基于临床特征和影像组学特征的联合模型预测磨玻璃结节型肺腺癌浸润性的价值

Value of combined model based on clinical and radiomics features for predicting invasiveness of lung adenocarcinoma manifesting as ground glass nodule

目的:探讨基于临床特征和影像组学特征的联合模型对磨玻璃结节(ground glass nodule,GGN)型肺腺癌浸润性的预测价值。方法:回顾性分析2019年1—12月在某院行胸部CT检查且经手术病理证实的GGN型肺腺癌患者的临床资料,使用基于Python的开放资源Pyradiomics进行影像组学特征的提取。根据单因素分析和多因素分析得到的独立危险因素构建临床模型,使用筛选出的影像组学特征建立影像组学模型,使用临床模型的预测值与影像组学分数(Radscore)构建联合模型。使用ROC曲线评估3种模型在训练集和测试集中的预测性能,使用Delong检验评估3种模型预测GGN型肺腺癌浸润性ROC曲线差异是否有统计学意义,使用临床决策曲线分析模型的净效益。结果:Logistic多因素分析结果显示,年龄(P=0.0202)、血管改变(P=0.0022)是预测肺腺癌浸润程度的独立因素。影像组学模型、临床模型和联合模型在训练集上的AUC分别为0.876、0.867、0.904,在测试集上的AUC分别为0.828、0.828、0.864。在测试集上,联合模型与临床模型、影像组学模型的ROC曲线之间比较差异无统计学意义(P>0.05)。临床决策曲线显示,在大部分阈概率的条件下,使用联合模型预测浸润性具有更高的临床效益。结论:基于临床特征和影像组学特征的联合模型可以提高对GGN型肺腺癌浸润性的预测性能。

Objective To evaluate the predictive value of a combined model based on clinical and radiomics features for the invasiveness of lung adenocarcinoma manifesting as ground glass nodule(GGN).Methods Clinical data of patients with GGN-type lung adenocarcinoma who underwent chest CT and were confirmed by surgical pathology at some hospital from January to December 2019 were analyzed retrospectively,and the extraction of imaging histological features was performed using Python-based open resource Pyradiomics.A clinical model was constructed based on independent risk factors obtained from univariate and multivariate analyses,a radiomics model was built using the screened radiomics features,and a combined model was established with the predictive values of the clinical models and radiomics scores(Radscore).The predictive performance of the three models in the training and test sets was assessed using ROC curves,the statistical significance of the differences in the ROC curves of the three models for predicting GGN-type lung adenocarcinoma was assessed using the Delong test,and the net benefits of the models were analyzed using clinical decision curves.Results Logistic multifactor analysis showed that age(P=0.0202)and vascular characteristics(P=0.0022)were the independent predictors of the degree of the invasiveness of lung adenocarcinoma.The AUCs of the radiomics model,clinical model and combined model were 0.876,0.867 and 0.904 on the training set,and 0.828,0.828 and 0.864 on the test set,respectively.The difference between the ROC curves of the combined model and the clinical and radiomics models was not statistically significant(P>0.05)on the test set.Clinical decision curves showed a higher clinical benefit when using the combined model to predict the invasiveness under most conditions of threshold probability.Conclusion The combined model based on clinical and radiomics features enhances the predictive performance for the invasiveness of GGN-type lung adenocarcinoma.