High resolution image analysis of cell nuclei in tissue sections of primary and metastatic carcinomas

The present study examines whether certain histological tumour types can be differentiated on account of their nuclear image with the aid of automated image analysis. For karyometric investigations three tumour types (adenocarcinomas, squamous cell carcinomas and mammary carcinomas) were chosen, which occur frequently as occult primary tumours. From each type ten primary tumours with their corresponding lymph node metastases were examined. 100 cell nuclei were measured from each case using 4 micron thick paraffin sections stained with gallocyanin-chromalum. For each cell nucleus 21 contour and texture features were determined. Through the application of linear classifiers 41 out of 52 cases (25 primary tumours, 27 metastases) of these three tumour types were correctly classified. Eight cases could not be classified with certainty and only three cases were wrongly classified. In addition, within the group of adenocarcinomas differences due to localisation were detected which allow us to draw conclusions on the seat of the primary tumour.