Characterization of mammographic masses based on local photometric attributes

Abstract

This paper proposes Local Photometric Attributes (LPA) for the characterization of mammographic masses as benign or malignant. LPA measures the local information over the optical density image which suppresses the background region and provides more details about the mass lesion. The evaluation of the proposed approach is conducted by incorporating the mammograms of two benchmark databases—mini-MIAS and DDSM where a ten-fold cross validation technique is employed with different classifiers—Fishers Linear Discriminant Analysis, Random forest, and Support vector machine after filtering the optimal set of features by utilizing stepwise logistic regression method. The best performance achieved by the introduced approach in terms of an area under the receiver operating characteristic (ROC) curve (Az value) and accuracy (Acc) are 0.94 and 86.90%, respectively for the mini-MIAS dataset while the same for the DDSM dataset are 0.89 and 80.76%, respectively. The competitive nature of the proposed scheme is evident by comparing the obtained results with schemes in the state-of-the-arts.