CT影像组学模型评估直肠癌周围神经或脉管侵犯的价值

The value of CT-based radiomics model in the evaluation of perineural or lympho-vascular invasion in rectal cancer

目的:探讨基于增强CT的影像组学在直肠癌周围有无神经或脉管侵犯的评估价值。方法:回顾性分析我院2017年12月至2021年06月经病理证实的196例直肠癌患者的临床及影像学资料。术后病理提示107例无神经及脉管侵犯,89例有神经或脉管侵犯。患者术前均接受全腹部CT平扫及增强检查。以8∶2的比例采用分层随机抽样法将患者分为训练集和测试集(157例vs 39例),利用PyRadiomics软件从术前CT影像中提取113个组学特征,利用Spearman相关分析与LASSO的方法进行特征的筛选与模型构建。最后采用受试者工作特征(receiver operating characteristic,ROC)曲线对训练集及验证集的模型验证,评价CT影像组学模型在直肠癌周围有无神经或脉管侵犯中的预测价值。结果:经过特征筛选,平扫图像中13个组学特征及静脉期图像中15个组学特征用于构建评估直肠癌周围有无神经或脉管侵犯的模型。训练组中预测模型的ROC曲线下面积(area under curve,AUC)为0.91(95%CI:0.86~0.96),敏感性、特异度与准确度分别为91.8%、76.6%、83.3%;测试组中AUC是0.91(95%CI:0.82~0.97),敏感性、特异度与准确度分别为85.7%、80.0%、82.7%。结论:基于增强CT的影像组学模型可用于评估直肠癌周围有无神经及脉管侵犯,且具有较高诊断价值。

Objective:To investigate the value of evaluating nerve or lympho-vascular invasion around rectal cancer based on CT radiomics model.Methods:The clinical and imaging data of 196 patients with colorectal cancer confirmed by pathology in our hospital from December 2017 to June 2021 were retrospectively analyzed.There were 107 cases without nerve and lympho-vascular invasion,and 89 cases with nerve or lympho-vascular invasion.All patients underwent preoperative total abdominal CT scan and enhanced scan.The stratified random sampling method divided patients into training group and test group by 8∶2 ratio(157 vs 39).A total of 113 radiomics features were extracted by PyRadiomics software based on preoperative CT images.Spearman correlation analysis and least absolute shrinkage and selection operator(LASSO)were used for feature selection and model building.Receiver operating characteristic(ROC)curve was used to evaluate the diagnostic potential of the CT-based radiomics model in the prediction of nerve or lympho-vascular invasion around rectal cancer.Results:After feature selection,13 radiomics features in plain scan and 15 in venous phase were used to construct a model for evaluating the presence or absence of nerve or lympho-vascular invasion around rectal cancer.In the training group,the area under curve(AUC)of the prediction model was 0.91(95%CI:0.86~0.96)with 91.8%sensitivity,76.6%specificity,and 83.3%accuracy.In the test group,the AUC was 0.91(95%CI:0.82~0.97),with 85.7%sensitivity,80.0%specificity,and 82.7%accuracy.Conclusion:The CT-based radiomics model could be used to evaluate the presence of nerve and lympho-vascular invasion around rectal cancer,which showed a high diagnostic performance.