基于内镜的人工智能辅助诊断系统在早期食管癌诊断中准确性的系统评价与Meta分析  

Accuracy of endoscopy-based artificial intelligence-assisted diagnostic system in the diagnosis of early esophageal cancer:A systematic review and meta-analysis

作　　者：洪子强 金大成 李宏超 成涛 白向豆 吴旭升 崔百强 苟云久[2] HONG Ziqiang;JIN Dacheng;LI Hongchao;CHENG Tao;BAI Xiangdou;WU Xusheng;CUI Baiqiang;GOU Yunjiu(The First Clinical Department of Gansu University of Traditional Chinese Medicine,Lanzhou,730000,P.R.China;Department of Thoracic Surgery,Gansu Provincial Hospital,Lanzhou,730000,P.R.China)

机构地区：[1]甘肃中医药大学第一临床医学院,兰州730000 [2]甘肃省人民医院胸外科,兰州730000

出　　处：《中国胸心血管外科临床杂志》2023年第9期1329-1336,共8页Chinese Journal of Clinical Thoracic and Cardiovascular Surgery

基　　金：甘肃省人民医院国家自然科学基金抚育项目(19SYPYB-28);甘肃省科技计划项目(20JR10RA388);甘肃省卫生健康行业科研管理项目(GSWSKY2020-50)。

摘　　要：目的系统评价基于内镜的人工智能(artificial intelligence,AI)辅助诊断系统在早期食管癌诊断中的准确性,为其诊断价值提供科学依据。方法计算机检索PubMed、EMbase、The Cochrane Library、Web of Science、万方数据库、维普数据库和中国知网数据库,搜索建库至2022年3月公开发表的基于内镜的AI辅助诊断系统诊断早期食管癌的相关文献。采用QUADAS-2评价工具对纳入研究进行质量评估。采用Stata 16、MetaDisc 1.4和RevMan 5.4软件对纳入的文献进行Meta分析。采用双变量混合效应回归模型计算AI辅助系统的综合诊断效能,并进行Meta回归分析以探索异质性的来源。结果共纳入17篇文献,其中13篇为回顾性队列研究、4篇为前瞻性队列研究。QUADAS-2评价结果显示纳入文献质量均较高。Meta分析结果显示,AI辅助系统诊断食管癌的合并灵敏性为0.94[95%CI(0.91,0.96)],特异性为0.85[95%CI(0.74,0.92)],阳性似然比为6.28[95%CI(3.48,11.33)],阴性似然比为0.07[95%CI(0.05,0.11)],诊断比值比为89[95%CI(38,208)],受试者工作特征曲线下面积为0.96[95%CI(0.94,0.98)]。结论AI辅助诊断系统对早期食管癌具有较高的诊断价值。然而,本Meta分析纳入研究大都为回顾性研究,因此需要进一步开展高质量的前瞻性研究来进行验证。Objective To systematically evaluate the accuracy of endoscopy-based artificial intelligence(AI)-assisted diagnostic systems in the diagnosis of early-stage esophageal cancer and provide a scientific basis for its diagnostic value.Methods PubMed,EMbase,The Cochrane Library,Web of Science,Wanfang database,VIP database and CNKI database were searched by computer to search for the relevant literature about endoscopy-based AI-assisted diagnostic systems for the diagnosis of early esophageal cancer from inception to March 2022.The QUADAS-2 was used for quality evaluation of included studies.Meta-analysis of the literature was carried out using Stata 16,Meta-Disc 1.4 and RevMan5.4 softwares.A bivariate mixed effects regression model was utilized to calculate the combined diagnostic efficacy of the AI-assisted system and meta-regression analysis was conducted to explore the sources of heterogeneity.Results A total of 17 articles were included,which consisted of 13 retrospective cohort studies and 4 prospective cohort studies.The results of the quality evaluation using QUADAS-2 showed that all included literature was of high quality.The obtained meta-analysis results revealed that the AI-assisted system in the diagnosis of esophageal cancer presented a combined sensitivity of 0.94(95%CI 0.91 to 0.96),a specificity of 0.85(95%CI 0.74 to 0.92),a positive likelihood ratio of 6.28(95%CI3.48 to 11.33),a negative likelihood ratio of 0.07(95%CI 0.05 to 0.11),a diagnostic odds ratio of 89(95%CI 38 to 208)and an area under the curve of 0.96(95%CI 0.94 to 0.98).Conclusion The AI-assisted diagnostic system has a high diagnostic value for early stage esophageal cancer.However,most of the included studies were retrospective.Therefore,further highquality prospective studies are needed for validation.

关 键 词：人工智能 内窥镜检查 早期食管癌 诊断 系统评价/META分析 

分 类 号：R735.1[医药卫生—肿瘤]