Klinisk Biokemi i Norden Nr 4, vol. 18, 2006 - page 8

| 4 | 2006
Klinisk Biokemi i Norden
(Fortsætter side 10)
Computerised Pattern Recognition Systems in Routine
Haematology
Per Simonsson
1
, Sven Björnsson
1
, Lennart Friis-Hansen
2
, Birgitta Swolin
3
E-post: per.simonsson@med.lu.se
1
Department of Clinical Chemistry, Malmö University
Hospital, Malmö, Sweden,
2
Department of Clinical
Biochemistry, Rigshospitalet, Copenhagen University
Hospital Denmark,
3
Department of Clinical Chemistry
and Transfusion Medicine, Sahlgrenska University
Hospital, Göteborg, Sweden
Abstract
Artificial neural network is a computerised tool
for pattern recognition that can be used in the
classification of cells in clinical haematology. We
have evaluated DiffMaster Octavia, an automated
image analysis system, in routine practice at three
different Scandinavian university laboratories. The
concept offers analytical quality often compatible
with that of a trained morphologist, at least when
classifying normal peripheral blood cells. However,
a validation based on a review by a medical tech-
nologist is necessary, in particular when investigat-
ing pathological samples. The digital system offers
additional advantages compared to traditional mor-
phology such as increased efficiency, quality assur-
ance, standardisation of interpretation, training,
image storage and transferability.
Introduction
Success in laboratory medicine requires that clinical
work interface with biotechnology and information
technology to produce a valuable clinical service.
Traditionally laboratory haematology has been very
strong in clinical relations. Biotechnology is devel-
oping apace as is information technology. The real
and important challenge is to link all three parts
together optimally.
Another of tomorrow’s challenges to the clinical
biochemists will be to meet the specific requirement
of the user to get results in time. Until now the
service from the clinical biochemical laboratories
has been very robust but rigid.
Haematology will be shaped by the progress in
science and technology, both of which are advanc-
ing rapidly. Immunophenotyping by flow cytometric
technique is developing very strongly and molecular
biology methods are also a rapidly developing diag-
nostic tool. Automated cell counters have revolution-
ised the workflow of the haematological laboratory,
and the routine cell counters have new features that
increase their capacity to characterize cell classes.
Recent advances in automated cell counting technol-
ogy open up for 6 or 7 part differential counts. The
additional groups quantitate nuclear red blood cells
(NRBCs) and immature leucocytes and are thus added
to the normal differential counts. Also, in many cases
knowledge of the exact numbers of each class of
immature cells is not needed. This also suggests that
an intelligent use of the flags could be possible. Thus
some results could be released while others are sup-
pressed until they can be validated. Potentially the
same type of algorithms could also be made to work
for the digital microscope thereby ensuring release of
auto-validated results.
In our departments results from about 80 % of
differentials can be released after counting with
high quality and great reproducibility in modern
cell counters, the remaining 20% needing further
examination by microscopy. That means we have
only the difficult samples left for microscopy. One
consequence of this is that there is loss of microsco-
py expertise, particularly when combined with one
staff generation retiring and another very differ-
ently educated generation coming into the labora-
tory. In many cities like Copenhagen and Göteborg
the treatment of patients with haematological
diseases has been centralized, and this has reduced
1,2,3,4,5,6,7 9,10,11,12,13,14,15,16,17,18,...44
Powered by FlippingBook