基于CT影像组学结合临床影像特征预测局部晚期鼻咽癌诱导化疗疗效

Prediction of the efficacy of induction chemotherapy in locally advanced nasopharyngeal carcinoma based on CT radiomics combined with clinic-radiological features

目的:探讨基于增强CT的影像组学结合临床影像特征的列线图在预测局部晚期鼻咽癌(LA-NPC)患者诱导化疗(ICT)疗效中的价值。方法:回顾性分析2014年7月至2022年3月178例LA-NPC(Ⅲ、Ⅳ期)患者的临床及CT图像资料,以7:3随机将患者分为训练组(n=125)和测试组(n=53)。采用3D-Slicer勾画容积感兴趣区(VOI)并用Pyradiomics包提取特征。使用单-多因素Logistic回归选择临床预测因子。采用最小绝对收缩与选择算法(LASSO)筛选组学特征,最后通过多变量Logistic回归构建临床、影像组学及联合模型,并绘制列线图。以受试者工作特征曲线(ROC)的曲线下面积(AUC)评估和比较三种模型的预测效能。应用决策曲线(DCA)观察列线图的临床净获益。结果:Logistic回归分析结果显示T分期(OR=0.45,P=0.004)、癌灶强化程度(OR=2.706,P=0.038)、血小板/淋巴细胞比值(PLR)(OR=0.289,P=0.024)是ICT疗效的临床预测因子,基于以上3者构建临床模型。基于22个与ICT疗效显著相关的组学特征构建影像组学模型。ROC曲线分析结果显示,联合模型的预测效能最佳;训练组中,联合模型、临床模型、影像组学模型的AUC分别为0.821、 0.732、0.798;验证组中,三者的AUC分别为0.836、0.793、0.779。DCA分析进一步表明,列线图模型对比单纯组学模型,其人群净获益率更高。结论:基于增强CT的影像组学联合传统临床影像特征的列线图能直观、量化、个性化地预测LA-NPC患者ICT的疗效,优于单一模型,可以作为一种无创的预测工具。

Objective:To explore the value of nomogram based on contrast-enhanced computed tomography radiomics integrated with clinic-radiological features in predicting response of locally advanced nasopharyngeal carcinoma(LA-NPC) to induction chemotherapy(ICT).Methods:The clinical and CT imaging data of 178 patients with LA-NPC from July 2014 to March 2022 were retrospectively analyzed.All patients were randomly stratified into training(n=125) and testing(n=53) cohorts at 7:3 ratios.3D-slicer was used to segment volume of interest(VOI) and features were extracted with Pyradiomic package.Predictive clinical factors were identified by univariate and multivariate logistic regression analysis.Least absolute shrinkage and selection operator(LASSO) was applied to select radiomics features.Finally, the clinic-radiological, radiomics and combined model were established by multivariate logistic regression.The performance of models were assessed by area of under curve(AUC) of the receiver operating characteristic curve(ROC).Decision curve analysis(DCA) was performed to evaluate the net benefit of the nomogram.Results:Logistic regression results demonstrated that T stage(OR=0.45,P=0.004),tumor enhancement degree(OR=2.706,P=0.038),platelet/lymphocyte ratio(PLR)(OR=0.289,P=0.024) were the clinical predictors of ICT response, and the clinical model was constructed based on the above three factors.Twenty-two optimal radiomics features significantly related to ICT efficacy were used to develop the radiomics model.The ROCs displayed that the combined model had the best predictive performance compared with clinical or radiomics models(AUC for training group: 0.821 vs 0.732、0.798;AUC for testing group: 0.836 vs 0.793、0.779).DCA analysis further showed that the combined nomogram model had a higher net benefit rate in population than the radiomics model alone.Conclusion:The nomogram based on enhanced CT radiomics combined with clinic-radiological features can intuitively and quantitatively predict the efficacy of ICT in LA-NPC patients, which is superior to the single model and can serve as a non-invasive predictive tool.