MGMT Promoter Methylation Prediction in Glioblastoma Using 3D CNNs with Advanced MRI Sequences

Tran Nguyen Tuan Minh, Hien Quang Kha, Viet Huan Le, Matthew Chin Heng Chua, Nguyen Quoc Khanh Le

Keywords: glioblastoma, MGMT methylation, deep learning, arterial spin labeling, apparent diffusion coefficient, MRI, non-invasive diagnosis

TL;DR: We show that combining advanced MRI sequences (i.e., ASL and ADC) significantly improves deep-learning prediction of MGMT promoter methylation in glioblastoma.

Abstract: Accurate determination of O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation status is essential for therapeutic planning in glioblastoma (GBM). Although molecular assays remain the reference standard, they are costly, invasive, and not always feasible in routine practice. This has motivated the development of non-invasive MRI-based deep learning approaches, particularly those leveraging advanced physiological imaging sequences. In this study, we investigated whether arterial spin labeling (ASL) and apparent diffusion coefficient (ADC) imaging provide complementary information for predicting MGMT methylation status in IDH-wildtype GBM. We analyzed 351 patients from the UCSF Preoperative Diffuse Glioma MRI dataset and trained 3D convolutional neural network models based on a ResNet-10 architecture using ASL, ADC, diffusion-weighted imaging (DWI), and conventional T2-FLAIR sequences. Among single-sequence models, ASL achieved the highest performance (accuracy of 0.76, precision of 0.75, and F1 score of 0.73). A dual-sequence model combining ASL and ADC further improved prediction, yielding an AUC of 0.83, significantly outperforming both the ASL-only model and the T2-FLAIR model (AUC 0.6524; DeLong test, $p<0.05$). These results demonstrate that integrating perfusion- and diffusion-based MRI captures complementary physiological characteristics relevant to MGMT methylation, offering a more accurate and fully non-invasive alternative for biomarker assessment. Incorporating advanced MRI sequences into deep learning pipelines may support more informed treatment planning and improve clinical decision-making for patients with GBM.

Primary Subject Area: Detection and Diagnosis

Secondary Subject Area: Application: Neuroimaging

Registration Requirement: Yes

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 367