糖尿病性视网膜病变图像黄斑水肿等级自动分析方法

Automatic Analysis Method of Macular Edema Classification for Diabetic Retinopathy Images

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。

Diabetic macular edema （DME） can appear in any stage of the diabetic retinopathy （DR） and is the leading cause of vision impairment in people with diabetes. Therefore, the automatic analysis of DME plays a key role in the DR screening program. According to the DME international clinical classification standard, detecting and judging the presence of hard exudates （HEs） close or relate to macula fovea （MF） is a standard method to assess DME in the fundus images. A synthetic improvement method based on existing mathematical morphology technique is selected for HEs detection. A novel method of macular fovea center location is proposed based on the directional local contrast filter and the local vessel density. Then the optic disk region can be easily located and removed for reducing the impact to the HEs detection. Only the large vessel segmentations were used for the vessel density calculation. From the testing results on the 169 images from HEI- MED public dataset, the sensitivity and specificity of HEs detection on image level are 100% and 92. 2% , respectively. The accuracy of macular fovea center location reaches 98.2%. And the accuracy for each DME classification is more than 88%. The proposed method would have important clinical application potentials.