Optic Cup Segmentation from Fundus Image Using Swin-Unet

Glaucoma is one of the main causes of blindness, it is characterized by an increase in the area ratio of optic cup (OC) and optic disc (OD). Therefore, OC segmentation is an important basis for computer-aided diagnosis of glaucoma. However, the accuracy of OC segmentation still needs to be improved due to the low contrast, occlusion by blood vessels, and different scales of OC in fundus images. This study applies swin-unet to do OC segmentation. The swin-unet is composed of U-shaped structure and swin-transformer block. U-shaped structure can extract features from different scales and fuse a large number of features to solve the difficulties of low contrast and different scales. In addition, swin-transformer block can extract the features of the relationship between each patches, which can solve the difficulties of occlusion by blood vessels. In this study, swin-unet obtained mean intersection of union (IoU) of \(77.76\%\) and mean DICE of \(86.94\%\) , which is better than the segmentation results of U-Net and swin-transformer. Therefore, the effectiveness of swin-unet in OC segmentation is proved in this study.