Cross-scale prediction of glioblastoma MGMT methylation status based on deep learning combined with MRI and pathology images

Motivation: Promoter methylation of MGMT in glioblastoma is associated with beneficial treatment outcomes and prognosis; however, MGMT testing is costly, and molecular detection is not available in some hospitals. Goal(s): This study aims to construct a model based on MRI and pathological images to predict the status of MGMT. Approach: MRI and pathological images from patients with glioblastoma and a machine learning model was established to predict the MGMT status. Results: The combination of MRI and pathological images demonstrated superior predictive performance for MGMT compared to either MRI or pathological images alone. Impact: This research provides a method for therapeutic decision-making and prognostic prediction in glioblastoma, offering a potential approach for the future integration of multi-information for accurate prediction and clinical application in other tumors.