A Water Immersion Algorithm for Cytological Image Segmentation

Abstract

A non-parametric multi-resolution method is presented for cell segmentation and some of the advantages it has over other techniques, when applied to this application, are outlined. The proposed algorithm is an extension on the method of Wilson and Spann [6] and uses a technique based on the concept of water immersion which ensures closed boundaries. Experiments with cervical cells have shown that the algorithm is robust, fast and accurate. 1 Introduction The ratio of the area of the cytoplasm to the area of the nucleus (CN) is one of the best discriminatory features for the classification of cells in the screening for cancer of the cervix. The larger the ratio, the more likely the cell is to be dysplastic (pre-cancerous) or even cancerous. This ratio naturally depends upon an accurate segmentation of the cytoplasm from the background and of the nucleus from the cytoplasm. Unfortunately, cytological images present difficult segmentation problems such as overlapping cells, clumped group...

Ähnliche Arbeiten