Maximum Likelihood Active Contours Specialized for Mammography Segmentation

We present a region-based active contour approach to segmenting masses in digital mammograms. The algorithm developed in a Maximum Likelihood approach is based on the calculation of the statistics of the inner and the outer regions (defined by the contour). The Poisson distribution that has been deemed in the past adequate for modeling mammograms is applied as the probability density function. The Poisson distribution parameters are assumed unknown and are also estimated by the algorithm. We evaluate the performance of the algorithm on real mammographic images, given from the digital database for screening mammography (DDSM). The quantitative validation results demonstrate an average segmentation accuracy of 81% for 100 test images using the presented method.