Attentional Dilated Convolution Neural Network for Nuclei Segmentation in Histopathology Images

The foundational task of pathological diagnosis is to accurately segment the nuclei in pathology images. Deep learning has become a well-validated method of nuclei segmentation, but the touching or overlapping nuclei are difficult to segment. In this research, we bring forward a new strategy for nuclei segmentation. The channel attention module is utilized for reweighting channel-wise features and further improving segmentation performance. Moreover, dilated convolution is used for collecting multi-scale contextual information and preserving original resolution. We name our method as Attentional Dilated Convolution Neural Network (ACDNN). Experiments on a large tissue image dataset show that our strategy is superior to other advanced methods.