基于人工智能的CT影像预测肺腺癌浸润性:Meta分析  

Predicting the invasiveness of lung adenocarcinoma using CT images based on artificial intelligence:a meta-analysis

作　　者：黄文君 周秀秀 周陶胡 葛艳明[2] 董鹏[2] 范丽 HUANG Wen-jun;ZHOU Xiu-xiu;ZHOU Tao-hu(School of Medical Imaging,Weifang Medical University,Shandong 261000,China)

机构地区：[1]潍坊医学院医学影像学院,山东261000 [2]潍坊医学院附属医院医学影像科,山东261000 [3]海军军医大学第二附属医院放射诊断科,上海200003

出　　处：《放射学实践》2022年第11期1359-1366,共8页Radiologic Practice

基　　金：国家自然科学基金(81871321,81930049,82171926);上海市青年科技英才扬帆计划(20YF1449000);上海长征医院2020年度院创新型临床研究项目(2020YLCYJ-Y24);上海长征医院金字塔人才工程。

摘　　要：目的:评价基于人工智能(AI)的CT影像预测肺腺癌浸润性的诊断性能。方法:通过PubMed、Embase、Cochrane图书馆、Web of Science、中国知网、SinoMed、万方和维普等数据库,检索2011年1月1日-2021年6月30日公开发表的基于AI的CT影像预测肺腺癌浸润性的所有文章。以诊断试验的纳入和排除标准筛选文章并提取关键特征信息。通过软件MetaDiSc 1.4和Stata 16.0进行Meta分析。以浸润性腺癌作为阳性结果、非浸润性腺癌作为阴性结果,计算合并后的敏感度、特异度、阳性似然比、阴性似然比和诊断比值比,绘制集成受试者操作特征(SROC)曲线并得出曲线下面积(AUC)。基于Meta回归探究异质性的来源。使用敏感性分析验证Meta分析结果的可靠性。结果:共12项研究入选,包括4066例患者的4325枚肺结节,磨玻璃结节占97%。研究间存在异质性,故指标合并采取随机效应模型。合并后的敏感度、特异度、阳性似然比、阴性似然比和诊断比值比分别为0.86、0.82、4.55、0.19和28.31,AUC为0.9110。Meta回归结果表明AI算法的不同是异质性的来源,敏感性分析显示Meta分析结果的可靠度高。结论:基于AI的CT影像对肺腺癌浸润性有较高的预测能力,可向医生提供更确切的诊疗信息,进而优化患者的治疗方案。Objective:To evaluate the diagnostic performance of artificial intelligence(AI) using CT images for predicting lung adenocarcinoma invasiveness.Methods:Systematically retrieved all published articles on AI-based CT imaging for predicting lung adenocarcinoma invasiveness from January 1,2011,to June 30,2021,through PubMed,Embase,Cochrane Library,Web of Science,CNKI,SinoMed,Wanfang Database,and VIP series database.Articles were screened with inclusion and exclusion criteria of diagnostic tests and characteristic information was extracted.Meta-analysis was performed by using Metadisc 1.4 software and Stata 16.0 software.Sensitivity,specificity,positive/negative likelihood ratio,and diagnostic ratio were measured for the combined results with invasive adenocarcinoma as the positive result and non-invasive adenocarcinoma as the negative result.And the summary receiver operating characteristic(SROC) curve was plotted and the area under the curve(AUC) was derived.The sources of heterogeneity was explored based on Meta-regression.Sensitivity analysis was used to validate the reliability of the Meta-analysis results.Results:Twelve studies were enrolled,including 4325 pulmonary nodules in 4066 patients,with ground glass nodules accounting for 97%.The indexes were combined in a random-effects model due to heterogeneity existed in those studies.The combined sensitivity,specificity,positive likelihood ratio,negative likelihood ratio,and diagnostic ratio were 0.86,0.82,4.55,0.19,and 28.31,respectively,with an AUC of 0.9110.Meta-regression analysis indicated that different AI algorithms may be the source of heterogeneity,and sensitivity analysis shows high reliability of Meta-analysis results.Conclusion:AI-based CT imaging has a robust predictive power for lung adenocarcinoma invasiveness,which can provide doctors with more precise diagnosis and treatment information,and thus optimize the therapeutic schedule.

关 键 词：人工智能 体层摄影术 X线计算机 腺癌 肺肿瘤 META分析 

分 类 号：R814.42[医药卫生—影像医学与核医学] R734.2[医药卫生—放射医学] R-05[医药卫生—临床医学]