基于术前IVIM影像组学预测老年胶质瘤MGMT甲基化状态

Prediction of MGMT methylation status in elderly gliomas patients based on preoperative IVIM imaging radiomics

中文关键词: 老年  胶质瘤  体素内不相干运动  O6-甲基鸟嘌呤-DNA甲基转移酶  机器学习  预测价值

英文关键词: elderly  glioma  IVIM  MGMT  machine learning  predictive value

中文摘要:

      目的 本研究采用机器学习算法,探讨体素内不相干运动(intravoxel incoherent motion,IVIM)对老年胶质瘤患者O6-甲基鸟嘌呤-DNA甲基转移酶(O6-methylguanine-DNA methyltransferase,MGMT)启动子甲基化状态的预测价值.方法 回顾性分析39例经病理确诊的老年胶质瘤病例.通过3D slicer软件在真实扩散系数(true diffusion coefficient,D)参数图上勾画肿瘤感兴趣区,使用Python从D、快扩散系数(pseudo-diffusion coefficient,D*)、灌注分数(perfusion fraction,f)图中分别提取影像组学特征,用F-test和最小绝对收缩和选择算子法(least absolute shrinkage and selection op-erator,LASSO)对影像组学特征降维和筛选.分别采用8个机器学习算法构建单个弥散指标模型和联合模型,对老年胶质瘤MGMT启动子甲基化状态进行预测.使用留一验证法(leave-one-out cross-validation,LOOCV),通过平衡准确率(balanced accuracy,BA)、召回率(recall,REC)、精确率(precision,PRE)、F1 分数(F1 score,F1)和受试者工作特征曲线下面积(area under the receiver operating characteristic curve,AUC)评价各模型的诊断效能.结果 对于老年胶质瘤患者MGMT启动子甲基化状态的预测,性能最佳的机器学习算法为朴素贝叶斯(naive bayes,NB)(AUC均达0.850以上).基于NB算法构建的IVIM预测模型中,D的预测效能最高(BA:0.841,REC:0.870;PRE:0.870;F1:0.870;AUC:0.902).结论 基于术前IVIM影像组学的机器学习模型可有效预测老年胶质瘤患者MGMT启动子甲基化状态.

英文摘要:

      Objective To explore the predictive value of intravoxel incoherent movement(IVIM)on the methylation status of O6-methylguanine-DNA methyltransferase(MGMT)promoter in elderly glioma patients based on machine learning al-gorithm.Methods A retrospective analysis was conducted on 39 pathologically confirmed elderly glioma cases.3D slicer soft-ware was used to delineate the tumor region of interest on the true diffusion coefficient(D)parameter maps.Python was used to extract radiomics features from D,pseudo-diffusion coefficient(D*),and perfusion fraction(f)maps,respectively.F-test and least absolute shrinkage and selection operator(LASSO)were used to reduce and screen the radiomics features.Eight machine learning algorithms were used to construct single dispersion index models and a combined model,respectively,to predict the methylation status of MGMT promoter in elderly gliomas patients.The leave-one-out cross-validation(LOOCV)was employed,and the diagnostic performance of each model was evaluated by balancing accuracy(BA),recall(REC),pre-cision(PRE),F1 score(F1),and area under the receiver operating characteristic curve(AUC).Results For predicting the methylation status of MGMT promoter in elderly glioma patients,the naive Bayes(NB)algorithm demonstrated the best per-formance(AUC≥0.850).Among the IVIM prediction models constructed based on NB algorithm,the D parameter achieved the highest prediction performance(BA:0.841,REC:0.870;PRE:0.870;F1:0.870;AUC:0.902).Conclusion The machine learning model based on preoperative IVIM radiomics can effectively predict MGMT promoter methylation status in elderly glioma patients.