Mass candidate detection and segmentation in digitized mammograms

This paper introduces a system for identifying candidate masses in digitized mammograms. Mass identification is a basic component in Computer-Aided Detection (CAD) systems for mammograms. The proposed algorithm is a cascaded filtering process that consists of several stages: First, a new breast fat model is introduced and the fat content in the image is estimated and removed from the image to obtain a fatless image at a standard resolution. Next, a Gabor filter is specially designed and tailored to fit the mass detection problem and then applied to the fatless image. Finally, the resulting image is segmented to obtain iso-contours. Candidate regions are then identified by contour processing and selection. The proposed algorithm obtained 100% sensitivity with 3.4 false positives per image.