Automated Segmentation of Breast Tissue and Pectoral Muscle in Digital Mammography

Abstract

Breast tissue and pectoral muscle segmentation are an important preliminary processing step in automated analysis of breast tumors from digital mammography. Without proper pectoral muscle exclusion, textural analysis of the breast tissue imagery would be hindered. This paper presents an automated segmentation of breast tissue and pectoral muscle in digital mammography based on simple histogram operation and k-means classification of the mammogram pixel intensity. We tested our algorithm in 25 digital medio-lateral oblique (MLO) mammograms with confirmed microcalcifications provided by the mini MIAS public database. Our results show that the proposed algorithm can successfully delineate the boundary of pectoral muscle next to the adjacent breast tissue. The best accuracy is achieved in mammograms of fatty-glandular dominant breast tissue and the worst accuracy is achieved in dense-glandular breast mammograms. Due to its simplicity and independence to mammogram intensity range, the algorithm may have potentials to be implemented in the pipeline of mammographic computer-aided diagnosis to assist large-scale screening of breast cancer.