微动脉瘤与出血自动检测的NPDR图像分型方法  

Grading method for non-proliferative diabetic retinopathy images based on microaneurysms and hemorrhages automatic detection

作　　者：郑绍华[1] 潘林[1] 陈健[1] 郭健[2] 余轮[1] 

机构地区：[1]福州大学物理与信息工程学院,福州350108 [2]福建医科大学附属第一医院眼科中心,福州350005

出　　处：《仪器仪表学报》2014年第1期59-67,共9页Chinese Journal of Scientific Instrument

基　　金：国家自然基金(60827002);福建省自然基金(2012J05116);福建省教育厅(JA13037)资助项目

摘　　要：为构建糖尿病视网膜病变(DR)自动筛查系统,依据新的糖尿病视网膜病变国际临床分型标准,提出了一种基于微动脉瘤(MAs)和视网膜内出血(HAs)自动检测的非增生性糖尿病视网膜病变(NPDR)图像分型方法。首先在红色通道上利用自适应阈值、形态学技术及区域主动轮廓模型对视盘中心进行精确定位;再者,在绿色通道预处理及现有算法改进的基础上,提出了一种新的基于Radon变换的方法对HAs和MAs进行识别,有效地抑制了血管或背景纹理的干扰,算法不依赖于血管的分割,也无需多尺度匹配滤波器。实验选取了120幅高分辨率的NPDR图像,对HAs和MAs的检测均得到较高的灵敏度和特异性,在各分型上准确率都达到了85%以上,不存在漏诊的情况。算法稳定,具有较高的实用价值。In order to develop an automated screening system for diabetic retinopathy （DR） ,a grading method for non- proliferative diabetic retinopathy （NPDR） images based on microaneurysms （MAs） and hemorrhages （HAs） automatic detection is proposed according to the new international clinical diabetic retinopathy classification standard. Firstly, in the red channel of the retinal image, the adaptive thresholding, morphological technique and region-based active contour model are adopted to accurately locate the optic disk center. Secondly, after pre-processing in the green channel, a novel method based on the Radon transform is presented to recognize the true MAs and HAs lesions. The Radon based method effectively suppresses the interference of blood vessels or background texture ; and the algorithm neither relies on the blood vessel segmentation,no requires the multi-scale matching filter. In an experiment, 120 high resolution color images were used to evaluate the performance of the developed method. The experiment results show that the proposed system achieves high sensitivity and specificity for the detection of both MAs and HAs, and the grading accuracy reaches above 85% for the NPDR images without a missed diagnosis. The proposed algorithm is stable and has high practical value.

关 键 词：糖尿病视网膜病变 微动脉瘤 出血 视盘 RADON变换 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TH79[自动化与计算机技术—计算机科学与技术]