基于影像组学和人工智能预测非小细胞肺癌放化疗疗效的可行性

Feasibility of predicting the efficacy of radiotherapy and chemotherapy for non-small cell lung cancer based on radiomics and artificial intelligence

目的:构建基于影像组学特征的预测模型,以预测非小细胞肺癌患者接受序贯放化疗(sequential chemoradiotherapy,SCRT)或同步放化疗(concurrent chemoradiotherapy,CCRT)后的病情部分缓解(partial response,PR)可能性。方法:回顾性收集2016年01月至2020年06月确诊为非小细胞肺癌并接受SCRT或CCRT患者资料。符合条件的患者纳入本研究中,并随机分为训练集和验证集。采用单因素方差分析及最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)算法,在训练集中筛选出最佳影像组学特征。在训练集中进行机器学习(Logistic regression,LR;Decision tree,DT;AdaBoost)模型构建。受试者工作特征曲线下面积(area under curve,AUC)、敏感性和特异性用于评估模型性能,使用列线图对模型进行可视化,决策曲线分析法检验模型应用效能。结果:共纳入75例患者,随机分为两组,训练集52例,验证集23例。在进行单因素方差分析和LASSO回归分析后,筛选出了6个放射学特征,使用机器学习方法构建预测模型。在训练集中,LR、DT、AdaBoost的模型的AUC为0.919、0.773及0.832,在验证集中为0.795、0.723及0.638。使用LR模型构建决策曲线表明,当风险阈值为0.1~0.92时,可增加患者的净效益。结论:本研究开发并验证了一个影像组学预测模型,可以预测接受SCRT/CCRT后肺癌患者的缓解概率。

Objective:According to the radiomic features,a prediction model was established and verified to predict the possibility of partial response(PR)of non-small cell lung cancer(NSCLC)patients after receiving sequential chemoradiotherapy(SCRT)or concurrent chemoradiotherapy(CCRT).Methods:Patients diagnosed with NSCLC and receiving SCRT or CCRT from January 2016 to June 2020 were retrospectively collected.Eligible patients were included in this study and randomly divided into a training group and a validation group.The one-way analysis of variance and the LEAST Absolute Shrinkage and Selection Operator(LASSO)algorithm were used to screen the optimal radiographic features in the training set.Machine learning(Logistic egression,LR.Decision tree,DT.AdaBoost)model building.The area under the receiver operating characteristic curve curve(AUC),sensitivity and specificity were used to evaluate the model performance,the nomogram was used to visualize the model,and the decision curve analysis(DCA)method was used to test the application efficiency of the model.Results:A total of 75 patients were included and randomly divided into two groups,52 in the training group and 23 in the validation group.Six radiomics features were selected after univariate analysis of variance and LASSO regression analysis,and predictive models were constructed using machine learning classifier.The model AUCs for LR,DT and AdaBoost were 0.919,0.773 and 0.832 in the training group and 0.795,0.723 and 0.638 in the validation group.The use of LR models to construct decision curves suggested that a risk threshold of 0.1~0.92 increased the net benefit for the patient.Conclusion:This study developed and verified an radiomics model,which can predict the remission probability of NSCLC patients after SCRT/CCRT.