Predictive efficacy of multimodal MRI-based machine learning models for glioblastoma multiforme MGMT promoter methylation states
10.19745/j.1003-8868.2025099
- VernacularTitle:基于多模态MRI的机器学习模型对胶质母细胞瘤MGMT启动子甲基化状态预测效能研究
- Author:
Hong-lin LI
1
;
Shi-ting HU
;
Zi-heng ZHOU
;
Bing LI
;
Zhi-ping QI
;
Ruo-qi LI
;
Kai LIU
;
Chun-feng HU
;
Hai-tao GE
Author Information
1. 徐州医科大学医学影像学院,江苏 徐州 221004
- Publication Type:Journal Article
- Keywords:
glioblastoma multiforme;
magnetic resonance imaging;
machine learning;
radiomics;
MGMT promoter methylation
- From:
Chinese Medical Equipment Journal
2025;46(6):7-13
- CountryChina
- Language:Chinese
-
Abstract:
Objective To explore the predictive efficacy of several multimodal MRI-based machine learning models for the promoter methylation states of O6-methylguanine-DNA methyltransferase(MGMT)of glioblastoma muliforme(GBM)patients in terms of the GBM heterogeneity and the complexity of the tumor microenvironment.Methods Firstly,the multimodal MRI images of 317 GBM patients from The University of Pennsylvania Glioblastoma(UPENN-GBM)dataset were pre-processed,with four sequences involved in including T1-weighted imaging(T1WI)sequence,T1-weighted contrast-enhanced imaging(T1CE)sequence,T2-weighted imaging(T2WI)sequence and fluid-attenuated inversion recovery(FLAIR)sequence,and the radiomics features were extracted for two regions of interest(ROIs)such as the tumor core region and the tumor edema region.Secondly,the data of the 317 GBM patients were randomly divided into a training set(254 cases)and a test set(63 cases),which underwent normalization with Z-scores and feature selection and dimensionality reduction with Lasso regression.Finally,three models were established respectively with particle swarm optimization-support vector machine(PSO-SVM),C-support vector classification(C-SVC)and adaptive boosting(adaptive boosting(Adaboost)algorithms,and the predictive efficacy of the three models for glioblastoma multiforme MGMT promoter methylation states were evaluated in terms of accuracy and AUC.Results The Adaboost model based on T2WI sequence and radiomics features of the tumor core region had the highest predictive efficacy with accuracy and AUC values of 67％and 0.74,respectively,higher than those of other combinations of sequences,models and regions of interest.Conclusion The multimodal MRI-based machine learning models can be used for the prediction of glioblastoma multiforme MGMT promoter methylation states,which provides powerful support for personalized treatment and prognostic assessment of GBM.[Chinese Medical Equipment Journal,2025,46(6):7-13]
