A commercially available interactive pattern recognition system for the characterization of blood cells: description of the system, extraction and evaluation of simple geometrical parameters of normal white cells

Abstract

Summary A programmable system, the Textur Analyse System (T. A. S.) of E. Leitz, is described for use in interactive work on pattern recognition of white blood cells. The system appears well suited for the task of finding new parameters for the characterization of normal and abnormal blood cells. Hardware advantages such as speed of operation are coupled with software flexibility. The first application of the machine has been the extraction of some of the ordinary parameters for characterization of leucocytes. The value of each parameter has been analysed with the interactive statistical pattern analysis program (ISPAHAN). A separation in the five normal classes of peripheral white blood cells can be achieved, in which the nuclear/cell area ratio and nuclear area together with the density histograms proved to be the most important parameters. The interesting feature of the system is, however, the possibility of finding new data for the recognition of normal and abnormal blood cells.

Relevance of morphometric parameters for the classification of normal white blood cells*

“…In spite of these differences nearly the same classification rate was obtained when only geometrical parameters were used. The shape of the densitogram was thought to be characteristic for each cell class in the previous study, and was not affected by variations in staining (Smit et al 1979). Indeed parameters describing its shape were highly successful (92.6%) in classification.…”

“…In a previous study (Smit et al 1979) only geometrical parameters were evaluated but similar results were obtained in this study, although considerable differences existed in the measuring methods; in the present study the grey level used for thresholding to separate nucleus and cytoplasm was set automatically by the TAS, while in the previous study this grey level was set interactively; in the present study granules in the cytoplasm and vacuoles in the nucleus were corrected for, while in the previous study no correction was made; and finally in this study virtually no classifiable cells were excluded from analysis, whereas in the previous study no atypical cells were used. In spite of these differences nearly the same classification rate was obtained when only geometrical parameters were used.…”

“…Automated morphometry may be of even greater importance in the examination of bone marrow smears, where morphological differences between cell classes can be very subtle, making classification a difficult matter even for the trained haematologist. This paper describes the use of a Leitz Texture Analysing System (TAS) (Smit et al 1979) with interactive pattern recognition (ISPAHAN) (Gelsema 1980) of geometricai features, densitogram features and the numbers of holes and particles at different grey levels in the cellular images.…”