Detection of clusters of microcalcifications using neural network-based schemes

The authors present a neural network-based approach for detection of clusters of microcalcifications in digitized mammograms. Histogram modification and adaptive morphological filtering are initially applied to the mammograms to improve the conspicuity of possible microcalcifications. This is followed by a segmentation procedure that uses local image statistics to define possible blocks containing individual microcalcifications. The 2D principal component (PC) transform is used to extract the energy related features of the candidate data blocks. These PCs are then processed using a multi-layer backpropagation neural network to detect the actual microcalcifications. Finally, clusters are formed based upon a preselected proximity criterion. The effectiveness of the proposed schemes is demonstrated on a number of digitized mammograms