基于拓扑结构的眼底图像分割评价新方法  

New Approach of Fundus Image Segmentation Evaluation Based on Topology Structure

作　　者：盛韩伟 戴培山[1] 刘智航 张文妙韵 赵亚丽[1] 范敏[2] 

机构地区：[1]中南大学地球科学与信息物理学院生物医学工程系,长沙410083 [2]湖南女子学院教育与法学系,长沙410004

出　　处：《生物医学工程学杂志》2015年第5期1100-1105,共6页Journal of Biomedical Engineering

基　　金：国家自然科学基金项目(81171420)

摘　　要：在眼底图像分割效果的评价中,针对传统评价方法只考虑像素点重合而未考虑视网膜血管拓扑结构的不足,本文提出一个新的评价方法。该方法首先利用数学形态学和细化算法得到血管的拓扑结构,然后统计并分析视网膜血管区域3个特征参数的分布情况,即以互信息、相关系数和节点率来获得基于拓扑结构的眼底图像分割评价结果。该方法的实验数据取自STARE公开数据库中专家手工分割及其腐蚀结果。实验结果表明:互信息、相关系数和节点率这三个特征参数可以从拓扑结构的角度来评价眼底图像视网膜血管的分割效果,且算法复杂程度较低,该方法对眼底图像视网膜血管分割评价方法的补充有重要意义。In view of the evaluation of fundus image segmentation, a new evaluation method was proposed to make up insufficiency of the traditional evaluation method which only considers the overlap of pixels and neglects topology structure of the retinal vessel. Mathematical morphology and thinning algorithm were used to obtain the retinal vas- cular topology structure. Then three features of retinal vessel, including mutual information, correlation coefficient and ratio of nodes, were calculated. The features of the thinned images taken as topology structure of blood vessel were used to evaluate retinal image segmentation. The manually-labeled images and their eroded ones of STARE da- tabase were used in the experiment. The result showed that these features, including mutual information, correlation coefficient and ratio of nodes, could be used to evaluate the segmentation quality of retinal vessel on fundus image through topology structure, and the algorithm was simple. The method is of significance to the supplement of tradi- tional segmentation evaluation of retinal vessel on fundus image.

关 键 词：视网膜血管分割 图像分割评价 拓扑结构 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]