糖尿病性视网膜病变图像黄斑水肿等级自动分析方法

发布时间：2018-10-12 20:54

【摘要】：糖尿病性黄斑水肿(DME)可以出现在糖尿病性视网膜病变(DR)的任何阶段,是导致糖尿病患者视力损伤的主要原因,因此DME自动分析是DR筛查的关键内容。依据DME国际临床分级标准,通过检测并判断硬性渗出(HEs)是否接近或涉及黄斑中心,可对眼底图像进行DME等级的自动分析。HEs检测选择基于现有的数学形态学方法的综合改进;黄斑中心定位则引入定向局部对比度滤波结合局部血管密度的新方法,可同步确定并去除视盘区域,以消除对HEs检测的影响,其中血管密度仅需提取粗血管网络。经开放的HEI-MED数据集中169幅眼底图像的测试,HEs检测在图像水平上获得100%敏感性和92.2%特异性;黄斑中心定位正确率98.2%;各DME等级评价正确率均在88%以上,具有重要的临床参考和应用价值。
[Abstract]:Diabetic macular edema (DME) can occur at any stage of diabetic retinopathy (DR) and is the main cause of visual impairment in diabetic patients. So DME automatic analysis is the key content of DR screening. According to the international clinical classification standard of DME, by detecting and judging whether the rigid exudation (HEs) is close to or involved in macular center, the DME grade of the fundus image can be automatically analyzed. The HEs detection choice is based on the comprehensive improvement of the existing mathematical morphology method. A new method of directional local contrast filtering combined with local vascular density is introduced to determine and remove the optic disc area synchronously in order to eliminate the influence on HEs detection in which the vessel density only needs to be extracted from the coarse vascular network. By testing 169 fundus images in an open HEI-MED data set, the sensitivity and specificity of HEs were 100% and 92.2% at the image level, the accuracy rate of macular localization was 98.2%, and the accuracy rate of DME evaluation was more than 88%. It has important clinical reference and application value.
【作者单位】：福州大学物理与信息工程学院;
【基金】：国家自然科学基金(60827002) 福建省教育厅项目(JA13037) 福建省电子信息基础教学团队建设项目
【分类号】：R587.2;R774.1;TP391.41

【参考文献】