Patch-based feature extraction algorithm for mammographic cancer images

Abstract

The study of mammography aims at identifying the presence of cancerous or non-cancerous tissue by using signs of bilateral asymmetry, masses, calcification and architectural distortion. The most vigilant one among them is the architectural distortion owing to speculated or random patterns. In this paper, a novel method for pectoral muscle removal and annotation removal is explained. A patch-based algorithm is implemented to extract textural features, and according to the features, a neural classifier has been classified into benign or malignant. The method was experimented on 88 images from MIAS database, and the proposed method has a total efficiency of 92.04% with respect to pectoral muscle and annotation removal.