Hybrid Microcalcification Detection In Mammograms

This paper presents a combined system for detecting microcalcifications in mammographic images. The proposed approach analyses the mammograms using the parallel combination of two methods. One of the methods is a heuristic approach that is based on features used by humans to descript microcalcifications. Second approach applies a hierarchical neural structure trained on images already interpreted by human experts. The results of the two methods are binary masks of the microcalcifications. These results are integrated on pixel level, currently the 'AND' relation of the binary masks are used. Tests show that the combined system could decrease false positive ratio significantly with a modest increase of false negative cases.