ADAPT-EPILEPSY: AN AI-AUGMENTED SPATIOTEMPORAL PREDICTIVE MODELING PROTOCOL FOR OPTIMIZING REPEAT MRI TIMING IN PEDIATRIC SEIZURES

Abstract

Currently, repeat MRI in pediatric epilepsy is reactive and empiric (mean 537 days) leading to delayed diagnosis of structural lesions, unnecessary sedation and suboptimal surgical referral rates. We developed the ADAPT-Epilepsy protocol, a spatiotemporal predictive model augmented with artificial intelligence, which integrates static clinical risks, dynamic EEG evolution and baseline MRI radiomics into a Structural Evolution Probability Score (SEPS) to optimize repeat MRI timing. The SEPS algorithm was applied retrospectively to a cohort of 260 children with seizures. Inputs included perinatal history, neurological deficits, evolving EEG abnormalities, GTC seizure semiology and radiomic features extracted from initial visually “normal” MRIs. Repeat MRI recommendation was based on a pre-specified high-risk threshold (SEPS ≥0.75). SEPS recommended repeat MRI in 1 9 (7.3%) patients and clinical examination in 28 (1 0.8%) patients. Diagnostic yield increased from 64.3% to 89.5%, mean interval to repeat MRI decreased from 537 to 1 89 days, and surgical referral rate from positive studies increased from 22.2% to 76.5%. SEPS was the only independent predictor of abnormal repeat MRI (OR 8.94, p<0.001). ADAPT-Epilepsy shifts pediatric neuroimaging from reactive to predictive, significantly increasing diagnostic yield, reducing unnecessary scans, and accelerating surgical candidacy. Prospective validation is needed.