基于背景估计和SVM分类器的眼底图像硬性渗出物检测方法  

Hard Exudates Detection Method Based on Background-Estimation and SVM Classifier

作　　者：肖志涛[1] 王雯[1] 耿磊[1] 张芳[1] 吴骏[1] 赵北方 张欣鹏[1] 苏龙[2] 陈莉明[3] 单春燕[3] 

机构地区：[1]天津工业大学电子与信息工程学院,天津300387 [2]天津医科大学眼科医院,天津300384 [3]天津医科大学代谢病医院,天津300070

出　　处：《中国生物医学工程学报》2015年第6期720-728,共9页Chinese Journal of Biomedical Engineering

基　　金：天津市科技支撑计划重点项目(13ZCZDGX02100);天津市应用基础与前沿技术研究计划一般项目(15JCYBJC16600)

摘　　要：硬性渗出物是糖尿病视网膜病变(DR)的早期病症,是糖尿病性黄斑水肿的最主要表现,因此对硬性渗出物的准确检测具有重要的临床意义。提出一种基于背景估计和SVM分类器的眼底图像硬性渗出物检测方法。首先通过背景估计,得到包含亮目标的前景图;然后利用基于Kirsch算子的边缘信息确定硬性渗出物的候选区域,再移除视盘;最后对候选区域进行形状特征、直方图统计特征以及相位特征的提取,采用SVM对候选区域进行分类,完成硬性渗出物的精确提取。对DIARETDB1和HEI-MED公共数据库中共248幅眼底图像进行实验,图像水平达到灵敏度97.3%和特异性90%,病灶水平达到灵敏度84.6%和阳性预测值94.4%。实验表明,所提出的方法能够实现眼底图像中硬性渗出物的自动检测。Hard exudates（ HE） are early symptoms of diabetic retinopathy（ DR） and main symptom of macular edema. Hence,HE detection is very important for clinical diagnosis. In this paper,a new method based on background-estimation and SVM classifier for hard exudates detection is presented. Firstly,foreground map containing all bright objects is obtained by background-estimation. The HE candidates are gotten using the edge information based on Kirsch operator,and then the optic disc is removed. Finally,the shape features,histogram statistic features and phase features of the HE candidates are extracted before using the SVM classifier so that the accurate extraction of HE is obtained. Our method has been tested on the public databases of DIARETDB1 and HEI-MED. The experiment results show that the method＇s sensitivity is 97. 3% and the specificity is 90% at the image level,the mean sensitivity is 84. 6% and the mean predictive value is 94. 4% at the lesion level. The performance of the proposed method shows considerable efficiency for hard exudates detection.

关 键 词：硬性渗出物检测 糖尿病视网膜病变 背景估计 相位一致性 支持向量机(SVM) 

分 类 号：R318[医药卫生—生物医学工程]