基于小波与粗概率先验双分支的眼底微细血管分割网络OA

Morphological parameters such as the diameter and bifurcation of retinal blood vessels serve as important references for clinical diagnosis,and accurate segmentation of microvessels is essential for the early diagnosis of related diseases.Existing segmentation methods often suffer from blurred edges,fractured topological structures,and inadequate fusion of thick and thin vessel features in microvessel detection.To address these issues,this paper proposes a dual-branch segmentation network.We adopt IterMiUnet as the coarse segmentation branch to extract global vascular features,and introduce Haar wavelet prior to enhance the frequency-domain representation and edge details of microvessels.Since coarse segmentation struggles to identify weak boundaries and fractured vessels accurately,we take uncertain pixels from coarse segmentation probability as prior guidance,enabling the fine segmentation model to focus on these challenging regions and boost segmentation precision.A dual-scale probability encoding fusion mechanism combined with temperature scaling and background suppression is also constructed.The outputs of coarse segmentation are optimized using fine segmentation results to realize adaptive fusion and reduce background interference.Experiments on DRIVE,CHASE_DB1 and STARE datasets demonstrate that the proposed algorithm outperforms mainstream methods,with F1 scores of 0.825 7,0.836 0 and 0.835 7 respectively.It achieves stable and reliable performance,and can effectively capture microvascular details while maintaining the topological integrity of blood vessels.