Segmentation and classification of various types of cells in cytological images

The authors are concerned with the segmentation of cytological images in order to automatically classify cellular categories identified thanks to their nucleus. They deal with microscopic images of human oesophagus cells, that have been dissociated and stained according to Feulgen and Rossenbeck methods (1924). These images come from the pathology department of the cancer-research centre F. Baclesse of Caen. This study comes within the framework of the optimisation of an image analyzer dedicated to rapid detection of abnormalities of ploidy in human tumors. In this application, the purpose of segmentation is to extract cellular objects to distinguish isolated cells from clusters and to calculate shape and texture measures to identify each kind of cell. By using knowledge about these cytological images, the authors describe the results obtained in a preliminary study of segmentation and region classification into isolated cells and clusters. First, they study the problem of segmentation of cytological images, and lay the emphasis upon the knowledge involved in this kind of image and its exploitation in the selection process of segmentation operators. Then they give a few segmentation methods based on thresholding before describing in detail the segmentation operators selected in order to solve this problem.<
>