人工智能辅助诊疗系统在肺小结节诊断中的应用——附1650例临床分析  

Application of artificial intelligence-assisted diagnosis and treatment system in the diagnosis of pulmonary nodules:a clinical analysis of 1650 cases

作　　者：王峥[1] 张洪[2] 吕军 罗翔凤 孙大强 Wang Zheng;Zhang Hong;Lv Jun;Luo Xiangfeng;Sun Daqiang(Department of Thoracic Surgery,Tianjin Chest Hospital Affiliated to Tianjin University,Tianjin,300222,China;Department of Radiology,Tianjin Chest Hospital Affliated to Tianjin University,Tianjin,30222,China;LinkDoc Science and Technoo gy Co.Ltd.,Beijing,100080,China)

机构地区：[1]天津大学附属天津市胸科医院胸外科,天津300222 [2]天津大学附属天津市胸科医院影像科,天津300222 [3]零氪科技有限公司,北京100080

出　　处：《中华胸心血管外科杂志》2023年第8期466-471,共6页Chinese Journal of Thoracic and Cardiovascular Surgery

基　　金：天津市科学技术委员会重大科研攻关项目(18ZXZNSY00400);天津市医学重点专科建设项目(TJYXZDXK-018A)。

摘　　要：目的探讨人工智能辅助诊疗系统在肺小结节诊断中的临床应用价值。方法回顾性分析2015年1月1日至2022年10月1日在天津大学附属天津市胸科医院胸外科手术治疗的1650例肺部小结节患者CT影像资料,对比分析术前AI辅助诊断的恶性概率与术后病理诊断结果。分析AI智能辅助诊疗系统对肺小结节的良恶性的预测能力。结果1650例肺小结节患者术后病理诊断分别为肺癌1391例(84.3%)和肺良性病变259例(15.7%)。两组患者术前AI诊断的平均恶性概率分别为(78.85±22.91)%和(54.91±28.68)%。对本组病例进行AI恶性概率分布的工作特征曲线分析,结果显示曲线下面积为0.768。当恶性概率为81.3%时,敏感度与特异性之和最高,敏感度为0.620,特异性为0.815。分层分析结果显示,AI智能系统对于肺结节诊断的准确性随直径增大升高,0~1cm、1~2cm和2~3cm大小的肺结节诊断的R0C曲线下面积分别为0.717、0.769和0.804。结论人工智能辅助诊疗系统计算出的恶性概率可以用于术前辅助鉴别肺小结节的良恶性情况,准确性与肺小结节直径相关。Objective To evaluate the effectiveness of the artificial intelligence-assisted diagnosis and treatment system in distinguishing benign and malignant lung nodules.Methods Clinical data of 1650 patients with pulmonary nodules admit-ted to the Tianjin Chest Hospital Affiliated to Tianjin University from January 2015 to October 2022 were retrospectively ana-lyzed,which included a total of 1650 nodules.The malignant probability was compared to and analyzedpredict the benign and malignant nodules.Results A total of 1650 nodules were included,which were divided into a malignant tumor group(n=1391,84.3%)and a benign lesion group(n=259,15.7%).Between the malignant tumor group and the benign lesion group,the malignant probability was significantly different,and the malignant probability could better distinguish malignant nodules and benign lesions[(78.85±22.91)%vs.(54.91±28.68)%,P<0.001].The area under the curve(AUC)was 0.768.The critical value of malignant probability for diagnosis of lung cancer was 81.3%with a sensitivity of 0.620 and speci-ficity of 0.815.Stratified analysis results showed that the accuracy of the Al intelligent system for diagnosing pulmonary nodules with the sizes of 0-1 cm,1-2 cm and 2-3 cm was also increased,and the areas under R0C curve were 0.717,0.769 and 0.804,respectively.Conclusion The malignant probability of lung nodules calculated by the artificial intelligence-assisted diagnosis and treatment system can be used in the asssment of the preoperative benign and malignant lung nodules.

关 键 词：人工智能 肺结节 辅助诊断 

分 类 号：R563[医药卫生—呼吸系统] R319[医药卫生—内科学]