Automatic classification of breast tissue density

Breast cancer is an international public health concern. Medical imaging is one of the key elements in diagnosis. However, the quality of the interpretation of mammograms remains variable. One of the important characteristics in breast anatomy and physiology is breast tissue density. Density is important for two main reasons: first, increased breast density is associated with decreased mammo graphic sensitivity for the detection of breast cancer (Schetter, 2014). Second, breast density is one of the strongest known risk factors for breast cancer (Prevrhal et al., 2002; Boyd et al., 1995). For these reasons, automatic tissue density classification is an important process in diagnosis. Moreover, the BI-RADS (Breast Imaging-Reporting And Data System) classification system identifies four levels of breast density, but the mini-MIAS (Mammographic Image Analysis Society) database is divided into three density categories. In this article we describe a method for overall breast density classification using artificial neural networks. This approach has the advantages of not requiring a preprocessing step and the ability to be adapted to different mammography databases. The validation of our method is demonstrated using 240 mammograms from the DDSM database and 180 mammograms from mini-MIAS database, with the correct classification rate of 87.50% and 86.11%, respectively.