摘要
目的:探讨基于增强CT影像组学特征和临床独立危险因素构建的联合模型及其列线图在术前预测进展期胃癌(AGC)周围神经侵犯(PNI)中的价值。方法:回顾性分析171例AGC患者的CT图像和临床资料。将171例患者按7:3的比例随机分为训练组119例(PNI阳性83例,阴性36例)和验证组52例(PNI阳性37例,阴性15例)。依次使用Spearman相关性分析及绝对收缩与选择算子(LASSO)对增强CT静脉期图像上提取的组学特征进行降维和筛选,并建立影像组学标签(V-Radscore)。使用单因素分析比较PNI阳性组与阴性组之间的V-Radscore和术前临床指标值,将差异具有统计学意义的指标纳入多因素logistic回归分析,得到PNI相关的独立危险因素,同时构建影像组学模型(V)、临床模型(C)和组合模型(V+C),并在训练组构建组合模型的列线图。采用受试者工作特征曲线(ROC)的曲线下面积(AUC)、敏感度、特异度和符合率来评价模型的诊断效能,使用校准曲线评价列线图模型在训练组和验证组中的拟合程度,使用决策曲线分析(DCA)来评价列线图的临床应用价值。结果:PNI的独立危险因素包括V-Radscore、CT报告的肿瘤部位、T分期和N分期(P均<0.05)。PNI阳性组的V-Radscore高于阴性组(Z=5.536,P<0.001)。在验证组中,列线图模型预测PNI的AUC值为0.865,显著高于临床模型(AUC=0.786,χ^(2)=2.108,P=0.035)和影像组学模型(AUC=0.681,χ^(2)=2.083,P=0.037),其预测PNI的敏感度、特异度和符合率分别为0.838、0.800和0.827。校准曲线显示列线图在训练组(χ^(2)=5.846,P=0.661>0.05)及验证组(χ^(2)=8.170,P=0.417>0.05)中的预测概率与实际概率的一致性良好。DCA显示模型具有良好的临床应用价值。结论:临床-影像组学列线图模型在AGC患者PNI的术前预测方面具有可行性,有望帮助临床医师优化术前决策。
Objective:To explore the value of nomogram model based on contrast-enhanced CT radiomics and clinical features in preoperative prediction of perineural invasion(PNI)of advanced gastric cancer(AGC).Methods:The CT images and clinical data of 171 patients with AGC were retrospectively analyzed.All patients were randomly divided into 119 patients(including 83 patients with PNI and 36 patients without PNI)in the training cohort,and 52 patients(including 37 patients with PNI and 15 patients without PNI)in the testing cohort at a rate of 7:3.The spearman correlation analysis(SPM)and least absolute shrinkage and selection operator(LASSO)were used for dimension reduction and selection of the radiomics features extracted from contrast-enhanced CT venous images,and the selected features were used to calculate the Radscore of venous images(V-Radscore).Univariate analysis was used to compare V-Radscore and the preoperative clinical indicators between PNI positive and PNI negative groups,and the statistically significant indicators were incorporated into multivariate logistic regression to obtain the independent risk factors of PNI,the radiomics model(V),clinical model(C)and combined model(C+V)were built at the same time,and then a nomogram was deve-loped to predict PNI in training group.The area under curve(AUC)of receiver operating characteristic(ROC),sensitivity,specificity and accuracy were used to evaluate the diagnostic efficiency of nomogram model.The calibration curves and decision curve analysis(DCA)were used to assess the calibration and clinical usefulness of nomogram model,respectively.Results:The independent risk factors of PNI included V-Radscore,CT-reported tumor site,T-stage,and N-stage(P<0.05),and the V-Radscore in PNI positive group was significantly higher than that in negative group(Z=5.536,P<0.001).In the testing cohort,the AUC of nomogram model for predicting PNI was 0.865,which was significantly higher than that of the clinical model(AUC=0.786,χ^(2)=2.108,P=0.035)and the radiomics model(AUC=0.681,χ^(2)
作者
黄钰迅
李瑞
张宝腾
牛猛
刘钊
郭顺林
HUANG Yu-xun;LI Rui;ZHANG Bao-teng(The First Clinical Medical College,Lanzhou University,Lanzhou 730000,China)
出处
《放射学实践》
CSCD
北大核心
2022年第12期1548-1554,共7页
Radiologic Practice
基金
国家自然基金资助项目(81960323)。
