AGC-UNet:A Global Context Feature Fusion Method Based On U-Net for Retinal Vessel Segmentation

Computer-aided retinal vascular segmentation plays an irreplaceable role in the diagnosis of hypertension, retinal vascular occlusion, diabetic and other diseases. In this paper, we propose a global context feature fusion retinal vessel segmentation model based on U-Net, named AGC-UNet, which utilizes the encoding and decoding network, and uses Globle Context Block (GCB) in the encoding and decoding path to enhance the global context fusion of vascular features, and did not introduce a large amount of computation. In addition, Attention Gate Block(AGB) is added into the jump connection part to enhance the spatial extraction ability of vascular features, to weaken the ability of learning unrelated areas, and to improve the ability of vascular segmentation. AGC-UNet model is experienced respectively in the open datasets DRIVE and CHASE_DB1 and evaluating indicators of accuracy(Acc) in these two datasets are 0.9653 and 0.9646 respectively, 0.8347 and 0.8206 in sensitivity(Se), 0.9851 and 0.9791 in specificity(Sp) and 0.8639 and 0.8095 in F1-Score(F1). Compared with the newest existing methods, this method performs an outstanding performance in retinal vascular segmentation.