Ets-Net: Edge Enhanced Two-Stream Network for Medical Image Segmentation

Over the past several years, deep learning methods were widely applied to some medical image processing task such as medical image segmentation and achieved adequate performance. However, universal and effective methods are still needed to segment the details such as the edge of object in some medical image segmentation task. And the medical images nowadays tend to contain richer detail information with the development of the medical imaging technology, which makes the above challenge more obvious. To this end, we propose an edge enhanced two-streams network (ETS-Net) to improve the segmentation near the edge of the object. The proposed network contains a regular stream aiming at getting the contextual information and an edge detail stream aiming at capturing more detail features near the edge. Moreover a multi-scale supplement module (MSSM) is applied to the regular stream to obtain more contextual information from different scales and a context enhance fusion module (CEFM) is used to fuse the features from the two streams. The proposed method is thoroughly validated in three different datasets, and the experimental results demonstrate that the proposed method generally outperforms other state-of-the-art algorithms for various metrics.