山西医科大学学报2025，Vol.56Issue(6)：589-595,7.DOI:10.13753/j.issn.1007-6611.2025.06.001
基于多模态MRI的胶质瘤IDH突变伴MGMT启动子甲基化预测
Multimodal MRI-based prediction of IDH mutation combined with MGMT promoter methylation in glioma
摘要
Abstract
Objective To develop combined predictive models based on preoperative multimodal MRI radiomics for the co-occurrence of isocitrate dehydrogenase(IDH)mutation and O⁶-methylguanine-DNA methyltransferase(MGMT)promoter methylation in gliomas.Methods A total of 246 glioma patients from the First Hospital of Shanxi Medical University and Shanxi Provincial People's Hospital from December 2011 to December 2019 were included in the study.Radiomic features from tumor regions were extracted using FAE.Patients were categorized into IDH mutation with MGMT promoter methylation(IDHmut&MGMTmet)group and other molecular sta-tus group,and then randomly divided into training dataset and testing dataset with an 80∶20 split ratio.A penalized Logistic regression was performed for robust feature selection.For IDHmut&MGMTmet,seven machine learning methods,including kernel partial least squares with genetic algorithm(GA-KPLS),random forest,LASSO,Logistic regression,neural network,K-nearest neighbor classifi-cation,and Naïve Bayes were performed to construct the predictive model,and the model performance was evaluated using different criteria.Results There were significant differences in age and pathological grade between IDHmut&MGMTmet group and other mo-lecular status group(P<0.05).A total of 1 686 features were extracted from both T1CE and T2FLAIR images,and 115 features from T1CE and 30 from T2FLAIR were selected using penalized Logistic regression.The results of Dunnett's multiple comparison test showed that the GA-KPLS model outperformed the other machine learning models in predictive performance,with sensitivity of 0.752,specificity of 0.836,accuracy of 0.802,area under the ROC of 0.877,F-measure of 0.755,G-means of 0.790,and MCC of 0.594.Conclusion The GA-KPLS-based predictive model for IDH mutation and MGMT promoter methylation status in glioma patients de-monstrates good discriminatory power and can be used for non-invasive molecular subtype diagnosis,chemotherapy decision-making,and prognosis assessment.关键词
多模态MRI/GA-KPLS/机器学习/脑胶质瘤/影像组学/IDH/MGMTKey words
multimodal MRI/GA-KPLS/machine learning/glioma/radiomics/IDH/MGMT分类
医药卫生引用本文复制引用
闫俊羽,杨琪,王效春,谭艳,张辉,曹红艳,杨国强..基于多模态MRI的胶质瘤IDH突变伴MGMT启动子甲基化预测[J].山西医科大学学报,2025,56(6):589-595,7.基金项目
国家自然科学基金资助项目(U21A20386,82473739,82071893,82371941) （U21A20386,82473739,82071893,82371941）
山西省基础研究计划项目(202303021211204,202303021211130) （202303021211204,202303021211130）
山西省回国留学人员科研资助项目(2024-081) （2024-081）
山西省高等教育"百亿工程"科技引导专项项目 （）
