A multistage methodology for detection of microcalcifications in digital mammography
Abstract
This paper describes a coarse-to-fine approach for detection and segmentation of clustered microcalcifications. Starting from digitized images of whole mammographic films, the method involves three main stages. First, the breast is segmented. Then a texture analyzer based on the fractional Brownian motion model is applied to select those regions that may contain a cluster. The last stage computes the isophote map of the selected regions and considers the relative positions of isophotes to recognize the microcalcifications.
The method was evaluated using a database of 150 mammographic images of patients who underwent a surgical biopsy for a cluster of microcalcifications. In 95% of mammograms, the clusters are correctly localized if we accept an average 1,5 false positive cluster per mammography.