Evaluating the Efficacy of Multi‐resolution Texture Features for Prediction of Breast Density Using Mammographic Images
N.A. Kriti, Jitendra Virmani · 2017
It is an established fact that the chances of developing breast cancer have a strong correlation with high breast tissue density. Radiologists often characterize the density patterns exhibited by the breast tissue based on visual assessment of texture. This visual characterization is subjective, and for atypical cases it is considered to be a difficult task even for the experienced radiologist. Therefore, the design of an efficient computer-aided diagnostic (in this chapter, CAD) system for breast tissue density classification is clinically significant. In this work, an efficient CAD system based on multiresolution texture descriptors (derived from ten different compact support wavelet filters) using 2D wavelet transform has been implemented using a smooth support vector machine (SSVM) classifier. The standard Mammographic Image Analysis Society (MIAS) data set has been used for classifying the breast tissue into one of the three classes: fatty (F), fatty-glandular (FG), and dense-glandular (DG). The performance of the SSVM-based CAD system has been compared with that of SVM-based CAD system design. The highest classification accuracy of 89.4% with sensitivity values of 86.7, 86.5, and 94.6% have been achieved for F, FG, and DG classes, respectively, using the SSVM classifier and features derived from the db1 (Haar) wavelet filter. The promising results obtained from the current study indicate that the proposed CAD system design can be routinely used in a clinical environment for characterization of breast tissue density.