Breast Cancer Detection Based DenseNet with Attention Model in Mammogram Images

Breast cancer has become a very interesting topic due to the massive number of deaths among women across the world. Radiologists can diagnose breast cancer faster and more accurately because of advances in the computer-aided diagnosis (CAD) system. In this paper, we presented a new breast cancer detection system based on the integration of self attention model in the pre-trained deep neural networks DenseNet. First, we extracted automatic high-level features from breast images using DenseNet extraction layers, and thereafter attention model was applied to focus the treatment on the relevant parts of the region of interest. The experiments were conducted on a multi-class Mammographic Image Analysis Society (MIAS) database, including three classes of breast cancer images. We achieved the accuracy of 0.9939 when applying both transfer learning, data augmentation, and self attention mechanism.