基于ADC和增强MRI的影像组学模型预测低级别胶质瘤TERT启动子突变状态

Radiomics models based on ADC and contrast-enhanced MRI in predicting the TERT promoter mutation status of low-grade gliomas

目的:探讨基于ADC和增强MRI的影像组学模型对低级别胶质瘤端粒酶逆转录酶基因(TERT)启动子突变状态的预测价值。方法:回顾性搜集109例经病理证实的低级别胶质瘤患者,所有患者术前均行MRI检查,在ADC和对比增强T_(1)WI(T_(1)CE)图像上选取病灶最大层面,沿肿瘤边缘勾画ROI,提取影像组学特征。采用三联法(Fisher,POE+ACC,MI)和最小绝对收缩选择算子(LASSO)进行特征筛选,然后行多因素logistic回归分析,构建影像组学预测模型。采用ROC曲线评估预测模型的诊断效能。结果:在ADC和T_(1)CE图像上分别提取279个影像组学特征,最终筛选出11个影像组学特征,分别建立ADC模型、T_(1)CE模型和联合分析(ADC+T1CE)模型共3个影像组学模型。联合分析模型的预测效能最佳,训练集中曲线下面积(AUC)为0.928(95%CI:0.859~0.996),验证集中AUC为0.878(95%CI:0.758~0.997)。结论:基于ADC和增强MRI的影像组学模型能有效预测低级别胶质瘤TERT启动子突变状态,将不同序列的影像组学特征结合可提高预测效能。

Objective:To explore the value of the radiomics models based on ADC and contrast-enhanced MRI in predicting the TERT promoter mutation status of low-grade gliomas(LGGs).Methods:A total of 109 LGG patients confirmed by pathology were retrospectively analyzed.All patients underwent MRI scan before surgery.On the ADC and contrast-enhanced T_(1)-weighted(T_(1)CE)images,ROIs were delineated in each tumor along its edge at the slice with the maximum area and radiomics features were extracted.The features were selected by triad method(Fisher,POE+ACC,MI)and least absolute shrinkage selection operator(LASSO)analysis.Then,multivariate logistic regression analysis was used to construct the radiomics prediction models.Receiver operating characteristic analysis was used to evaluate the performance of models.Results:From ADC and T_(1)CE images,279 radiomics features were extracted respectively and 11 radiomics features were selected to construct three radiomics models named ADC model,T_(1)CE model and conjunct analysis(ADC+T_(1)CE)model.Among them,the conjunct analysis model showed the best prediction performance with AUC of 0.928(95%CI:0.859~0.996)in the training dataset and 0.878(95%CI:0.758~0.997)in validation dataset,respectively.Conclusions:The radiomics models based on ADC and contrast-enhanced MRI can effectively predict the TERT promoter mutation status in LGG patients,and the combination of the radiomics features from ADC and contrast-enhanced T_(1)-weighted sequences may improve the prediction performance.