摘要
针对传统卷积神经网络(convolutional neural network,CNN)受感受野大小的限制,无法直接有效地获取空间结构及全局语义等关键信息,导致宽血管边界及毛细血管区域特征提取困难,造成视网膜血管分割表现不佳的问题,提出一种基于图卷积的视网膜血管分割细化框架。该框架通过轮廓提取及不确定分析方法,选取CNN粗分割结果中潜在的误分割区域,并结合其提取的特征信息构造出合适的图数据,送入残差图卷积网络(residual graph convolutional network,Res-GCN)二次分类,得到视网膜血管细化分割结果。该框架可以作为一个即插即用模块接入任意视网膜血管分割网络的末端,具有高移植性和易用性的特点。实验分别选用U型网络(U-neural network,U-Net)及其代表性改进网络DenseU-Net和AttU-Net作为基准网络,在DRIVE、STARE和CHASEDB1数据集上进行测试,本文框架的Sp分别为98.28%、99.10%和99.04%,Pr分别为87.97%、88.87%和90.25%,证明其具有提升基准网络分割效果的细化能力。
Due to the limitation of the size of the receptive field,the traditional convolutional neural network(CNN)cannot directly and effectively obtain the key information such as spatial structure and global semantics,which leads to the difficulty in feature extraction of wide blood vessel boundary and capillary region,resulting in poor performance of retinal vascular segmentation.Therefore,a retinal vascular segmentation refinement framework based on graph convolution is proposed in this paper.In the framework,through contour extraction and uncertainty analysis method,the potential false segmentation regions in the coarse segmentation results of CNN are chosen,and the appropriate graph data is constructed by combining the extracted feature information.The graph data is sent to the residual graph convolutional network(Res-GCN)for secondary classification,and the retinal vascular refinement segmentation results are obtained.The framework can be used as a plug-and-play module to access the end of any retinal vascular segmentation network,which has the characteristics of high portability and usability.In the experiment,U-neural network(U-Net)and its representatively improved networks DenseU-Net and AttU-Net are selected as the benchmark networks,and tested on the DRIVE,STARE and CHASEDB1 datasets.The Sp of the framework is 98.28%,99.10%and 99.04%,respectively,and the Pr is 87.97%,88.87%and 90.25%,respectively,which prove that it has the refinement ability to improve the segmentation effect of the benchmark network.
作者
吕佳
梁浩城
王泽宇
LV Jia;LIANG Haocheng;WANG Zeyu(College of Computer and Information Sciences,Chongqing Normal University,Chongqing 401331,China;National Center for Applied Mathematics in Chongqing,Chongqing Normal University,Chongqing 401331,China)
出处
《光电子．激光》
CAS
CSCD
北大核心
2023年第6期654-662,共9页
Journal of Optoelectronics·Laser
基金
国家自然科学基金重大项目(11991024)
重庆市教委“成渝地区双城经济圈建设”科技创新项目(KJCX2020024)
重庆市教委重点项目(KJZD-K202200511)
重庆市科技局技术预见与制度创新项目(2022TFII-OFX0044)资助项目。
