Retinal vascular segmentation based on local feature enhancement
Retinal blood vessels are small and complex.When segmentating retinal blood vessels,noise,fracture and undersegmentation often occur.To solve this problem,a lightweight network named LRU-Net based on local fea-ture enhancement is proposed to capture more features of small blood vessels.Firstly,a feature extraction module is added to the channel attention module to extract secondary features from input features so as to obtain more detailed features.Secondly,a feature fusion module is designed,which can fuse the high and low features more effectively in the decoder,and strengthen the final feature representation.Finally,a context aggregation module is designed to ex-tract multi-scale information with different resolutions of the deepest features,and then splicing it to make the input features into the upper sampling more detailed.Experimental results on FIVES and OCTA-500 data sets show that compared with U-Net,the proposed method not only achieves lightweight,but also improves the accuracy and Dice coefficient of retinal vessel segmentation to a certain extent.
NETL
NSTL
万方数据
