Breast Cancer Histopathological Image Classification Based on Convolutional Neural Networks

Convolutional neural networks (CNNs), with the remarkable success in a variety of computer vision tasks, have recently achieved the effective breakthrough for breast cancer histopathological image classification. In this study, we further explore the performance of CNN architectures for this task. More specifically, we systematically study two recent milestones of CNNs, i.e., VggNet and ResNet, for breast cancer histopathological image classification. Considering large variety among within-class images, we adopt larger patches of the original image as the input of network to combine global and local features. We also study the effect of data augmentation for this task, which reduces overfitting problem to a certain degree. We also conduct extensive experiments on the BreakHis dataset and draw some interesting conclusions. Particularly, the optimal classification accuracies achieved by ResNet-50 with 40× images reach to 92.68% on image level and 93.14% on patient level respectively, illuminating the effectiveness of the employed CNN model.