A novel algorithm for pectoral muscle removal and auto-cropping of neoplasmic area from mammograms

Presence of pectoral muscle has always been a hindrance in neoplasm detection in screening mammography. Mediolateral-oblique (MLO) x-ray view of the breast taken while screening mammography shows the presence of pectoral muscle. The intensity range shared by pectoral muscle, masses and calcification clusters being almost the same makes pectoral muscle removal a vital or necessary step to attain proper segmentation of actual region of interest (ROI) i.e. the neoplasmic region. This paper provides a novel algorithm for automatic detection and removal of pectoral muscle along with breast boundary detection and several artefacts removal present in digital mammograms. A concatenation of an auto-cropping algorithm to pectoral removal step gives a précise RoI which helps in stepping up the lesion detection accuracy of the Computer-Aided Detection (CAD) system. This composite method has been has been implemented and applied to mini-MIAS which is one of the most challenging digital database consisting 322 MLO view mammograms. The algorithm shows an accuracy of around 83.89% on a set of 298 mammogram images.