Nerve Cell Segmentation via Multi-Scale Gradient Watershed Hierarchies

Automated segmentation of nerve cell in microscopic image is an important task in neural researches. We proposed a multi-scale watershed-based approach to cope with this microscopic image analysis problem. There are three stages in the proposed segmentation algorithm: (1) a multi-scale watershed scheme is used to estimate an initial location of nerve cell nuclei; (2) we can identity nerve cell nuclei according to properties of nerve cell and watershed results in different scale; (3) Once the possible nerve cell is identified as a true one, a fuzzy rule based active contour model (ACM) is applied to find the optimal outer contour. Our approach can segment nerve cell automatically and accurately. The cell detection rates in the experiments are above 95%. Moreover, the fuzzy rule based ACM provides flexible alternative to handle cell contour detection