基于CT影像组学的肺癌生存预后预测分析

Prognostic prediction in non-small cell lung cancer based on CT radiomics

目的探讨非小细胞肺癌(NSCLC)患者基于治疗前CT增强图像的影像组学特征对其生存期的预后价值。方法研究数据来源于癌症影像档案(the cancer imaging archive,TCIA)中的NSCLC-Radiomics公共数据集,使用数据库中的421例NSCLC患者的基线资料和CT影像数据,然后从每例患者的CT三维影像数据中提取组学特征,将所有病例按照7∶3的比例随机分为两组:训练集(296例)和测试集(125例),在训练组中以最小绝对收缩和选择算子(LASSO)算法筛选预测总生存(OS)的影像组学特征,基于Cox比例风险回归模型,建立预测模型,将患者分为高、低风险2组,Kaplan-Meier生存曲线比较两组间生存差异,纳入临床特征建立预后模型,曲线下面积(AUC)评价其预测效能。基于预后模型绘制列线图。结果共提取1409个组学特征,经降维后得到11个最有价值的组学特征。建模后计算组学标签,高、低风险2组在训练集和验证集中OS均有显著性差异(P<0.001)。Cox单因素和多因素分析显示影像组学标签均是影响OS[风险比(HR)值:1.529、1.369,95%CI:1.389~1.684、1.201~1.552,均P<0.0001]的独立预后因素。建立的LASSO-COX模型预测1年、3年和5年的OS,训练集AUC分别为0.696、0.718、0.749,在测试集中预测OS的AUC分别为0.689、0.667、0.661。结论基于CT图像的影像组学特征建立的预后模型有助于预测NSCLC患者OS状态。

Objective To evaluate the prognostic significance of CT-based radiomics features in predicting overall survival(OS)of patients with non-small cell lung cancer(NSCLC).Methods NSCLC-Radiomics public dataset was obtained from the Cancer Imaging Archive(TCIA).Pre-treatment CT,Region of interest(ROI)and baseline information of 421 NSCLC patients were retrospectively reviewed,and radiomic features were extracted from the three-dimensional CT images of each patient.The cases were randomly divided into two groups the training set(296 cases)and the validation set(125 cases)in a 7∶3 ratio.Feature selection was performed in the training set using the Least Absolute Shrinkage and Selection Operator(LASSO)algorithm for predicting the radiomics features of OS,and prediction model was constructed based on the COX proportional risk regression model.The patients were divided into high and low-risk groups.The Kaplan-Meier survival curves were used to compare the survival differences between the two groups.The area under the curve(AUC)was used to evaluate the predictive power of the prediction model incorporating clinical features.Additionally,a visualized nomogram was further constructed based on the prediction model.Results A total of 1409 radiomics features were extracted,and after dimensionality reduction 11 most valuable radiomics features were obtained.After modelling,radiomics score was calculated and there was a significant difference in OS between the high and low-risk groups in both the training and validation sets(P<0.001).Cox univariate and multifactor analysis showed that CT radiomics score was an independent prognostic factor affecting OS[hazard ratio(HR)value:1.529,1.369;95%CI:1.389~1.684,1.207-1.552,P<0.0001].The LASSO-COX model achieved an AUC of 0.696,0.718 and 0.749 for predicting 1-year,3-year,and 5-year OS,respectively,in the training set,and 0.689,0.667 and 0.661,respectively in the validation set.Conclusion Prediction models based on radiomics features of CT image may be useful in predicting OS in NSCLC patients.