Classifying Breast Cancer Histopathological Images Using A Robust Artificial Neural Network Architecture

Pathological diagnosis is the standard for the diagnosis and identification of breast malignancies. Computer-aided diagnosis (CAD) is widely applied in pathological image analysis to help pathologists improving the accuracy, efficiency, and consistency in diagnosis. The traditional CAD methods rely on the expert domain knowledge, time-consuming feature engineering, which is insufficient to real-world systems. In recent studies, deep learning methods have been explored to improve the performance of pathological CAD. However, typical deep methods mainly suffer from the following limitations on pathological image classification. (i) The model cannot extract rich and informative features due to the shallow network structure. (ii) The commonly adopted patch-wise classification strategy makes it impossible to obtain the global features at the image level. To address the two issues, in this paper we propose to use a deep ResNet structure with Convolutional Block Attention Module (CBAM), in order to extract richer and finer features from pathological images. Moreover, we abandon the patch-wise classification strategy and perform an end-to-end training instead. The public BreakHis dataset is used to evaluate our proposed method. The results show that our model achieves a significant improvement over the baseline methods.