周围型小细胞肺癌与腺癌、鳞癌的CT影像组学诊断  

CT-based radiomics in the diagnosis of peripheral small cell lung cancer,adenocarcinoma and squamous carcinoma

作　　者：汪靖婷 李婷[1] 张可萌 龙云 甘甜 廖美焱[1] WANG Jingting;LI Ting;ZHANG Kemeng;LONG Yun;GAN Tian;LIAO Meiyan(Dept.of Radiology,Zhongnan Hospital of Wuhan University,Wuhan 430071,Hubei,China)

机构地区：[1]武汉大学中南医院放射科,湖北武汉430071

出　　处：《武汉大学学报（医学版）》2024年第9期1057-1065,共9页Medical Journal of Wuhan University

基　　金：武汉大学中南医院临床研发项目(编号:lcyf202104)。

摘　　要：目的:探讨CT影像组学对周围型小细胞肺癌(SCLC)和非小细胞肺癌(NSCLC)的鉴别诊断价值。方法:回顾性分析2017年10月至2022年10月武汉大学中南医院周围型肺癌患者的临床及CT影像资料,根据病理类型分为SCLC-vs-SCC(鳞癌)组和SCLC-vs-ADC(腺癌)组,两组分别按照7∶3的比例分为训练集和验证集,使用最小绝对收缩和选择算子回归分析筛选影像组学特征,使用单因素和多因素Logistic回归筛选临床影像因素,分别建立影像组学、临床-影像及综合诊断模型。使用受试者操作特征(ROC)曲线、校准曲线和临床决策曲线评价验证模型诊断效能。结果:SCLC-vs-SCC组中,综合模型由3个临床影像因素和12个筛选的影像组学特征构成,训练集和测试集ROC曲线下面积(AUC)分别为0.977、0.937;SCLC-vs-ADC组中,综合模型由3个临床影像因素和17个筛选的影像组学特征构成,训练集和测试集AUC分别为0.979、0.961。两组校准曲线表明,训练集和验证集的校准曲线和理想曲线拟合较好;决策曲线分析显示,列线图在两组训练集和验证集均有较高的临床获益性。结论:CT影像组学联合临床影像资料的综合模型鉴别诊断周围型SCLC与ADC、SCC具有良好的临床价值。Objective:To explore the value of CT radiomics features combined with clinical-radiographic da-ta for differential diagnosis of peripheral small cell lung cancer(SCLC)and non-small cell lung cancer(NSCLC).Methods:The clinical and CT imaging data of peripheral lung cancer patients in our hospi-tal from October 2017 to October 2020 were retrospectively analyzed.All subjects were divided into the SCLC-vs-SCC group and SCLC-vs-ADC group based on the pathology type and then into training and validation sets according to the 7∶3 ratio.The least absolute shrinkage and selection operator re-gression were employed to select radiomics features generated from the training set.The rad-score was calculated for each lesion and the radiomic model was constructed.The clinical-radiographic model was constructed with clinical and imaging features selected by univariate and multivariate re-gression.Based on clinical,CT image,and radiomics features,the comprehensive model was con-structed,and the nomogram was further developed.The diagnostic performances of these models were assessed with the areas under the receiver operating characteristic curve,calibration curve,and decision curve analysis.Results:The comprehensive model indicated good discrimination ability,with AUCs of 0.977(training set)and 0.937(validation set)in the SCLC-vs-SCC group,0.979(training set)and 0.961(validation set)in the SCLC-vs-ADC group,respectively.Decision curve analysis showed that the comprehensive model had higher clinical value than the radiomics and clinical-radiographic models.Decision curve analysis demonstrated good clinical utility.Conclusion:The nomogram combining clinical,CT imaging,and radiomics features provided a potential noninva-sive method to distinguish peripheral SCLC from NSCLC,which could help clinicians make individu-al treatment decisions.

关 键 词：周围型肺癌 小细胞肺癌 体层摄影术 X线计算机 影像组学 诊断模型 

分 类 号：R734.2[医药卫生—肿瘤]