Automated cell nucleus segmentation using improved snake

Accurate cell nucleus segmentation is crucial for the development of automated cytological cancer diagnosis system. This paper presents a novel cell nucleus segmentation method for esophageal cell image. Firstly the ultimate erosion is used to detect the localizations of nuclei. Then we use an improved active contour model to isolate each cell nucleus. A growing energy based on region similarity is added to the energy function to overcome the initialization problem of conventional snake. During energy minimization, the contour points arc restricted to move along the radial directions, which reduces the computation cost. The presented method has been tested on a number of cell images obtained from esophageal smear slide and the results are encouraging. It has been shown that the proposed method performs well on both well-separated nuclei and some overlapped nuclei.