Deep Attentive Feature Learning for Histopathology Image Classification

Pengxiang Wu, Hui Qu, Jingru Yi, Qiaoying Huang, Chao Chen, Dimitris N. Metaxas

Abstract: In this paper, we present a new deep learning-based approach for histopathology image classification. Our method is built upon standard convolutional neural networks (CNNs), and incorporates two separate attention modules for more effective feature learning. In particular, the attention modules infer the attention maps along different dimensions, which help focus the CNNs on critical image regions, as well as highlight discriminative feature channels while suppressing the irrelevant information with respect to the classification task. The attention modules are light-weight, and enhances the feature representation with small extra computational overhead. Experimental results on the publicly available BreakHis dataset demonstrate that our method outperforms the state-of-the-arts by a large margin.