摘要
提出了一种糖尿病视网膜病变图像的血管分割方法.该方法通过对视网膜图像进行一系列形态学操作,获取精确的病变区域检测结果,然后将其从血管分割结果中去除,并将病变区域内断裂的血管根据梯度强度和方向相似的原则进行连接,得到最终的血管网络.在Stare公共数据库上进行实验结果表明:该方法获得的血管分割平均准确率达到94.67%,真阳性率达到74.96%;与未进行病变检测的分割结果相比,本文方法能获得较高准确率和真阳性率,分割结果更为精准.
A vessels segmentation method of diabetic retinopathy image was proposed in this paper. The accurate detection results of lesion areas were obtained by a series of morphological operations. Then,the lesion areas were removed from the result of vessels segmentation.Finally,the final net-work of vessels was obtained via connecting the broken vessels in the lesion areas according to the principle of gradient intensity and direction similar.The average accuracy rate of the segmentation method tested on Stare public database in this paper is 94.67%,and TPR value reaches 74.96%.In addition,the segmentation result is higher accuracy and TPR value compared with the segmentation method of non-lesion detection.
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2015年第S1期376-380,共5页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
湖北省自然科学基金资助项目(2014CFA130)
湖北省教育厅重点资助项目(D20141505)
武汉工程大学科学研究基金资助项目(K201401)
关键词
糖尿病视网膜病变
数学形态学
血管分割
图像增强
病变检测
diabetic retinopathy
mathematical morphology
vessels segmentation
image enhancement
lesion detection
