Diagnosis of Breast Cancer from Mammogram Images Based on CNN

Abstract

Breast cancer has become the most common malignant tumor with the highest incidence of death in women. The MIBCAD (Medical Image Based Computer-Aided Diagnosis) system currently in use has a low diagnostic accuracy rate of only 85%. Furthermore, this system has major limitations for image processing of mammogram. To address these issues, this paper proposed a breast cancer diagnosis method based on an improved CNN (Convolutional Neural Networks). To avoid the image overfitting problem, transfer learning and data augmentation methods were used. The image classification accuracy was improved by using different CNN structures and changing the classifier type. Our results showed that the classification accuracy of the model reached 91.4%, which was significantly improved compared with the existing MIBCAD system.