Estimated platelet and differential leucocyte counts by microscopy, Sysmex XE-2100 and CellaVision[TM] DM96.Introduction
The increased workload experienced by many haematology laboratories coupled with reductions in staff numbers has lead to the introduction of automated morphological analysis in laboratories. The improved processing capacity of computers combined with high resolution digital imaging technology has allowed for the development of systems which can pre-classify haematological cells, mostly leucocytes (1). The CellaVision[TM] DM96 is an imaging device intoduced into Canterbury Health Laboratories (CHL) in 2007. This paper reports on the performance of the DM96 to determine the leucocyte differential and compared the results with both manual differentials and the differential result from the Sysmex XE-2100 automated cell counter. In addition, the DM96 derived platelet estimate was performed and compared to the standard microscopic platelet estimate and the platelet count from the Sysmex XE-2100.
Materials and methods
The study utilised the Sysmex XE-2100 haematology analyser and the Sysmex SP1000i slidemaker/autostainer. The CellaVision DM96 system was used for the automated leucocyte differential counting and platelet count estimates. These analysers are used in the core laboratory at CHL and provide CBC testing of samples submitted from inpatients and outpatients of the public hospital in Christchurch. Standard light microscopes were used for microscopy.
Samples were selected from the routine work which arrived in the haematology laboratory during a single 24-hour period. Flagging functions on the Sysmex XE-2100 indicated which samples required blood film examination and films were prepared and stained by the Sysmex SP1000i slidemaker/autostainer. Fifty slides were selected for the project by the author's (AP) supervisors from the results of the CBC values. Samples were selected to include both normals and abnormals containing qualitative and/or quantitative abnormalities in the main cell classes. Slides from patients with known acute leukaemia were excluded. The slides were anonymised to show only the laboratory identification numbers. Of the fifty slides four samples were excluded from platelet analysis due to the presence of platelet clumps observed during microscopy. Three samples were also excluded from the differential count analysis as the Sysmex was not able to count suffficient cells to produce a valid differential.
Each slide was analysed by the DM96 with 100 cells counted. Cells were pre-classified into: band neutrophils, segmented neutrophils, lymphocytes, monocytes, eosinophils, basophils, metamyelocytes, myelocytes, promyelocytes, blast cells, lymphocyte variant forms and plasma cells. For data collection purposes, band and segmented neutrophils were counted together. Lymphocytes, lymphocyte variants and plasma cells were counted as one group. Precursor cells from myeloid blast cells through to metamyelocytes were counted together as immature granulocytes. Unidentified cells were categorised as 'other'. Results were stored as the pre-classification DM96 data for each slide and were then viewed. Incorrectly classified cells were moved to their correct classification based on their morphological appearance. Cells classified as 'other' were similarly reclassified where possible and moved intoone of the leucocyte classification groups. Non-leucocyte classifications were also examined and the cells incorrectly identified as nucleated red cells, giant platelets, platelet aggregates, smudge cells or arte facts were reclassified. Results were recorded as the reclassified DM96 data.
For each slide a manual 100-cell differential was performed using a light microscope at 500x magnification. Cells were recorded in the same categories as for the reclassified DM96 data.
The platelet count estimation facility on the DM96 was used to derive an estimate of platelet numbers on each of the stained blood films. Platelet numbers present in 16 sub-images were counted, and the average value for those fields multiplied by the platelet estimation factor (pre-set at 10) provided the estimated platelet result. For the DM96, the platelet estimation corresponded to counting the equivalent of eight high power microscopy fields as per the CellaVision Users Manual (4). Results were recorded as the DM96 platelet estimation.
Manual platelet estimates were performed by counting the number of platelets in 10 successive high power fields at 1000x magnification and multiplying the count by 109 to give the platelet estimation as n/[10.sup.9]/L. These results were recorded as the manual platelet estimation.
Following collation of manual and DM96 results for both the leucocyte differentials and the platelet estimates, patient data from the laboratory information system was accessed and the verified data for the Sysmex differential and platelet count were obtained. Correlation data was derived using Bland-Altman plots and regression analysis.
The platelet estimates from the DM96 and the visual method followed closely the platelet counts generated by the Sysmex XE-2100 (Table 1 and Figure 1). Of interest was the tendency for the DM96 to over estimate the platelet count as compared with the manual method. In comparison with the Sysmex results, the Rz values were 0.91 and 0.93 for the DM96 and visual platelet estimation methods, respectively (Figure 2).
The average ratio of Sysmex to DM96 platelet counts was 0.86 with the DM96 counts an overestimate in 80% of samples. In absolute values the average platelet overestimate was about 66 platelets. The average ratio of Sysmex to visual platelet counts was 0.93 with the visual result either over- or underestimating the Sysmex result by about 40 platelets.
The correlation between the Sysmex data and either reclassified DM96 or visual results for cell lines are summarised by the [R.sup.2] values in Table 2. The correlation data presented for the leucocyte counts in Table 2 was derived from the data presented in figures 3-14. Both DM96 and manual methods showed a strong correlation with the Sysmex data for the neutrophil counts, and an acceptable correlation was observed for lymphocyte and eosi nophi I counts. The correlations for monocytes, basophils and immature granulocytes were weak. Table 2 also shows the [R.sup.2] values for the pre-classification data. The pre-classification data was correct in 69% of cases. In most cases the results paralleled the reclassified results, except for the monocyte population, where the pre-classification count correlated strongly with the Sysmex data.
