Leukocyte Detection Using Nucleus Contour Propagation

We propose a new technique for medical image segmentation, focused on front propagation in blood smear images to fully automate leukocyte detection. The current approach also incorporates contextual information, which it is especially important in direct general algorithms to the applied problem. A Bayesian classification of pixels is used to estimate cytoplasm color and is embedded in the speed function to accomplish cytoplasm boundary estimation. We report encouraging results, with evaluations considering difficult situations as cell adjacency and filamentous cytoplasmic projections.