Classification of Mammograms Based on Gray Level Intensity

Abstract

Breast cancer is one of the leading causes of death among women. To find out the type of breast tissue is the first step towards knowing whether a woman has cancer /likely to develop the risk of cancer or not. Out of fatty, glandular and dense breast tissue types; dense tissue type has the most probability of having/developing malignant tissues. In this work, mammograms are treated as textures and a method is proposed to classify them into fatty, glandular and dense breast tissue type. Three new features based on gray level intensity are proposed in this work and are merged with well-known gray-level co-occurrence matrix (GLCM) features for tissue classification. In total 16 features are extracted from 150x150 region of interest (ROI). Out of 16 features, optimal features are selected through correlation feature selection (CFS). Finally, classification is done using Naïve Bayes classifier, Random Forest classifier and k-nearest neighbor (k-NN) classifier on four sets of features which are: only GLCM features, three proposed features, combination of both (GLCM+proposed) and seven optimal features. The results of all the set of features are compared on the basis of their accuracy. Subsequently, mammograms are classified into fatty and dense using the same procedure.