Breast density classification using a bag of features and an SVM classifier

Breast cancer is one of the most common types of cancer, as well as the leading cause of mortality among women. Numerous attempts have been made in developing automated systems for identifying cancerous tissues from mammographic images. Breast density, in particular, is seen as the main characteristic used in Computer-Aided Diagnosis (CAD) systems. As such, breast density classification is seen as a crucial preprocessing stage in cancer detection. Here, we propose a new image preprocessing density classification approach for early detection of cancer from mammogram images using a set of robust texture and edge related features combined with an SVM classifier. The proposed algorithm achieves an accurate density classification of 93.56% for Low and High densities. The results were validated using the IRMA database. A comparison to state-of-the-art has been carried and showed that the proposed approach achieves improved performance in terms of density classification accuracy.