Swin Transformer Encoder and Local Attention Network Via Transfer Learning Strategy for Retinal Vessel Segmentation

The effective segmentation of retinal blood vessels is an important means for diagnosis of fundus diseases. However, recent segmentation methods often ignore the global dependence of retinal blood vessels and have poor segmentation capability to small blood vessels. In addition, these methods are trained on limited datasets, resulting in a weak generalization ability. To solve these problems, we propose a new method based on Swin Transformer encoder and local attention network. First, Swin Transformer is used as an encoder to extract the global dependence of retinal blood vessels which can improve the continuity of blood vessels in segmentation results. Then, a local attention network based on edge detection module and coordinate attention module is developed to enhance the segmentation ability of the method to small blood vessels. Finally, a transfer learning strategy is also utilized in this paper. The encoder of our proposed method is pre-trained on ADE20K to enhance the generalization ability. Our proposed method is evaluated on three retinal image datasets: DRIVE, STARE and CHASE-DB1. The results demonstrate that the method proposed in this paper has better performance than others.