 h 20 min–4 h 10 min),
with the inexperienced taking much longer than the Advances in technology mean that for most blood
experienced, this is probably because of the inclusion samples automated haematology analysers can
 2007 The Authors
Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60
 C. BRIGGS ET AL.
Table 2. Percentage agreement for red cell morphology
on the CellaVision DM96
Red cell Preclassification Reclassification
abnormality agreement (%) agreement (%)
Polychromasia 76 76
Hypochromia 87 84
Microcytosis 85 84
Macrocytosis 42 85
Anisocytosis 51 74
Poikilocytosis 59 74
Preclassification is the number of correct suggestions by
the DM96 and Reclassification is agreement with the
manual method after the operator has altered the results
originally presented by the instrument. Two hundred
and eighty-six blood films were evaluated.
provide precise and accurate WBC differential counts.
However, when abnormal cells are present a manual
differential or examination of a stained blood film is
still required. Manual examination of blood films
requires highly trained personnel and is time con-
suming. Because of these difficulties in manual cell
counting automated image analysis systems have
been developed, but earlier systems were slow and
failed to identify abnormal cells (Tatsumi & Pierre,
2002).
We have evaluated a system capable of automated
analysis of peripheral blood films and present data
that shows that the system can generate WBC differ-
ential results comparable with a manual differential
performed by a laboratory scientist. 89.2% of cells
were classified correctly with only 0.9% of abnormal
cells misclassified as normal.
The DM96 can be loaded with blood films and the
operator can leave the instrument to perform other
duties, then on return be presented with the WBCs
on the computer screen. These can be reviewed,
quickly reclassified if necessary and the results
released into the laboratory information system. In
busy laboratories, such as ours, which processes 2000–
2500 samples a day, it would be necessary to have
two DM96 instruments for the optimum processing of
blood films. Using local decision rules for blood film
review, similar to those published by the International
Consensus Group for Hematology Review (Barnes
et al., 2005), but adapted for our patient population,
and restricting the number of times a film is looked at
 2007 The Authors
Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60
EVALUATION OF THE CELLAVISION DM96 57
on all haematology in-patients to twice per week
only, despite having daily full blood counts, our blood
film review rate is less than 10% (approximately 200
blood films per day). Final WBC differential results
from the DM96 (after reclassification by the operator)
were compared with the 100-cell manual differential
and for most cell types, including blast cells and
NRBCs, correlation was very good. Correlation for the
eosinophils is not as good as the manual method, this
may be a staining issue which could be modified, the
eosinophils appear pinker on the DM96 screen than
under the microscope.
For basophils and immature myeloid cells auto-
mated classification was different from the reference
method and this may be because of the small number
of cells being counted (Rumke, 1985a) or, for the
immature granulocytes, the misclassification of cells
into the earlier or later maturity class, e.g. myelocyte
instead of promyelocyte. This is demonstrated by the
excellent correlation (r2 = 0.95) between the reference
method and the DM96 when the total immature cells
were compared. The subjective classification of imma-
ture granulocytes into an earlier or later maturity class
is also a human failing.
The classification of red blood cells was less suc-
cessful with the instrument reporting macrocytosis
and anisocytosis on many samples which had normal
red blood cells. However, in 95% of samples the qual-
ity of the red cell and platelet images were sufficient
for the operator to clearly identify morphological fea-
tures and reclassify the results. In this evaluation, the
default red cell precharacterization settings were used
which may have contributed the percentage disagree-
ment; settings can be readjusted by the user to
achieve a level of classification equivalent to that by
manual microscopy.
The overall accuracy of the manual differential by
five different operators does not exceed that of the
DM96. This is demonstrated by the comparison of
each individual’s manual differential as well as their
pre- and reclassified differentials from the DM96 to
the 400-cell reference differential. Correlation to the
reference method is similar for an operator whether
the differential is performed manually or by the
DM96 with reclassification of cells. For some cell types
and for some operators, the accuracy of the DM96
compared with the reference method was better than
that of the individual’s 100-cell manual differential.
58 C. BRIGGS ET AL. EVALUATION OF THE CELLAVISION DM96

Table 3. Correlation coefficients (r2 values) for comparison of five different operators’ differentials to the 400-cell reference
differential

Ref vs. Ref vs. Ref vs. Ref vs. Ref vs. Ref vs. Ref vs. Ref v Ref v
Operator man reclass preclass Man reclass pre-class man Re-class Pre-class

