The study of retinal vessel segmentation based on improved U-net algorithm

Abstract

Fundus retinal vascular images are an important basis for clinical diagnosis and prevention of many diseases. Due to the complex structure of retinal vessels and the resistance to differentiation from the image background, it results in poor retinal vessel segmentation. In order to improve the accuracy and efficiency of fundus retinal image segmentation, the paper added multi-scale convolutional kernels based on the U-net network structure. This structure can extract the image feature information of different layers and fully learn the features of retinal images; and the residual network structure is added to prevent the problem of gradient disappearance and explosion when the network level deepens. Experiments show that the model proposed in this paper can effectively obtain multi-level feature information of images and improve the effect of retinal image segmentation. The accuracy of this algorithm in DRIVE dataset achieves 97.45% and 82.56% sensitivity.