基于计算机视觉的糖尿病视网膜病变自动筛查系统

发布时间：2018-07-17 19:26

【摘要】：目的本研究旨在利用计算机视觉相关技术自动识别眼底影像中糖尿病视网膜病变(diabetic retinopathy,DR,以下简称糖网)的病变特征,开发能够用于DR筛查的计算机自动筛查系统。方法利用数学形态学和支撑向量机(support vector machine,SVM)分类技术设计出检测DR包括出血、渗出、微血管瘤等各类病变的算法,再根据DR的临床诊断标准,对眼底影像进行自动分级诊断,实现自动筛查。结果利用建立完成的糖网自动筛查系统对国际Messidor数据库进行了筛查判断,以经过专家认证的诊断结果作为判定标准。在总共1 200张眼底图中,系统的判定灵敏度(sensitivity)为93.8%,特异度(specificity)为94.5%,检测时间为9.83 s。结论基于计算机视觉算法开发的糖网自动筛查系统能准确、高效的完成眼科影像的糖网筛查工作,能大幅减少阅片医生的工作量和人为的主观性,具有很好的临床应用前景和社会效益。
[Abstract]:Objective to identify the features of diabetic retinopathy (diabetic retinopathy) in fundus image automatically by using computer vision related technique, and to develop a computer automatic screening system for Dr screening. Methods using mathematical morphology and support vector machine (support vector machine) classification technique, an algorithm was designed to detect various lesions of Dr, such as hemorrhage, exudation, microhemangioma and so on. According to the clinical diagnostic criteria of Dr, the fundus images were automatically classified and diagnosed. Automatic screening is realized. Results the international Messidor database was screened and judged by the established automatic screening system of sugar mesh, and the diagnostic results certified by experts were taken as the criterion. In a total of 1 200 fundus images, the system has a sensitivity of 93. 8, a specificity of 94. 5 and a detection time of 9. 83 s. Conclusion the sugar mesh automatic screening system based on the computer vision algorithm can accurately and efficiently complete the sugar mesh screening of ophthalmic images, and can greatly reduce the workload and artificial subjectivity of the radiographers. It has good clinical application prospect and social benefit.
【作者单位】：北京大恒图像视觉有限公司;首都医科大学附属北京同仁医院内分泌科;
【基金】：国家十二五重大专项基金(2013BAH19F02)~~
【分类号】：R587.2;R774.1

【参考文献】