Fast Statistical Level Sets Image Segmentation for Biomedical Applications

In medical microscopy, image analysis offers to pathologist a modern tool, which can be applied to several problems in cancerology: quantification of DNA content, quantification of immunostaining, nuclear mitosis counting, characterization of tumor tissue architecture. However, these problems need an accurate and automatic segmentation. In most cases, the segmentation is concerned with the extraction of cell nuclei or cell clusters. In this paper, we address the problem of the fully automatic segmentation of grey level intensity or color images from medical microscopy. An automatic segmentation method combining fuzzy clustering and multiple active contour models is presented. Automatic and fast initialization algorithm based on fuzzy clustering and morphological tools are used to robustly identify and classify all possible seed regions in the color image. These seeds are propagated outward simultaneously to refine contours of all objects. A fast level set formulation is used to model the multiple contour evolution. Our method is illustrated through two representative problems in cytology and histology.