Classification of whole mammogram and tomosynthesis images using deep convolutional neural networks

Mammography is the most popular technology used for the early detection of breast cancer. Manual classification of mammogram images is a hard task because of the variability of the tumor. It yields a noteworthy number of patients being called back to perform biopsies, ensuring no missing diagnosis. The convolutional neural network (CNN) has succeeded in a lot of image classification challenges during the recent years. In this paper, we proposed an approach of mammogram and tomosynthesis classification based on CNNs. We had acquired more than 3000 mammograms and tomosynthesis data with approval from an institutional review board at the University of Kentucky. Different models of CNNs were built to classify both the 2-D mammograms and 3-D tomosynthesis, and every classifier was assessed with respect to truth-values generated by histology results from the biopsy and two-year negative mammogram follow-up confirmed by expert radiologists. Our outcomes demonstrated that CNN-based models we had built and optimized utilizing transfer learning and data augmentation have good potential for automatic breast cancer detection based on the mammograms and tomosynthesis data.