Automatic Segmentation of Pectoral Muscle in Mammogram Images Using Global Thresholding and Weak Boundary Approximation

Abstract

Removal of pectoral muscle is a major challenge in automatic detection of tumors breast mammograms. The intensity value for pectoral muscle is same as tumor presented (if any) in the breast region. It interferes in the detection of breast tumor and maximizes false positive error. In this paper, an efficient approach is introduced to remove the pectoral muscle from breast mammograms. Unwanted labels are removed by segmenting high intensity areas using convex-hull. Median filter is used to remove salt and pepper noise. Then global thresholding is used for the removal of breast tissues which appear along with pectoral muscle region during thresholding by pixel count on the connection of pectoral muscle and breast lesions. If the number of pixels in a small area are less than preset value on the boundary of pectoral muscle than it is considered not to be the part of pectoral muscle. Using this approximation regions outside the pectoral muscle are removed. Also minimal connectivity array is used to segment the portion connected to pectoral muscle which is not a part of it. The well-known mini-MIAS database with 322 mammograms is used to study the performance of the proposed method. 92.86% images are well segmented; 4.97% images are also segmented to an acceptable level. This performance is significantly better than that of several existing techniques.