AI辅助诊断联合医师C-TIRADS分类对甲状腺结节的诊断效能分析  

Analysis of Diagnostic Efficacy of AI-Assisted Diagnosis Combined with C-TIRADS Classification for Thyroid Nodules

作　　者：徐可 石波 周春美 曾卓华 谢杨 刘家开 XU Ke;SHI Bo;ZHOU Chunmei;ZENG Zhuohua;XIE Yang;LIU Jiakai(Department of Ultrasound Medicine,The 2nd Affiliated Hospital of Chengdu Medical College,Nuclear Industry 416 Hospital,Chengdu Sichuan 610057,China)

机构地区：[1]成都医学院第二附属医院·核工业四一六医院超声医学科,四川成都610057

出　　处：《中国医疗设备》2022年第10期134-138,151,共6页China Medical Devices

基　　金：国家卫计委心血管病高危人群早期筛查与综合干预项目(GW-2015-005);四川省卫生健康委员会普及应用项目(17PJ437)。

摘　　要：目的探讨人工智能(Artificial Intelligence,AI)诊断系统联合中国(超声)甲状腺影像报告和数据系统(Chinese Thyroid Imaging Reporting and Data System,C-TIRADS)在甲状腺结节中的诊断价值。方法将符合纳入及排除标准的225个甲状腺结节作为研究对象,分别采用医师C-TIRADS分类诊断、AI诊断系统、AI诊断系统联合医师C-TIRADS分类诊断的方法,以粗针穿刺或手术病理诊断结果为金标准,绘制受试者工作特征(Receiver Operating Characteristic,ROC)曲线;比较3种诊断方法的ROC曲线下面积(Area Under Curve,AUC)、灵敏度、特异性、约登指数。结果联合诊断的AUC(0.952)高于C-TIRADS分类诊断(0.900)及AI诊断系统(0.829);联合诊断的特异性(95.74%)高于AI诊断系统(84.04%)和C-TIRADS分类诊断(82.98%);C-TIRADS分类诊断的灵敏度(96.95%)高于联合诊断(94.66%)及AI诊断系统(81.68%)。在≤10 mm结节诊断中,联合诊断的AUC(0.978)高于AI诊断系统(0.897)和C-TIRADS分类诊断(0.778);联合诊断的灵敏度(98.36%)高于C-TIRADS分类诊断(93.44%)和AI诊断系统(90.16%);联合诊断的特异性(97.30%)高于AI诊断系统(89.19%)和C-TIRADS分类诊断(62.16%)。在>10 mm结节诊断中,C-TIRADS分类诊断的AUC(0.982)高于联合诊断(0.931)和AI诊断系统(0.775);C-TIRADS分类诊断的灵敏度(100%)高于联合诊断(91.43%)和AI诊断系统(74.29%);C-TIRADS分类诊断的特异性(96.49%)高于联合诊断(94.74%)和AI诊断系统(80.7%),差异均具有统计学意义(P<0.05)。结论AI诊断系统联合医师C-TIRADS分类可提高甲状腺结节诊断的准确度及特异性,尤其对于≤10 mm结节,AI与医师联合诊断有更好的临床应用价值。对于>10 mm结节,AI诊断系统临床应用价值有限,医师C-TIRADS分类诊断灵敏度、特异性和准确度更高。Objective To explore the diagnostic value of artificial intelligence(AI)diagnostic system combined with Chinese thyroid imaging reporting and data system(C-TIRADS)in thyroid nodules.Methods The 225 cases of thyroid nodules that met the inclusion and exclusion criteria were selected as the research objects,and C-TIRADS classification diagnosis,AI diagnosis,and AI diagnosis combined with C-TIRADS classification diagnosis method were used for diagnosis respectively.The diagnosis results of coarse needle puncture or surgical pathological were used as the gold standard,and the receiver operating characteristic(ROC)curves were drawn.The area under the ROC curve(AUC),sensitivity,specificity,and Youden index of the three diagnosis methods were compared.Results The AUC of combined diagnosis(0.952)was higher than that of C-TIRADS classification diagnosis(0.900)and AI diagnosis(0.829),the specificity of combined diagnosis(95.74%)was higher than that of AI diagnosis(84.04%)and C-TIRADS classification diagnosis(82.98%),the sensitivity of C-TIRADS classification diagnosis(96.95%)was higher than that of combined diagnosis(94.66%)and AI diagnosis(81.68%).In the diagnosis of nodules≤10 mm,the AUC of combined diagnosis(0.978)was higher than that of AI diagnosis(0.897)and C-TIRADS classification diagnosis(0.778),the sensitivity of combined diagnosis(98.36%)was higher than that of C-TIRADS classification diagnosis(93.44%)and AI diagnosis(90.16%),the specificity of combined diagnosis(97.30%)was higher than that of AI diagnosis(89.19%)and C-TIRADS classification diagnosis(62.16%).In the diagnosis of nodules>10 mm,the AUC of C-TIRADS classification diagnosis(0.982)was higher than that of combined diagnosis(0.931)and AI diagnosis(0.775),the sensitivity of C-TIRADS classification diagnosis(100%)was higher than that of combined diagnosis(91.43%)and AI diagnosis(74.29%),the specificity of C-TIRADS classification diagnosis(96.49%)was higher than that of combined diagnosis(94.74%)and AI diagnosis(80.7%),and the differences were statistical

关 键 词：人工智能 辅助诊断系统 甲状腺结节 中国(超声)甲状腺影像报告和数据系统 

分 类 号：R736.1[医药卫生—肿瘤] R445.1[医药卫生—临床医学]