Improving MGMT methylation status prediction of glioblastoma through optimizing radiomics features using genetic...

Area of Science:

• Neuro-oncology
• Medical imaging analysis
• Machine learning in medicine

Background:

• O6-Methylguanine-DNA-methyltransferase (MGMT) promoter methylation is a key predictor of temozolomide resistance and progression-free survival in glioblastoma multiforme (GBM).
• Current molecular techniques for determining MGMT methylation status face challenges including specimen acquisition, cost, and tumor heterogeneity.
• Noninvasive preoperative prediction of MGMT methylation status is highly desirable for optimizing GBM patient management.

Purpose of the Study:

• To develop and validate a radiomics-based machine learning (ML) model for noninvasive, preoperative prediction of MGMT methylation status in GBM patients.
• To assess the feasibility of using multimodal magnetic resonance imaging (MRI) features for this prediction.
• To evaluate the model's performance on both high-grade (GBM) and low-grade glioma (LGG) datasets.

Main Methods:

• Retrospective analysis of radiomics features extracted from multimodal MRI scans of GBM patients with known MGMT methylation status from the TCIA database.
• A two-stage feature selection process involving an eXtreme Gradient Boosting (XGBoost) model followed by a genetic algorithm (GA)-based wrapper model.
• Model performance was evaluated using cross-validation, reporting sensitivity, specificity, and accuracy.

Main Results:

• The GA-based wrapper model demonstrated high predictive performance in GBM, achieving a sensitivity of 0.894, specificity of 0.966, and accuracy of 0.925 for MGMT methylation status.
• When applied to an LGG dataset, the model achieved a sensitivity of 0.780, specificity of 0.620, and accuracy of 0.750.
• The selected radiomics features showed potential for application across both low- and high-grade gliomas.

Conclusions:

• A novel radiomics-based ML model can accurately and noninvasively predict MGMT methylation status in GBM.
• The model's ability to perform well on LGG data suggests broader applicability in glioma classification.
• This approach offers a promising tool for improving preoperative assessment, prognosis, and treatment strategies for glioma patients.