Automatic extraction of pectoral muscle in the MLO view of mammograms

Abstract

A mammogram is the standard modality used for breast cancer screening. Computer-aided detection (CAD) approaches are helpful for improving breast cancer detection rates when applied to mammograms. However, automated analysis of a mammogram often leads to inaccurate results in the presence of the pectoral muscle. Therefore, it is necessary to first handle pectoral muscle segmentation separately before any further analysis of a mammogram. One difficulty to overcome when segmenting out pectoral muscle is its strong overlapping with dense glandular tissue which tampers with its extraction. This paper introduces an automated two-step approach for pectoral muscle extraction. The pectoral region is firstly estimated through segmentation by mean of a modified Fuzzy C-Means clustering algorithm. After contour validation, the final boundary is delineated through iterative refinement of edge point using average gradient. The proposed method is quite simple in implementation and yields accurate results. It was tested on a set of images from the MIAS database and yielded results which, compared to those of some state-of-the-art approaches, were better.