ABSTRACT
Automatic mammogram registration provides information about gradual changes in temporal mammograms. Segmentation of landmarks such as the breast, breast skin line, and pectoral muscle are required for this process. This paper presents a new machine learning algorithm, known as margin setting algorithm (MSA), to segment the breast and pectoral muscle. MSA creates multiple prototypes to enclose patterns belonging to different classes. In this research, we applied MSA to segment the breast and pectoral muscle. The performance of our algorithm is compared with four different algorithms; neural network (NN), and three thresholding algorithms; ant colony optimization (ACO), global, and Otsu. These algorithms were tested on 554 mammograms from 125 patients. Subjective evaluation, by four researchers in the area of pattern recognition, was used to compare the outcomes. MSA outperformed NN algorithm in 84.21% of the mammograms. Also, MSA outperformed the other three algorithms in 98.12% of the mammograms.
