A CNN-Transformer Hybrid Network for Joint Optic Cup and Optic Disc Segmentation in Fundus Images

Glaucoma is one of the main causes of blindness in the world. Accurate segmentation of optic cup (OC) and optic disc (OD) from retinal fundus images is an important task in glaucoma screening. In this paper, we propose a segmentation network combining CNN and Transformer to jointly segment OC and OD from retinal fundus images. First, we employ CNN to obtain local features and combine the ASPP module to capture multi-scale features. The Transformer is introduced for long-distance modeling to obtain global context information. Then, the two parts of features are fused to further improve the segmentation performance of the OC and OD. Finally, the proposed network is trained and tested on the public REFUGE dataset. Compared with the current mainstream methods, it is verified that the proposed network has better performance on OC and OD segmentation. It provides the basis for the realization of large-scale glaucoma screening.