人工智能对早期肺癌结节良恶性鉴别能力探究  

Ability of artificial intelligence in differentiating benign and malignant nodules of early lung cancer

作　　者：卢辉 李峰[1] 胡庆 庞中好 陈圣杰[1] 蒋志华[1] 顾添一 基红云[1] LU Hui;LI Feng;HU Qing;PANG Zhong-hao;CHEN Sheng-jie;JIANG Zhi-hua;GU Tian-yi;JI Hong-yun(Thoracic Surgery Department,the Affiliated Hospital of Jiangsu University,Zhenjiang,Jiangsu 212001,China)

机构地区：[1]江苏大学附属医院胸外科,江苏镇江212001

出　　处：《临床肺科杂志》2021年第12期1870-1875,共6页Journal of Clinical Pulmonary Medicine

基　　金：吴阶平医学基金(No.320.6750.2021-12-3);镇江市社会发展重大研发项目(No.SH2019063);江苏大学人才启动基金(No.jdfyRC2017002)。

摘　　要：目的探讨人工智能(Artificial Intelligence,AI)诊断早期肺癌结节的良恶性鉴别的能力。方法本研究采用回顾性分析的方法,在2015年1月-2020年10月的病历资料中收集病理确诊为肺恶性肿瘤患者的肺结节CT图像资料382例及良性肺结节的CT图像资料156例,共538例,分别采用AI阅片和医师阅片两种方法判读所选取的所有患者的胸部CT,对肺结节良恶性鉴别结果进行统计学分析和ROC曲线对比,并用AI二次阅片,采用Kappa检验评定AI阅片的可靠性。结果本研究设计的优化算法的AI阅片读取孤立实性结节(SN)和磨玻璃结节(GGN)CT图像的敏感度、阳性预测值、准确率皆略高于医师诊断的敏感度、阳性预测值、准确率。AI对两类结节两次重复阅片的Kappa值接近1。AI阅片与医师阅片诊断肺结节良恶性的AUC分别为0.934、0.888。结论AI阅片诊断肺结节良恶性的阳性预测值及准确率比临床医师阅片更高,且两次AI阅片结果高度一致。AI在一定程度上可以辅助放射科医师进行SPN良恶性的诊断。Objective To investigate the ability of artificial intelligence(AI)in the differential diagnosis of benign and malignant lung cancer nodules.Methods In this study,a retrospective analysis method was used to collect the CT image data of 382 patients with lung malignant tumor and 156 patients with benign pulmonary nodules from January 2015 to October 2020.Their chest CT was analyzed by AI film reading and doctor reading,and statistical analysis and ROC curve were used to compare the results of benign and malignant pulmonary nodules,and the reliability of AI film reading was evaluated by Kappa test.Results The sensitivity,positive predictive value and accuracy of AI in reading CT images of solitary solid nodules(SN)and ground glass nodules(GGN)were slightly higher than those of physicians.The Kappa value of AI was close to 1.The AUC of AI and physician were 0.934 and 0.888 respectively.Conclusion The positive predictive value and accuracy of AI film reading in the diagnosis of benign and malignant pulmonary nodules are higher than that of clinicians,and the results of twice AI film reading are highly consistent.AI can assist radiologists in the diagnosis of benign and malignant SPN to a certain extent.

关 键 词：人工智能 肺癌 孤立性肺结节 实性肺结节 磨玻璃样结节 

分 类 号：R734.2[医药卫生—肿瘤]