Classification of breast tissues in mammographic images in mass and non-mass using McIntosh's diversity index and SVM

Abstract

This paper introduces three approaches, which use McIntosh's Diversity index to extract breast tissue features from mammographic images, for later classification, through Support Vector Machine (SVM), into mass and non-mass. In order to implement the diversity index, it is necessary to define the element that will represent the species in the image. So, in the first approach, the intensities of the pixels of the image are treated as species, and the texture statistic used is the histogram. Considering the spatial relations of direction and distance between pixels, we adopted a second approach, using GLCM as texture statistic, where the species are represented by pairs of pixels, and the third approach, using GLRLM as texture statistic, where the species are represented by gray level run lengths. We achieved an accuracy of 60.25% with the first approach, 99.00% with the second one and 99.75% with the third one. © 2012 Springer-Verlag.