A novel region growing based method to remove pectoral muscle from MLO mammogram images

Abstract

Digital Mammography is the most efficient screening technique for early detection of breast abnormalities. Automated computer aided methods have been very effective in identifying subtle signs of breast cancer like microcalcification and masses. However such methods for detection of masses are highly affected by the presence of pectoral muscle in mediolateral oblique view. So it is highly recommended to remove pectoral muscle as a preprocessing step. In this paper, a novel method for pectoral muscle removal from Mediolateral Oblique mammogram images is presented. The method has three main phases. In the first phase a triangular region is defined over the mammogram that separates the pectoral muscle from the rest of the tissue. In the second phase, a local region growing method is applied within the triangular area defined to suppress the pectoral muscle. In the last phase, the pectoral edge is refined using gradient information of the edge. Results and Conclusion: The system is tested over 150 images taken from Mini-MIAS dataset. Hand-drawn segmentation masks are used to compare the segmentation accuracy for the proposed method.