Quantitative MRI Characterization of Fluid-Related Phenotypes in Pediatric Epilepsy: Deep Learning-Assisted Assessment of Enlarged Perivascular Space Burden, Regional Distribution, and Clinical Associations

Abstract

Objective: To quantitatively characterize EPVS burden and regional distribution, together with CP and ventricular CSF structural imaging phenotypes, in pediatric epilepsy and to explore within-epilepsy clinical associations of EPVS measures differing from controls. Methods: This retrospective cross-sectional study included 123 children with epilepsy and 108 controls aged 10–17 years. Enlarged perivascular space (EPVS), choroid plexus (CP), and ventricular cerebrospinal fluid (CSF) structural imaging phenotypes were quantitatively assessed using SHIVA-PVS, FreeSurfer, and 3D Slicer with expert visual quality control. Primary between-group comparisons used general linear models adjusted for age and sex, with estimated total intracranial volume additionally included for absolute volumetric measures, and multiple comparisons were controlled using the false discovery rate (FDR). Exploratory within-epilepsy clinical-association analyses were subsequently performed for EPVS measures showing FDR-significant between-group differences. Results: Global EPVS burden was higher in the epilepsy group than in controls (q < 0.01), with the most pronounced regional differences in the basal ganglia (BG; q < 0.001). Relative distribution measures showed higher BG burden and lower centrum semiovale (CSO) burden and CSO/BG ratios in epilepsy (q < 0.01). No statistically significant between-group differences were observed in the CP or ventricular CSF phenotypes. In exploratory within-epilepsy analyses, longer disease duration and drug resistance were associated predominantly with BG-EPVS measures, whereas associations with antiseizure medication burden were limited. Bilateral epileptiform activity was also associated with higher BG-EPVS burden after FDR correction (q < 0.05). Conclusions: Primary between-group analyses identified greater EPVS burden in pediatric epilepsy, with the most pronounced regional differences involving the BG. Exploratory within-epilepsy analyses identified associations between selected EPVS measures and specific clinical characteristics. These cross-sectional structural imaging findings characterize a regional EPVS phenotype for further longitudinal and multimodal investigation but do not establish an underlying glymphatic mechanism.