CT影像组学联合临床风险因素预测非小细胞肺癌纵隔内隐匿性淋巴结转移的研究  

Prediction of mediastinal occult lymph node metastasis in non-small cell lung cancer(NSCLC)by CT imaging combined with clinical risk factors

作　　者：李悦 胡丛阳 许晓燕[1] 周永[1] LI Yue;HU Congyang;XU Xiaoyan;ZHOU Yong(Diagnostic Imaging Center,Xinjiang Medical University Affiliated Cancer Hospital,Urumqi 830011,China)

机构地区：[1]新疆医科大学附属肿瘤医院影像诊断中心,乌鲁木齐830011

出　　处：《新疆医科大学学报》2025年第1期75-81,共7页Journal of Xinjiang Medical University

基　　金：新疆维吾尔自治区自然科学基金(2021D01C392)。

摘　　要：目的探讨基于CT影像组学及临床风险因素术前预测非小细胞肺癌(Non-small cell lung cancer,NSCLC)纵隔内隐匿性淋巴结转移的价值。方法收集2019年1月-2023年12月在新疆医科大学附属肿瘤医院放射科行胸部增强CT并行肺癌根治性手术且病理证实为非小细胞肺癌,伴或不伴纵隔淋巴结(Mediastinal lymph nodes,MLN)转移的患者147例。患者按照7∶3的比例随机分为训练组102例(MLN阳性29例、阴性73例)和测试组45例(MLN阳性12例、阴性33例)。对所有临床风险因素进行单因素及多因素逻辑回归分析,建立预测非小细胞肺癌纵隔内隐匿性淋巴结转移的临床模型。在薄层增强CT图像手动绘制感兴趣区(Region of Interest,ROI),提取影像组学特征。使用Mann-Whitney U检验、Spearman秩相关系数、最大相关最小冗余(Minimum redundancy maximum relevance,mRMR)及最小绝对收缩和选择算法(Least Absolute Shrinkage and Selection Operator,LASSO)进行特征降维和特征选择。基于最佳影像组学特征建立支持向量机(SVM)模型。结合临床风险因素构建联合模型。结果临床模型、影像组学模型及临床-影像组学联合模型在训练组的曲线下面积(Area under curve,AUC)分别为0.740、0.863、0.887;测试组的AUC分别为0.699、0.712、0.758;在3组模型中临床-影像组学联合模型的预测效能最佳,其AUC在训练组和测试组分别为0.887、0.758。结论基于CT影像组学特征及临床风险因素构建的模型在预测非小细胞肺癌纵隔淋巴结转移中有一定的价值,临床-影像组学联合模型的诊断效能优于单一模型。Objective Investigated the significance of preoperative estimation of hidden lymph node spread in the mediastinum of non-small cell lung cancer(NSCLC)using CT imaging and clinical risk factors.Methods A total of 147 patients diagnosed with NSCLC with or without Mediastinal lymph node(MLN)metastasis who received thoracic thin-layer-enhanced CT scans followed by radical lung cancer surgery and confirmed NSCLC in the hospital from January 2019 to December 2023 were included in the study.The patients were randomly divided into the training group about 102 cases(29 MLN positive cases,73 MLN negative cases)and the test group about45 cases(12 MLN positive cases,33 MLN negative cases)accord-ing to a ratio of 7∶3.Single-factor and multifactor Logistic regression analyses were performed for all clinical risk factors to establish a clinical model for predicting latent mediastinal lymph node metastasis in NSCLC.ROIs were manually outlined in thin-layer CT images to extract imaging histologic features.The Mann-Whitney U test,Spearman's rank correlation coefficient,maximum correlation minimum redun-dancy(mRMR),and the least absolute shrinkage and selection operator(LASSO)were utilized to reduce the dimensionality of features and select relevant features.A support vector machine(SVM)model was built based on the best imaging histologic features.And the joint model was constructed by combining clinical risk factors.Results The AUCs of the clinical model,the imaging histology model and the com-bined clinical-imaging histology model in the training group were 0.740,0.863 and 0.887,respectively;the AUCs of the test group were 0.699,0.712 and 0.758,respectively;the prediction efficacy of the com-bined clinical-imaging histology model was the best among the 3 models.Its AUC in the training group and testing group were 0.887 and 0.758,respectively.Conclusion The model based on CT imaging features and clinical risk factors has a certain value in predicting mediastinal lymph node metastasis of NSCLC.The diagnostic efficacy of the combi

关 键 词：非小细胞肺癌 淋巴结转移 影像组学 预测 

分 类 号：R734.2[医药卫生—肿瘤]