Application of Structural Pattern Recognition in Histopathology

Abstract

Structural pattern recognition in histopathology can be used for a) Assistance in difficult diagnoses b) Measurements of interactions between different cell populations c) Estimation of proliferation activity of cancerous tissue in relation to survival of the patients. The following system based upon interactive measurements was developed: Images of HE-stained and immuno-stained histopathological specimens were projected onto a graphic pad connected to a 4051 TEKTRONIX-computer. Coordinates of interesting structures (glands, epithelial cancerous cells, positively immuno-stained cells differentiating cancerous subpopulations) were marked interactively. Centers of interesting structures were considered as vertices, neighboring structures as edges. A modified neighborhood condition based upon O’CALLAGHAN’s definition was used. Measurements were performed at low and high microscopic magnification. The following parameters were measured: Number of neighbors, cyclomatic number, n-simplices, n-stars, distance between neighboring cells. Results were analyzed by non-hierarchic discriminant analysis (test on cohesion andcentroids). The following results were obtained: a) Measurements of healthy mucosa, tubulo-villous adenoma and highly to moderately differentiated adenocarcinoma of colon could be correctly separated and regrouped in 83% of the cases. 11/15 cases (73%) could be classified correctly in a prospective group. Similar percentage of correct separation and reclassification was obtained in a teaching set of 20 cases and in a training set of 18 cases with metastatic adenocarcinoma of pleura of epithelial-biphasic mesothelioma. b) Differences of tumour cell clones detectable by immunohistology were related to geometrical distance of positively stained tumour cells. Difference of nearest neighboring cells between negatively stained cells was undistinguishable for different antibodies opposite to significant differences in distance of positively stained nearest neighboring cells. The data indicate clonal origin of cells reacting positively to the applied antibodies. c) Minimum distance of nearest neighboring cells was measured in order to determine proliferation activity of tumour cells. Survival of 60 patients with small cell anaplastic carcinoma of the lung showed close relation to distance of nearest neighboring cells (p<0.05).