眼科

• 论著 • 上一篇    下一篇

糖尿病视网膜病变筛查智能检测与人工检测的一致性分析

王娟  陈淑娟  董方田  巩迪  唐轶  金蒙  赵雷 常海莲  刘凤桐  张健  任安娜  张帆   

  1.  100080 北京大恒普信医疗技术有限公司(王娟、陈淑娟、唐轶、金蒙、赵雷);100730 北京协和医院眼科(董方田、巩迪);101200 北京市平谷区医院眼科(常海莲、刘凤桐、张健、任安娜、张帆)
  • 收稿日期:2018-01-08 出版日期:2018-07-25 发布日期:2018-07-27
  • 通讯作者: 陈淑娟,Email: 452835085@qq.com

Comparison of computer automated screening system with ophthalmologist for grading of diabetic retinopathy

WANG Juan1, CHEN Shu-juan1, DONG Fang-tian2, GONG Di2, TANG Yi1, JIN Meng1, ZHAO Lei1, CHANG Hai-lian3, LIU Feng-tong3, ZHANG Jian3, REN An-na3, ZHANG Fan3   

  1. 1. Daheng Prust Medical Technology Co. Ltd, Beijing 100080, China; 2. Department of Ophthalmology, Peking Union Medical College Hospital, Beijing 100730, China; 3. Department of Ophthalmology, Beijing Pinggu Hospital, Beijing 101200, China
  • Received:2018-01-08 Online:2018-07-25 Published:2018-07-27
  • Contact: CHEN Shu-juan, Email: 452835085@qq.com

摘要:

目的  探讨糖尿病视网膜病变(DR)自动筛查系统与眼科医生判定DR评级的一致性。设计  诊断试验。研究对象  北京市平谷地区40岁以上糖尿病人群DR筛查时的1155例糖尿病患者眼底图片。方法  所有受试者均行免散瞳眼底照相,每眼分别以黄斑和视盘为中心拍摄两张45°彩色眼底像。智能检测采用北京大恒普信医疗技术有限公司研发的慧眼软件进行。检测时根据每眼的两张眼底像综合分析,给出判读结果。以人工检测结果作为金标准,分析智能检测的敏感度、特异度、漏诊率和误诊率以及Kappa检验的一致性分析。主要指标  Kappa值、敏感度、特异度。结果  智能检测与人工检测对DR有无判定结果的Kappa值为0.58,智能检测有无DR的敏感度为87.47%,特异度为74.39%。智能检测与人工检测对DR分期判读结果的Kappa值0.35。智能检测重度非增殖性DR的敏感度为86.36%,特异度为87.72%。结论  DR自动筛查系统可作为糖尿病人群DR筛查工作的辅助工具,以减轻医生工作量。(眼科, 2018, 27: 254-257)

关键词: 糖尿病视网膜病变/诊断, 眼底照相, 人工智能, 一致性

Abstract:

Objective Comparison of the sensitivity and specificity of computer automated screening system and ophthalmologist in grading diabetic retinopathy(DR). Design Diagnosis tests. Participants One thousand, one hundred and fifty-five cases of diabetic patients over 40 years old in Beijing Pinggu district were selected. Methods Two-field non-mydriatic fundus photographs center on macular and optic disc were performed. Grading of DR was carried out by two retinal images of one eye. Computer automated screening system carried out by the Software Huiyan from Daheng Prust Medical Technology Co. Ltd. Main Outcome Measure Kappa value, sensitivity, specificity. Results Computer automated screening system and ophthalmologist on the presence of DR with the result of the Kappa values was 0.58. The sensitivity and specificity of the presence of DR on computer automated screening system and ophthalmologist were 87.47% and 74.39% respectively. Computer automated screening system and ophthalmologist on the presence of severe non-proliferative DR with the result of the Kappa values was 0.35. The sensitivity and specificity of the presence of severe non-proliferative DR on computer automated screening system and ophthalmologist were 87.47% and 74.39% respectively. Conclusion These results show that computer automated screening system can efficiently screen retinal images from diabetes populations to reduce the workload of doctors. (Ophthalmol CHN, 2018, 27: 254-257)

Key words: