Automatic mass detection in mammograms using multiscale spatial weber local descriptor

Automatic mass detection in mammograms is a challenging problem. The importance of this problem has attracted several researchers during the last decade and many algorithms have been proposed to deal with this problem. However, almost all these algorithms result in a large number of false positives/false negatives. For this problem, we introduce a new technique. The key idea of our approach is to represent textural properties of mammograms using Weber Local Descriptor (WLD), which has been shown outperforming stat-of-the-art best texture descriptors. The basic WLD descriptor is holistic by construction because it integrates the local information content into a single histogram. We extend it into a spatial WLD descriptor, which better encodes both the local region appearance and the spatial structure of the masses. Support Vector Machines (SVM) are employed for detecting masses and normal but suspicious parenchymal regions. The detection accuracy of the proposed system is Az = 0.988±0.006 on DDSM database; it outperforms the state-of-the-art best algorithms in the reduction of false positive/false negatives.