DMDR-UNet:An algorithm for retinal blood vessel segmentation
SUN Junding
ZHANG Hongying
Abstract：Aiming at the difficulty of collecting microvascular features of retinal vessels in fundus,an im-proved DMDR-UNet(deformable multiscale dense residual-UNet)segmentation model was proposed based on U-Net model.Firstly,a deformable convolutional network module was proposed to enhance the extraction of microvascular features,considering the free trend and rich morphology of retinal vessels.Secondly,multi-scale cavity convolution module was used to aggregate multi-scale information of retinal vessels under differ-ent receptive fields.Finally,a dense residual connection module was designed to reduce the semantic gap be-tween codecs and enhance the interaction and supplement of feature information.Experiments based on DRIVE dataset showed that the proposed method could accurately identify and segment retinal microves-sels,and achieve better segmentation results.
Keywords：retinal vascular segmentationU-Netdeformable convolutionmultiscaledense residual
Publication Date：2023-12-28
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：7( 142-148 )