The percentage agreement with Sysmex data for each cell class according to test method is presented in Figure 15. The results from each method were divided into the proportion of test results which were in agreement with the Sysmex data, and those results which were not. Results which were not in agreement were further divided according to whether the result was an over- or underestimate of the Sysmex value.
All test methods identified neutrophils with the highest percentage agreement, followed by lymphocytes. Cell classes with smaller populations represented in the blood generally showed a lower percentage agreement.
Figure 16 shows the proportion of each cell class in an average differential for each test method. On average, all test methods had higher neutrophil, and lower lymphocyte and monocyte counts compared to the Sysmex. All methods were approximately equal in their estimates of smaller cell classes. Overall, values obtained by the DM96 and manual methods were almost identical.
Both DM96 and visual platelet estimates correlated well with the Sysmex XE-2100 platelet counts, however, there was a consistent overestimation of the DM96 platelet count. The overestimation could be accounted for by the platelet estimation factor used on the machine, which set at 10 appeared to be too high. A change of the estimation factor to 0.86 would yield a platelet estimation very close to the Sysmex platelet count making this method more accurate than the manual method. Forty-seven samples were analysed in this study which was more than the 30 sample minimum recommended by the manufacturer in the setup of the platelet estimate function (3). The slides used in the study included 3 samples with platelet counts below 100 x [10.sup.9]/L and 5 above 450 x [10.sup.9]/L. Given these small numbers it may be worthwhile repeating the estimated platelet count on larger numbers of samples containing platelet numbers above and below the normal range.
While both the platelet estimate methods were time consuming, the DM96 method took longerto scan slides to estimate the platelet numbers in the film. An advantage of the DM96 method was the overlay grid which aided counting, but the major disadvantage was poor image quality. In our hands the imaging was sometimes out of focus. With traditional microscopy a scratched section could be bypassed, but the DM96 has no 'reserve' images for such a situation.
While the platelet estimate function of the DM96 could at times be useful in the laboratory, platelet estimates are rarely required. The DM96 is capable of producing accurate platelet estimates once properly setup, but its use for screening all slides probably cannot be justified given the extra time taken.
In this study the identification and counting of neutrophils and lymphocytes by manual and DM96 methods showed good correlation with the Sysmex data although the results as mentioned showed both over- and under-estimation by both techniques. The smaller cell populations (eosinophils, basophils and blast cells) showed less variation in the results between the the manual and DM96 methods, however, these did not correlate well with the Sysmex data, the eosinophil population being the exception. The DM96 to Sysmex correlation was particularly poor for the monocyte population, and immature granulocytes were not well identified by the manual method when compared to the Sysmex values.
The pre-classification data correlated well with the reclassification values in most cases except for the monocyte population, which had a high pre-classification Rz value. This resulted from other cells (often smudge cells) being incorrectly classified as monocytes. In a previous study by Briggs et al the pre-classification monocyte count as compared to the manual method resulted in a low correlation (2). Monocytes are generally poorly classified by both automated and manual differential systems, mainly due to smear method and the area of the film examined (4). The relatively low number of monocytes in most samples is also likely to contribute to this error.
Sample size is an important factor to consider when assessing the relevance of results. In this study 50 samples were initially selected, of which 46 were used for the platelet estimates and 47 for leucocyte analyses. Three other studies compared the DM96 and manual differential, analysing 136, 322 and 400 samples respectively, and performed either 200 or 400-cell differentials (2,4,5). The manual 100 cell differential is the standard method used in laboratories today. This method lacks both accuracy and precision which can be improved on by counting 200-400 cells. This was not performed in this study. Additionally, the comparison of the manual differential results to the Sysmex XE-2100 differential, which counts many thousands of cells, highlighted the manual differential inaccuracies. The DM96 and manual methods counted the same number of cells and neither had any statistical advantage over the other.
One recurrent source of error was the total number of cells present in the reclassification differential. As only 100 cells were counted and classified by the DM96 in our study, when reclassification took place the differential was more or less than 100 cells in half of the samples. Where a 'reciprocal reclassification' took place (eg. a neutrophil classified as an eosinophil as well as an eosinophil classified as a neutrophil), the differential remained 100. However, when cells were classified as non-leucocytes or vice versa, the differential was higher or lower than 100 cells after the reclassification. Other studies have set the automated microscope to count 105, 110 or 200 cells to take this problem into account (2,4,5). In our experience setting the count value to 110 should be adequate for most samples.
