8论著影像研究与医学应用2024年2月第8卷第3期基于深度学习人工智能在肺结节定性诊断中的临床应用研究叶文卫1,2，刘碧华1,3（通信作者），郭天畅2（1广东医科大学研究生学院广东东莞523808）（2东莞市松山湖中心医院放射科广东东莞523326）（3南方医科大学附属东莞医院<东莞市人民医院>放射科广东东莞523039）ClinicalapplicationofdeeplearningartificialintelligenceinqualitativediagnosisofpulmonarynodulesYEWenwei1,2,LIUBihua1,3(Correspondingauthor),GUOTianchang21GraduateSchoolofGuangdongMedicalUniversity,Dongguan,Guangdong523808,China;2DepartmentofRadiology,DongguanSongshanLakeCentralHospital,Dongguan,Guangdong523326,China;3DepartmentofRadiology,DongguanHospitalAffiliatedtoSouthernMedicalUniversity,Dongguan,Guangdong523039,China【Abstract】ObjectiveToexploretheclinicalapplicationofdeeplearningartificialintelligenceinthequalitativediagnosisofpulmonarynodules.MethodsAretrospectiveanalysiswasconductedonthedataof325pulmonarynodulesfrom230patientswhomettheinclusioncriteriaatDongguanSongshanLakeCentralHospitalfromJanuary2020toDecember2022.TheyweredividedintogroupA(doublereading),groupB(AIreadingalone),andgroupC(doublereading+AIcomprehensivereading)accordingtothereadingmethod.Usingpathologyasthegoldstandard,thereadingtime,sensitivity,specificity,positivepredictivevalue,negativepredictivevalue,andaccuracyof325lungnodulesdetectedineachgroupwerecountedandcompared.ResultsThefilmreadingtimeinGroupAwas(14.37±2.12)minutes,thefilmreadingtimeinGroupBwas(1.34±0.12)minutes,andthefilmreadingtimeinGroupCwas(8.34±1.26)minutes.ThefilmreadingtimeofGroupBwasstatisticallyshorterthanthatofGroupAandGroupC,andthefilmreadingtimeofGroupCwasstatisticallyshorterthanthatofGroupA,withastatisticallysignificantdifference(P＜0.05).Thesensitivity,specificity,positivepredictivevalue,negativepredictivevalue,andaccuracyofGroupAdiagnosiswere85.97%,36.17%,88.85%,30.36%,and78.76%,respectively,andtheindicatorsofGroupBdiagnosiswere91.01%,42.55%,90.36%,44.44%,and84.00%,respectively,andindicatorsofgroupCdiag...
