抄録
It is important to identify the nipple region in CAD (computer-aided diagnosis) schemes for mammograms. The purpose of this study was to develop a computerized method to automatically identify the nipple location on mammograms. Our database consisted of 50 mammograms which were randomly selected from DDSM (Digital Database for Screening Mammography in the University of South Florida). In our computerized method, the breast region was first segmented by a gray-level thresholding technique based on the gamma correction. The margin of the segmented breast region was defined as the skin line. The pixels on the skin line were selected as pixels of interest (POIs). The curvature of skin line, the convergence index of mammary glands, and the distance from best wall were extracted at each POIs. The nipple region was identified based on the output of an artificial neural network with these three objective features. With this computerized method, the detection accuracy of nipple region was 80.0% (40/50). This method would be useful in CAD schemes for mammograms.