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
[FIGURE 3 OMITTED]
[FIGURE 4 OMITTED]
[FIGURE 5 OMITTED]
[FIGURE 6 OMITTED]
[FIGURE 7 OMITTED]
[FIGURE 8 OMITTED]
[FIGURE 9 OMITTED]
[FIGURE 10 OMITTED]
[FIGURE 11 OMITTED]
[FIGURE 12 OMITTED]
[FIGURE 13 OMITTED]
[FIGURE 14 OMITTED]
[FIGURE 15 OMITTED]
[FIGURE 16 OMITTED]
It is also worth remembering that the DM96 only examines the area of the slide where it can detect an erythrocyte monolayer, thus if a sample does not contain many leucocytes in that area, the differential value may be provided on less than 100 WBCs. In this respect traditional microscopy has the advantage in that an area with overlapping erythrocytes can still be examined and leucocytes accurately identified.
Another important factor to take into account when examining correlation data, is the experience of the morphologists examining the blood films. Logically it would be expected that more experience would correlate with greater accuracy of identification, and this has been shown to be the case in other studies (2).
In conclusion, this study has shown that the results from a 6-part differential (including immature granulocytes) performed by the DM96 is similar to the results obtained by a manual differential. Neutrophil, lymphocyte and eosinophil values can be expected to correlate well with results obtained from the analyser differential using the Sysmex XE-2100, however monocyte, basophil and immature granulocyte numbers can be expected to correlate less well. This work supports the findings of others that the DM96 is particularly suited to laboratories (eg. community laboratories) processing large numbers of normal samples (2). More complex samples from patients with haematological malignancies, recent bone marrow transplantation and morphological changes associated with some infections, require the attention of a morphologist to accurately identify and classify immature cells. In the setting of a larger laboratory, such as that at Canterbury Health Laboratories, the DM96 can reduce the number of 'normal' films examined by morphologists, and can be used in conjunction with microscopy to investigate abnormal cases. The results of this study have demonstrated that microscopy and the trained morphologist still occupy an important place in the haematology laboratory. For the moment at least the human eye continues to occupy the high ground ahead of the modern mechanical morphologist.
This work was part of the requirement for the Massey University BMLSc degree and was conducted during the 4th year clinical laboratory placement in Haematology at Canterbury Health Laboratories in 2008. Many thanks to the haematology staff at Canterbury Health Laboratories, particularly Mr. Ken Beechey, Mrs. Linda Henshaw, and Mrs. Christine Harper for their assistance with this project.
(1.) Hutchinson CV, Brereton ML, Burthem J. Digital imaging of haematological morphology. Clin Lab Haematol 2005; 27: 357-62.
(2.) Briggs C, Longair I, Slavik M, Thwaite K, Mills R, Thavaraja V, et al. Can automated blood film analysis replace the manual differential? An evaluation of the CellaVision DM96 automated image analysis system. IntJ Lab Hematol 2009; 31: 48-60.
(3.) CellaVision[TM] DM96 User's Manual. CeIlaVision A.B. 2006.
(4.) Cornet E, Perol JP, Troussard X. Performance evaluation and relevance of the CellaVision TM DM96 system in routine analysis and in patients with malignant hematological disease. International Journal of Hematology. IntJ Lab Hematol 2008; 30: 536-42.
(5.) Swolin B, Simonsson P, Backman S, Lofqvist I, Bredin I, Johnsson M. Differential counting of blood leukocytes using automated microscopy and a decision support system based on artificial neural networks--evaluation of Diff Master Ocavia. Int J lab Hematol 2003; 25: 139-47.
Anthea Povall , BMLSc, Medical Laboratory Scientist
Christopher John Kendrick , Grad Dip Sci, MSc, MNZIMLS, Senior Lecturer
 Haematology Department, Canterbury Health Laboratories, Christchurch
 Massey University, Palmerston North
Address for correspondence: Anthea Povell, Haematology, Canterbury Health Laboratories, PO Box 151, Christchurch.
Table 1. Platelet count raw data (x 10 9/L) for slides 1-50. Slide Platelets No Sysmex DM96 Manual 1 358 450 354 2 173 264 246 3 171 151 147 4 254 194 233 5 59 66 64 6 38 51 30 7 165 171 180 8 197 259 207 9 594 663 545 10 142 150 152 11 291 345 319 12 241 229 201 13 128 186 174 14 225 283 295 15 222 355 244 16 222 288 250 17 357 496 424 18 229 324 275 19 480 493 441 20 197 270 222 21 158 208 231 22 371 543 490 23 367 449 382 24 362 539 413 25 338 420 419 26 255 303 260 27 889 1105 974 28 190 264 200 29 236 280 211 30 165 203 221 31 232 295 247 32 495 483 504 33 159 158 158 34 514 624 594 35 341 510 423 36 32 43 40 37 175 234 173 38 449 556 602 39 333 388 355 40 445 484 437 41 251 225 262 42 345 305 285 43 354 504 481 44 350 326 346 45 105 79 96 46 360 391 424 * Four samples were clotted and were excluded from all platelet analyses. Table 2. [R.sup.2] values for Sysmex vs DM96 and manual differentials for each cell class. DM96 Cell type Pre-classification Reclassified Manual Neutrophils 0.91 0.92 0.93 Lymphocytes 0.85 0.84 0.77 Monocytes 0.70 0.17 0.45 Eosinophils 0.69 0.71 0.81 Basophils 0.12 0.12 0.24 Immature grans 0.44 0.48 0.05