Neutrophils Basophils Metamyelocytes

1 0.994 0.993 0.995 0.295 0.001 0.056 0.274 0.330 0.822
2 0.993 0.991 0.974 0.060 0.012 0.031 0.537 0.701 0.723
3 0.968 0.986 0.988 )0.053 )0.049 )0.152 0.845 0.238 0.758
4 0.843 0.994 0.989 0.023 0.009 0.063 0.789 0.928 0.914
5 0.990 0.990 0.990 0.089 )0.197 )0.284 0.863 0.728 0.956
Lymphocytes Blasts Myelocytes
1 0.897 0.752 0.218 0.998 0.989 0.804 0.804 0.670 0.707
2 0.788 0.841 0.240 0.999 0.998 0.999 0.763 0.712 0.963
3 0.752 0.764 0.324 0.996 0.998 0.982 0.264 0.927 0.831
4 0.442 0.077* 0.179 0.965 0.975 0.952 0.688 0.333 0.329
5 0.686 0.782 0.193 0.998 0.999 0.986 0.712 0.503 0.712
Monocytes Nucleated red blood cell Promyelocytes
1 0.624 0.663 0.677 0.804 0.978 0.973 0.309 0.000 0.690
2 0.674 0.768 0.707 0.995 0.991 0.973 0.948 0.702 0.936
3 0.752 0.540 0.823 0.861 0.960 0.956 0.227 0.423 0.643
4 0.848 0.822 0.805 0.921 0.953 0.915 0.018 0.600 0.706
5 0.521 0.805 0.724 0.992 0.962 0.956 0.551 0.540 0.600
Eosinophils Immature granulocytes
1 0.802 0.461 0.323 0.831 0.987 0.910
2 0.451 0.528 0.155 0.909 0.971 0.917
3 0.624 0.554 0.301 0.750 0.748 0.950
4 0.338 0.395 0.465 0.777 0.670 0.851
5 0.694 0.394 0.112 0.898 0.887 0.956

Ref vs. man, the operators’ manual differential compared with the reference differential; Ref vs. reclass, the differential
from the CellaVision DM96 after re-classification by the operator; Ref vs. preclass, the differential from the CellaVision
DM96 before reclassification by the operator.
*Operator did not reclassify blast cells in one sample that had been wrongly classified by the DM96 as lymphocytes, if
this one sample is excluded r2 = 0.61.

One sample contained a large number of blast cells The wrongly identified cells in the reclassified results
which the DM96 classified as lymphocytes, all opera- from this operator were clearly seen by recalling the
tors, apart from one, reclassified the cells correctly. differential from the DM96 database. An experienced
laboratory scientist and the individual reanalysed the
Table 4. Comparison of time taken to complete the 30 blood film on a cell-by-cell basis to re-educate the
differentials on the CellaVision DM96 including member of staff who had made the misclassification.
reclassification of cells with time taken to perform the Unlike a manual blood film it is exactly the same 100
same differentials manually cells seen the first time that are being reanalysed. The
ability to determine how an individual has classified
Time for analysis Time for manual
Operator on DM96 differential analysis cells makes training and competency testing much
simpler and more effective.
1 1 h 5 min 1 h 45 min CellaVision can provide specific software for profi-
2 1 h 10 min 1 h 40 min ciency testing and education in manual blood cell dif-
3 1 h 30 min 3 h 45 min ferentials. CellaVision Diff IQ has not been
4 1 h 40 min 4 h 10 min
evaluated in this study but should allow laboratories
5 1 h 14 min 3 h 10 min
to ensure that staff are trained to report results consis-
 2007 The Authors
Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60
 C. BRIGGS ET AL. EVALUATION OF THE CELLAVISION DM96 59

