Abstract Retinal vessel segmentation can obtain rich ocular information which is important for the diagnosis of fundus diseases. To address the problems of existing segmentation methods such as poor capillary segmentation and incorrect segmentation of pathological information, an AFNet vessel segmentation network combining location attention, semantic aggregation module and multi-scale feature fusion module is proposed. Add positional attention to the feature codec block of the network for modeling global dependencies and reducing intra-class inconsistencies, the multiscale feature fusion module is used in the last layer of the coding part to extract multiscale feature information to solve the difficult problem of large variation of retinal vessel width and size, and the designed semantic aggregation module can fully utilize the contextual semantic information to improve the segmentation accuracy of capillaries. Extensive experiments are conducted on three publicly available fundus image databases DRIVE, STARE and CHASE_DB1, and the results show that AFNet can effectively improve the accuracy of retinal vessel segmentation and achieve better comprehensive performance compared with other methods.
展开▼
