基于CT影像组学的列线图模型预测肺部肿瘤立体定向放射治疗疗效

The value of nomogram model based on CT radiomics in predicting the efficacy of stereotactic body radiotherapy for lung tumors

目的:基于治疗前CT图像筛选放射组学特征构建列线图模型预测早期非小细胞肺癌(early stage-non-small cell lung cancer, ES-NSCLC)和肺部寡转移癌的放疗疗效。方法:本研究纳入122例接受立体定向放射治疗(stereotactic body radiotherapy, SBRT)的ES-NSCLC和肺部寡转移癌的患者,随机分为训练集和验证集。使用最小绝对收缩和选择算子(least absolute shrinkage and selection operator, LASSO)和逻辑回归(logistic regression)筛选训练集中与放疗疗效相关的放射组学特征以建立列线图模型。用受试者工作特征曲线(receiver operating characteristic curves, ROC)下面积(area under the curve, AUC)、校准曲线和决策曲线(decision curve analysis, DCA)评价模型性能。结果:经筛选得出6个放射组学特征形成放射组学特征分数(radiomics score, Rad-score)以建立列线图模型。模型训练集的AUC值为0.808(95%CI:0.712~0.884,P<0.001),验证集的AUC为0.741(95%CI:0.556~0.879,P=0.003)。Delong检测显示模型表现均衡(P=0.496),校准曲线和DCA均显示了模型较好的预测性能和较高的临床价值。结论:我们基于治疗前CT图像开发并验证了用于预测肺部肿瘤SBRT治疗疗效的列线图模型,该模型具有较高的预测性能和临床实用性。

Objective:To construct a nomogram model based on pretreatment CT images to screen radiomics characteristics to predict the radiotherapy efficacy of early stage-non-small cell lung cancer(ES-NSCLC) and oligometastatic lung cancer.Methods:122 patients with ES-NSCLC and lung oligometastatic carcinoma who received stereotactic body radiotherapy(SBRT) were randomly divided into training set and validation set. Radiomics characteristics associated with radiotherapy efficacy in the training set were screened using least absolute shrinkage and selection operator(LASSO) and logistic regression to build nomogram models. Model performance was evaluated with area under the receiver operating characteristic curves,calibration curves,and decision curve analysis(DCA).Results:Six radiomics features were screened to form radiomics score(Rad-score) to establish nomogram model. AUC values were 0. 808(95% CI:0. 712 ~ 0. 884,P< 0. 001) for the model training set and 0. 741(95% CI:0. 556 ~ 0. 879,P=0. 003) for the validation set. Delong test showed that the model performed equally(P= 0. 496),and both the calibration curve and DCA showed a better predictive performance of the model and a higher clinical value.Conclusion:We developed and validated a nomogram model for predicting SBRT treatment efficacy in lung tumors based on pretreatment CT,which has high predictive performance and clinical utility.