tent with laboratory procedures and helps to guaran-
tee the quality of the differential results. A digital test
case is set up by the examiner who also establishes
the correct cell classification. A member of staff per-
forms a WBC differential and red cell morphology
comments using the DM96 and the results are auto-
matically compared with those of the examiner and
other participants. There is automatic extraction of
cells possibly misidentified encouraging post-test com-
parison and discussion. The automatic storage of all
digital images from previous differentials is also useful
for allowing cell comparisons to previous encounters
on a particular patient.
The timing study clearly demonstrates that the
DM96 differential is faster than the manual differen-
tial this is consistent with earlier findings where on
average the manual differential took 1.3 min longer
to perform than when the DM96 was used (Kratz
et al., 2006).The time saved when using the DM96 is
greatest for the less experienced laboratory scientists.
With increased instrument familiarity with the instru-
ment there may be potential for even more time
saved.
The precision of the DM96 when analysing the
same slide five times is similar to that of the manual
differential performed on the same slide by five differ-
ent operators. This is unsurprising as the DM96 was
set to count 100-cells for equivalence with the man-
ual count. The error of these methods is associated
with the number of cells counted (Rumke, 1985b) as
well as cell distribution and interobserver variability.
The DM96 can be adjusted to count more than 100-
cells or more than one slide per sample can be used
with only a small increase in acquisition time. Statisti-
cally, the more cells counted the more precise the
results will be. The DM96 was reliable throughout the
3-month evaluation period with no breakdowns.
Other advantages of the DM96, outside this evalua-
tion, are the possibility of remote viewing of blood
cells. Software can be installed on any PC which
allows for verification of cell types from a different
location. A small satellite laboratory without morphol-
ogy expertise can send images to the central labora-
tory for classification and diagnosis.
So can automated blood film analysis replace the
manual differential?
We have demonstrated that the differential from
the DM96 is as good as that by a laboratory scientist;
however, the laboratory scientist operating the DM96
must be skilled in blood cell morphology. The instru-
ment gives results of comparable precision with that
of different individuals performing manual differen-
tials on the same sample. The speed of results from
the DM96 is impressive; quicker than all the labora-
tory scientists involved in the study, including those
with considerable morphological experience. The
DM96 is reliable and certainly has a place in the hae-
matology laboratory where it should improve work-
flow, make more efficient use of experienced
laboratory scientists’ time and make training and
monitoring of staff in blood cell morphology skills eas-
ier and more efficient.
University College Hospital London was in receipt
of an unrestricted educational grant at the time of this
study from Sysmex Europe GMBH.

REFERENCES
Barnes P.W., McFadden S.L., Machin S.J.
& Simson E. (2005) The international
consensus group for hematology
review: suggested criteria for action fol-
lowing automated CBC and WBC differ-
ential analysis. Laboratory Hematology
11, 83–90.
Bentley S.A. (1990) Automated differen-
tial white cell counts: a critical apprai-
sal. Bailliere’s Clinical Haematology 3,
851–869.
Clinical and Laboratory Standards Insti-
tute. (2007) Reference Leukocyte
(WBC) Differential Count (Proportional)
and Evaluation of Instrument Methods;
Approved Standard, 2nd edn. Vol 27,
No 4.
Koepke J.A., Dotson M.A. & Shifmann M.A.
(1985) A critical evaluation of the man-
ual/visual differential leukocyte counting
method. Blood Cells 11, 173–186.
Kratz A., Bengtsson H., Casey J.E., Keefe
J.M., Beatrice G.H., Grzbek D.Y., Lew-
androwski K.B. & Van Cott E.M. (2006)
Performance evaluation of the CellaVi-
sion DM96 system. American Journal of
Pathology 124, 770–781.
Lewis S.M. & Rowan M. (1991) Assess-
ment of the need for blood film exami-
nation with blood counts by aperture-
impedance systems. In: Standard Hae-
matology Practice (ed B. Roberts), pp.
34–42. Blackwell Scientific Publications,
Osney Mead, Oxford, UK.
Prewitt J.M. & Mendlesohn M.L. (1966)
The analysis of cell images. Annals of

 2007 The Authors

Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60
60 C. BRIGGS ET AL. EVALUATION OF THE CELLAVISION DM96

the New York Academy of Sciences
128, 1035–1053.
Rumke C.L. (1985a) Statistical reflections
on finding atypical cells. Blood Cells 11,
141–144.
Rumke C.L. (1985b) Imprecision of ratio-
derived differential leucocyte counts.
Blood Cells 11, 311–314, 315.
Tatsumi N. & Pierre R.V. (2002) Auto-
mated image processing. Past, present
and future of blood cell morphology
identification. Clinics in Laboratory
Medicine 22, 299–315.

 2007 The Authors

Journal compilation  2007 Blackwell Publishing Ltd, Int. Jnl. Lab. Hem. 2009, 31, 48–60