Prediction of glioma-related epilepsy by Brain Age Index: a multicenter study

Abstract

Glioma frequently induces widespread structural and functional alterations extending beyond the tumor site, with epilepsy being one of its most common clinical manifestations. Conventional brain-age models are rarely applied to neurosurgical diseases because focal structural damage violates the assumption of global anatomical integrity. To address this limitation, we propose a novel Brain Age Index (BAI) that integrates bias-corrected brain-age estimations with chronological-age normalization, computed exclusively from non-tumorous brain regions. Using T1-weighted MRI data from 307 glioma patients across three centers and 671 healthy controls, we trained a residual convolutional neural network model for brain-age prediction (mean absolute error, 3.35 ± 4.19 years) and derived the BAI to quantify systemic cerebral alterations. Glioma patients exhibited significantly higher BAI values than healthy controls (p < 0.001). Notably, patients with glioma-related epilepsy showed reduced brain-age acceleration compared with non-epileptic patients, suggesting possible adaptive neural reorganization. A combined clinic-radiomic model incorporating BAI achieved an Area Under Curve (AUC) of 0.79 for epilepsy prediction. Collectively, these findings establish the BAI as a promising imaging biomarker for detecting tumor-related cerebral alterations and for enhancing prognostic modeling and functional network assessment in glioma.