NMF-density: NMF-based breast density classifier

The amount of tissue available in the breast, commonly characterized by the breast density, is highly correlated with breast cancer. In fact, dense breasts have higher risk of developing breast cancer. On the other hand, breast density influences the mammographic interpretation since it decreases the sensitivity of breast cancer detection. This sensitivity decrease is due to the fact that both cancerous regions and tissue appear as white areas in breast mammograms. This paper introduces new features to improve the classification of breast density in digital mammograms according to the commonly used radiological lexicon (BI-RADS). These features are extracted from non-negative matrix factorization (NMF) of mammograms and classified using machine learning classifiers. Using ground truth mammographic data, the classification performance of the proposed features is assessed. Simulation results show that the latter significantly outperforms existing density features based on principal component analysis (PCA) by achieving higher classification accuracy.