Pathological Image Classification of Breast Cancer Based on Residual Network and Focal Loss

Wenting Shu, Shaoyu Wang, Qiang Chen, Yun Hu, Zhengwei Cai, Runlong Lin

Abstract: This paper proposes an improved deep residual neural network for the classification of breast cancer as either benign or malignant. Inspired by the success of using focal loss in object detection, we present a new focal loss for pathological image classification to solve the class imbalance problem encountered during training stage. Furthermore, we introduce a multi-scale acquisition structure into ResNet to get a larger range of receptive fields for each network layer and represent features at multiple scales. Data enhancement and migration learning are also used to optimize the initial parameters solving the problem of overfitting in the network. Experimental results show that our approach achieves higher accuracy of classification compared to previous methods.