基于深度学习技术的获得性免疫缺陷综合征合并巨细胞病毒性视网膜炎诊断模型的构建  

A deep learning assisted diagnostic system for the detection of cytomegalovirus retinitis in patients with AIDS

作　　者：杜葵芳 董力[2] 李小娜[1] 孔文君 谢连永[1] 董宏伟[1] 黄晓婕 魏文斌[2] Du Kuifang;Dong Li;Li Xiaona;Kong Wenjun;Xie Lianyong;Dong Hongwei;Huang Xiaojie;Wei Wenbin(Department of Ophthalmology,Beijing Youan Hospital,Capital Medical University,Beijing 100069,China;Beijing Tongren Eye Center,Beijing key Laboratory of Intraocular Tumor Diagnosis and Treatment,Beijing Ophthalmology&Visual Sciences Key Lab,Medical Artificial Intelligence Research and Verification Key Laboratory of the Ministry of Industry and Information Technology,Beijing Tongren Hospital,Capital Medical University,Beijing 100730,China;Department of Infectious Diseases,Beijing Youan Hospital,Capital Medical University,Beijing 100069,China)

机构地区：[1]首都医科大学附属北京佑安医院眼科、北京市感染性眼病诊疗中心,北京100069 [2]首都医科大学附属北京同仁医院、北京同仁眼科中心、眼内肿瘤诊治研究北京市重点实验室、北京市眼科学与视觉科学重点实验室、医学人工智能研究与验证工信部重点实验室,北京100730 [3]首都医科大学附属北京佑安医院感染中心一科,北京100069

出　　处：《中国医学前沿杂志（电子版）》2023年第6期10-14,共5页Chinese Journal of the Frontiers of Medical Science(Electronic Version)

基　　金：国家自然科学基金(82220108017,82141128);首都卫生发展科研专项(首发2020-1-2052);北京市科委科技计划项目(Z201100005520045,Z181100001818003)。

摘　　要：目的利用巨细胞病毒视网膜炎患者(cytomegalovirus retinitis,CMVR)的超广角眼底照片构建出深度学习系统,用于智能筛查。方法选取2017年9月至2020年10月15日就诊于北京佑安医院眼科815例人类免疫缺陷病毒感染者的6300张欧堡超广角(ultra-wide-field,UWF)图像构建一个深度学习系统,识别活动性CMVR、非活动性CMVR和非CMVR。同时采用其中的前瞻性数据集进一步验证该系统。通过计算曲线下面积、准确性、敏感度和特异度来评估系统的性能。结果在内部交叉验证和前瞻性验证中,深度学习模型检测活动性CMVR和非CMVR的曲线下面积分别为0.988(95%CI:0.978~0.998)和0.951(95%CI:0.947~0.955);准确性分别为0.955(95%CI:0.940~0.969)和0.942(95%CI:0.937~0.946)。该模型还显示出区分活动性CMVR与非CMVR和非活动性CMVR的能力,以及识别活动性CMVR和非活动性CMVR的能力(两个独立数据集中的所有曲线下面积均>0.886)。结论本研究使用超广角眼底图像构建的深度学习模型显示了可靠的获得性免疫缺陷综合征合并CMVR诊断性能。Objective To construct a deep learning(DL)system by using ultra-wide-field retinal photographs of patients with cytomegalovirus retinitis(CMVR)for intelligent screening purposes.Methods We developed and validated a DL-based system to identify active CMVR,inactive CMVR,and non-CMVR in 6300 UWF images from 815 patiens with acquired immunodeficiency syndrome(AIDS)from September 2017 to October 15th 2020.Then the system was further validated in a prospective data set.The performance of the system was evaluated by calculating the area under curve(AUC),accuracy,sensitivity,and specificity.Results The DL system achieved AUCs of 0.988(95%CI:0.978-0.998)and 0.951(95%CI:0.947-0.955)for detecting active CMVR from non-CMVR in the internal cross validation and prospective validation,respectively.The accuracy reached 0.955(95%CI:0.940-0.969)and 0.942(95%CI:0.937-0.946)for detecting active CMVR from non-CMVR in the internal cross validation and prospective validation,respectively.The system also showed the ability in differentiating active CMVR from non-CMVR and inactive CMVR as well as identifying active CMVR and inactive CMVR from non-CMVR(all AUCs in the three independent data sets>0.886).Conclusions Our DL system using UWF fundus images showed reliable performance for detecting AIDS-rated CMVR.

关 键 词：深度学习 获得性免疫缺陷综合征 巨细胞病毒性视网膜炎 超广角眼底成像 诊断 

分 类 号：R512.91[医药卫生—内科学] R774.1[医药卫生—临床医学]