The segmentation of the optic disc in color fundus images is crucial for the identification of ophthalmic diseases. To address the issues of inaccurate disc edge segmentation due to various factors and the low efficiency of segmentation algorithms,an automatic disc segmentation is proposed which is based on the lightweight U2-Net and integrated with an edge attention mechanism. This method employs the lightweight U2-Net as the backbone network,utilizes a preprocessing approach of extracting the region of interest of the optic disc to remove irrelevant features,and simultaneously introduces an edge attention mechanism to enhance the extraction capability of optic disc edge features. On two public datasets,Drishti_GS and REFUGE,the F1 score reaches 97.82％ and 97.36％,the Dice similarity indices reaches 97.15％ and 96.64％,and the IOU reaches 94.47％ and 93.50％,respectively. Compared to other network models,the proposed method demonstrates superior segmentation performance,showcasing its clinical application value.
color fundus imagesoptic disc segmentationU2-Netregion of interest extractionedge attention
万方数据
维普
