Retinal Fundus Image Segmentation Based on Channel-Attention Guided Network

With the frequent use of electronic products by modern people, retinal diseases have become one of the most common maladies. Among them, Glaucoma is a serious visual disorder that contributes to irreversible visibility loss and blindness due to the destruction of the optic nerve. The clinical diagnosis of glaucoma is performed manually by experienced ophthalmologists to evaluate the diameter ratio of the optic cup (OC) to the optic disc (OD) called the cup to disc ratio (CDR). In this paper, we propose a deep learning framework named CG-Net for OC and OD segmentation simultaneously using the channel attention method. The proposed CG-Net is inspired by the U-Net with multi-scale input. In addition, a novelty channel attention block is introduced to incorporate the feature map between different layers and reduce the background noise for better attention to the target area. We extensively evaluate our framework on the ORIGA dataset and a comparison with several state-of-the-art methods also demonstrates the segmentation effectiveness of our method. The results indicate that the proposed model outperforms other methods in terms of segmentation performance and achieves a total average overlap error of 0.139 in the metrics.