AI系统应用于老年人肺结核远程筛查的最佳临界值的确定  

Optimal cut-off value for remote screening of pulmonary tuberculosis in the elderly by artificial intelligence system

作　　者：吴世幸 黄宝杨[1] 梁佳佳 赖文胜[1] 杨俊 韦杨珏 周平[1] 马逸飞 周丽芳 WU Shixing;HUANG Baoyang;LIANG Jiajia;LAI Wensheng;YANG Jun;WEI Yangjue;ZHOU Ping;MA Yifei;ZHOU Lifang(Center for Disease Control and Prevention of Liuzhou City,Liuzhou 545007,Guangxi Province,China)

机构地区：[1]柳州市疾病预防控制中心,广西柳州545007

出　　处：《预防医学情报杂志》2023年第12期1457-1463,共7页Journal of Preventive Medicine Information

基　　金：广西壮族自治区卫生健康委员会自筹经费科研课题(项目编号:Z-B20221605);柳州市科技计划项目(项目编号:2022SB035)。

摘　　要：目的探索人工智能(Artificial Intelligence,AI)影像诊断系统(以下简称AI系统)在老年人中开展肺结核远程筛查的最佳临界值。方法纳入广西柳州市鹿寨县、鱼峰区、柳江区基层医疗卫生机构2022年6至12月期间应用AI系统进行肺结核远程筛查的老年人,收集其人口学特征与胸片诊断情况。通过配对χ^(2)检验考察不同诊断结果间的差异;通过Kappa检验考察不同诊断结果间的一致性;通过受试者工作特征曲线(Receiver Operator Characteristic Curve,ROC)确定AI系统肺结核疑似度的最佳临界值。除其他特殊说明外,均以P<0.05为差异有统计学意义。结果本次研究共纳入12354名研究对象,男女性别比为1∶1.22,年龄范围在65~98岁,中位年龄为70(67,75)岁(D=0.135,P<0.001)。基层医疗卫生机构影像医(技)师、结核病定点医院影像主任及AI系统诊断为疑似肺结核者分别为745人(6.03%)、565人(4.57%)、1603人(12.98%);非肺结核者分别为11609人(93.97%)、11789人(95.43%)和10751人(87.02%)。配对χ^(2)检验结果显示,基层医疗卫生机构影像医(技)师、定点医院影像主任及AI系统诊断结果间的差异均有统计学意义(P均<0.001);Kappa检验结果显示,基层医疗卫生机构影像医(技)师与定点医院影像主任诊断结果的Kappa值为0.791(P<0.001);基层医疗卫生机构影像医(技)师与AI系统诊断结果的Kappa值为0.578(P<0.001);定点医院影像主任与AI系统诊断结果的Kappa值为0.453(P<0.001)。AI系统中,所有研究对象的肺结核疑似度在0~0.996之间,中位值为0(0,0)(D=0.507,P<0.001);肺结核疑似度的ROC曲线有统计学意义(P<0.001),曲线下面积(Area Under the Cure,AUC)为0.950(H0:AUC=0.500,95%CI:0.938~0.962);当肺结核疑似度约为0.193时为最佳临界值,灵敏度约为93.50%,特异度约为92.20%,此时所对应的约登指数最大,为0.857。AI系统的阳性预测值约为33.13%。结论AI系统在广西柳州市老年人肺结核远程筛查中具�Objective To explore the optimal cut-off value of imaging diagnostic system for remote screening pulmonary tuberculosis by using artificial intelligence(AI)system among the elderly.Methods The elderly who had been remotely screened for tuberculosis by AI system in primary medical and health institutions in Luzhai County,Yufeng District and Liujiang District of Liuzhou,Guangxi Province from June 2022 to December 2022 were included.Demographic characteristics and chest X-ray diagnosis were collected.The McNemar Test was used to investigate the difference between different methods of diagnoses.The Kappa test was used to investigate the consistency of different diagnostic results.The receiver operator characteristic Curve(ROC)was used to determine the optimal critical value of AI system.P<0.05 was considered statistically significant unless otherwise specified.Results A total of 12354 subjects were included in this study,with a male to female ratio of 1∶1.22.The age of the elderly ranged from 65 to 98 years,and the median age of 70(67,75)years old(D=0.135,P<0.001).Among all diagnosis of the medical imaging practitioners in primary medical institutions,the director of imaging at designated tuberculosis hospitals and AI system,the suspected tuberculosis cases were 745(6.03%),565(4.57%)and1603(12.98%),respectively;the non-pulmonary individuals were 11609(93.97%),11789(95.43%)and 10751(87.02%),respectively.There were statistically significant differences among the diagnostic results of the medical imaging practitioners in primary medical institutions,the director of imaging at designated tuberculosis hospitals and the AI system(P<0.001).The Kappa values between the medical imaging practitioners in primary medical institutions and the director of imaging at designated tuberculosis hospitals,the medical imaging practitioners in primary medical institutions and the AI system,the director of imaging at designated tuberculosis hospitals and the AI system were 0.791(P<0.001),0.578(P<0.001),0.453(P<0.001),respectively.In the

关 键 词：人工智能 肺结核 老年人 筛查 

分 类 号：R521[医药卫生—内科学]