Utilization of DenseNet201 for diagnosis of breast abnormality

As one of the leading killers of females, breast cancer has become one of the heated research topics in the community of clinical medical science and computer science. In the clinic, mammography is a publicly accepted technique to detect early abnormalities such as masses and distortions in breast leading to cancer. Interpreting the images, however, is time-consuming and error-prone for radiologists considering artificial factors including potential fatigue. To improve radiologists' working efficiency, we developed a semi-automatic computer-aided diagnosis system to classify mammograms into normality and abnormality and thus to ease the process of making a diagnosis of breast cancer. Through transferring deep convolutional neural network DenseNet201 on the basis of suspicious regions provided by radiologists into our system, we obtained the network we termed DenseNet201-C, which achieved a high diagnostic accuracy of 92.73%. The comparison results between our method and the other five methods show that our method achieved the highest accuracy.