Higher accuracy and throughput in computer-aided screening of mammographic microcalcifications

This paper outlines a new methodology for computer-aided screening (CAS) of microcalcifications. The proposed approach takes advantage of the research already performed in the field of CAS and facilitates its translation to practical screening systems by exploring two concepts that address the key issues of (1) higher accuracy and (2) higher throughput. The first concept involves the utilization of a front-end fractal encoding scheme that analyzes digitized mammograms and generates focus-of-attention regions (FARs). The second concept is that of distributed segmentation, that is, subjecting FARs to multiple microcalcification segmentation techniques. To validate the impact of FAR generation on the ensuing processes in CAS, specifically that of segmentation, an existing segmentation technique was selected and applied to 100 digitized mammograms (45 with and 55 without microcalcifications). Utilizing the selected segmentation technique in conjunction with the proposed fractal encoding scheme reduced the number of false detections from 310 to 53 (83% reduction) in images with microcalcifications and from 931 to 197 (79% reduction) in images without microcalcifications. This was accomplished while maintaining a coverage rate of 88%. In addition, the input data were reduced by up to 99%.