Klinisk Biokemi i Norden Nr 4, vol. 18, 2006 - page 12

12
| 4 | 2006
Klinisk Biokemi i Norden
(Fortsat fra side 11)
them pathological. As defined in the NCCLS-H20A
protocol, 400 cells were counted, 200 cells each by
two different technologists on two different slides.
For the evaluation of the DiffMaster system , these
slides were processed according to the standard
instrument procedure. Instrument pre-classifica-
tion was undertaken and then manual validation
was performed. Generally the results show there
was a good correlation between the two methods
with concordance of about 89 %. For the 6 normal
cell types and blasts cells r
2
was 0.8 – 0.98 while
concordance was lower for pathological cell classes
(55 %). The correlation of physiologically occurring
cell types is excellent as is shown for lymphocytes
in figure 4. Thus, the DiffMaster is good for imag-
ing cells from the blood smear as well as presenting
them to the microscopist.
But how accurate is the neural network classifica-
tion of different cell populations compared to human
classification? This question can be addressed by
comparing the uncorrected values obtained from
DiffMaster with the corresponding values from
manual reclassification. The correlation between
the differential counts obtained with or without
manual reclassification was between 0.72 – 0.98 for
the major cell classes (Table 1) and showed that the
DiffMaster preclassification is clinically usable even
without human reclassification. The cell class with
largest differences between obtained with/without
manual reclassification were the nucleated re blood
cells (NRBC) but the number of observations was
low. The classification of lymphocytes and mono-
cytes also showed some discrepancy between com-
puter and human classification. The classification
of these cells may indeed be difficult and does also
vary between trained morphologists (8-10).
An advantage of computerised classification is
the presentation of unidentified as well as smudge
cells. The proportion of cells in these new classes
provide two important quality parameters that tell
the validity of the classification in a particular sam-
ple. This validity of the DiffMaster preclassification
is presently being further investigated in Malmö.
Since the differential leukocyte count is qualitative,
it is desirable to know the sensitivity of the method.
Does the digital differential count identify the sam-
ples that would be classified as pathological by the
traditional reference method? This was shown to be
the case. The clinical sensitivity was 98 %. Only 2%
Figure 4: Comparison of uncorrected DiffMaster values
for neutrophil granulocyte classification and corrected
values after human reclassification.
Figure 5: Comparison of uncorrected DiffMaster values
for lymphocyte classification and corrected values after
human reclassification.
Figure 6: Comparison of uncorrected DiffMaster values
for monocyte classification and corrected values after
human reclassification.
1...,2,3,4,5,6,7,8,9,10,11 13,14,15,16,17,18,19,20,21,22,...44
Powered by FlippingBook