Professional Documents
Culture Documents
WBC Differentials by Automated Digital Image Analysis Supported by An Artificial Neural Network
WBC Differentials by Automated Digital Image Analysis Supported by An Artificial Neural Network
DOI: 10.1309/XMB9K0J41LHLATAY
devices have been developed (recently reviewed by Tatsumi estimation of the platelet count. Advantages of this approach
and Pierre2). Capitalizing on advances in digital photography over slide review with a microscope include possible labor
and data processing and storage, these devices have made it savings owing to the localization and preclassification of the
possible to envision a “virtual blood film” that can be WBCs by the instrument, more reproducible results, and the
reviewed, shared with experts in distant locations, stored, ability to review cases from a remote location or to rereview
retrieved, and rereviewed at low cost.4,5 cells at a later date.
The latest such automated image analysis system to Direct microscopic review of blood smears is a well-estab-
become commercially available is the CellaVision DM96 lished procedure; clinicians and laboratory workers have exten-
(CellaVision AB, Lund, Sweden; distributed in the United sive experience with this method, are comfortable with it, and
States by Sysmex America, Mundelein, IL). The successor have confidence in the results obtained. Introduction of an instru-
model to the DiffMaster Octavia (CellaVision AB),6 the ment for automated image analysis, such as the CellaVision
CellaVision DM96 is an automated image analysis system for DM96, into the clinical laboratory will necessitate a change in
Hematology System and flagged for microscopic review. The performance of the CellaVision DM96. These cases included
use of cases prescreened by an automated cell counter was slides with increased numbers of immature WBCs and the
intended to mimic the standard workflow in a hematology lab- presence of abnormal WBCs and some cases with very low
oratory, where usually only cases flagged by the automated WBC counts. All slides had required a full microscopic differ-
cell counter are subject to slide review. Wedge blood smears ential count.
were prepared manually with the push-pull method with a Three laboratory employees (2 certified laboratory tech-
spreader slide or with a semiautomated method, the Miniprep nologists [J.E.C. and J.M.K.] and 1 certified medical laboratory
blood smearing instrument (Geometric Data, Wayne, PA), air technician [G.H.B.]) were asked to participate in this part of the
dried, and stained with a Sakura Stainer RSG-61 (Sakura study; they were selected for representing the range of expertise
Fineteck USA, Torrance, CA) with a Wright-Giemsa stain.7 in the performance of manual differentials found in the labora-
tory. Expertise of employees was determined by their level of
Manual Differential Counts (Reference Method) formal training (certified technologists vs certified technicians),
A B
100% 100%
90% 90%
80% 80%
70% 70%
60% 60%
50% 50%
40% 40%
30% 30%
20% 20%
10% 10%
0% 0%
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 on direct microscopic review. A repeated standard microscopic
3 cases in which the discrepancy could not be resolved by statis- differential also did not show these immature myeloid cells on
tical variation, differences in nomenclature, or misclassification the slides. When a third direct microscopic differential based on
of digital images presented by the CellaVision DM96; in these up to 300 cells was performed, populations of immature myeloid
cases, there were clearly cells consistent with immature myeloid cells (between 2.3% and 8%) were identified.
cells (promyelocytes, metamyelocytes, and/or myelocytes) pres- As noted in the preceding paragraph, most of the discrepan-
ent on the CellaVision DM96 images that had not been reported cies were insignificant statistically (within the 95% confidence
❚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 (%)
❚Table 5❚
Rereview of Cases With Positive Findings by the CellaVision DM96 and Negative Findings by Direct Microscopy (False-Positives
on the CellaVision DM96)*
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
❚Table 6❚
Rereview of Cases With Negative Findings by the CellaVision DM96 and Positive Findings by Direct Microscopy (False-Negatives
on the CellaVision DM96)*
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
(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 (%)
❚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)
Very experienced technologist with special expertise in WBC differentials (n = 76) 1.4 1.5
Technician with no special expertise in WBC differentials (n = 160) 4.1 3.0
slide review by technologists ranged from 34% to 100%, To evaluate the impact of the CellaVision DM96 on labor
depending on the abnormality. Review of the discrepant cases costs, we performed 2 studies. First, we timed 6 technologists
in our study showed that the majority of discrepant cases were performing 10 fully manual differentials with the microscope
due to variation that was not statistically significant (within and with the CellaVision DM96. It took the technologists, on
the 95% confidence interval of the number of cells on which average, 5.1 minutes per slide to perform a manual differential;
the differential was based) or to clinically insignificant varia- the time required with the CellaVision DM96 was longer, 6.4
tions in nomenclature (eg, blasts vs other cells, query blasts). minutes. The limitations of this study are that the technologists
When we corrected for these statistically or clinically insignif- had years of experience in performing differentials with a
icant discrepancies, the sensitivity of the CellaVision DM96 microscope; in contrast, the CellaVision DM96 was new to
was between 95% and 100% and the specificity between 88% them, and they lacked any prolonged experience on this instru-
and 97%, depending on the abnormality. ment. It therefore is possible that with increased exposure to the
The remaining discrepancies were mostly (16/19 dis- CellaVision DM96, the average time per manual differential
might be offset partly by the costs of the instrument, its asso- * Ms. Casey received an honorarium from Sysmex for a