Learning MRI-based classification models for MGMT methylation status prediction in glioblastoma

Area of Science:

• Neuro-oncology
• Radiology
• Molecular Diagnostics

Background:

• O-6-methylguanine-DNA-methyltransferase (MGMT) promoter methylation is a predictive marker for glioblastoma (GBM) treatment response.
• Current methods for determining MGMT methylation status require invasive tissue biopsy.

Purpose of the Study:

• To assess the noninvasive prediction of MGMT promoter methylation status using quantitative and qualitative MRI variables in GBM patients.
• To establish imaging biomarkers for MGMT methylation status.

Main Methods:

• Retrospective analysis of MRI scans from GBM patients.
• Application of machine-learning methods to develop multivariate prediction models.
• Validation of models using data from The Cancer Genome Atlas (TCGA) database.

Main Results:

• MGMT promoter methylation status was predicted with up to 73.6% accuracy.
• Key imaging variables included edema/necrosis ratio, tumor/necrosis ratio, edema volume, and tumor location/enhancement characteristics.
• These variables significantly correlated with MGMT methylation status in GBM.

Conclusions:

• Standard preoperative MRI variables are associated with MGMT methylation status in GBM.
• Noninvasive prediction of MGMT methylation status via MRI is feasible.
• This approach may guide personalized glioblastoma treatment strategies.