Authors :
Presenting Author: Davide Giampiccolo, MD – NHNN, UCL
Fenglai Xiao, MD PhD – UCL Institute of Neurology
Aidan O'Keeffe, PhD – University of Nottingham
Silvia Bonelli, MD PhD – University of Vienna
Christian Dorfer, MD – University of Vienna
Florian Fischmeister, PhD – University of Vienna
Jan Van Dijk, MSc – UCL Institute of Neurology
Michele Rizzi, MD – Instituto Neurologico Besta
Victoria Morgan, PhD – Vanderbilt University Medical Center
Dario Englot, MD PhD – Vanderbilt University Medical Center
Beate Diehl, MD PhD – UCL Institute of Neurology
Fahmida Chowdhury, FRCP PhD – National Hospital for Neurology and Neurosurgery
Andrew McEvoy, FRCS – National Hospital for Neurology and Neurosurgery
anna miserocchi, MD – NHNN, UCL
Matthias Koepp, MD PhD – UCL Queen Square Institute of Neurology/NHNN
John duncan, DM FRCP FMedSci – NHNN, UCL
Rationale: Epilepsy is recognized as a network disorder extending beyond cortical regions. Different epilepsy syndromes likely involve distinct cortico-subcortical networks, as suggested by syndrome-specific targets for resection and neuromodulation. Cortical involvement has been extensively studied, but subcortical contributions are poorly characterized. We investigated subcortical (thalamic, striatal, pallidal, and cerebellar) volumetric abnormalities across frontal lobe epilepsy (FLE), temporal lobe epilepsy (TLE), idiopathic generalized epilepsy (IGE), and healthy controls.
Methods:
Preoperative T1-weighted MRI scans from adult and paediatric epilepsy patients at NHNN and six independent cohorts were combined with healthy control scans from NHNN and four publicly available databases. Images underwent cortical and subcortical segmentation (FreeSurfer, THOMAS), with multi-center data harmonized using neuroCombat (covariates: age, sex, epilepsy syndrome). After normalizing to healthy controls, a MANCOVA with Bonferroni correction (covariates: age, sex, intracranial volume) identified syndrome-specific volume changes. A Random Forest Classifier with 10-fold cross-validation was used to classify epilepsy syndromes, with performance assessed by AUC and permutation testing. Permutation feature importance analysis was used to identify discriminative features.
Results:
1158 epilepsy patients [697 TLE, 213 FLE, 248 IGE] were compared with 989 controls. Overall, epilepsy syndromes were characterised by thalamic atrophy and striatal hypertrophy compared to controls. TLE patients had hippocampal atrophy, thalamic atrophy involving the anterior nucleus (ANT), pulvinar (PUL) and mediodorsal thalamic nuclei (MD) and pallidal atrophy. FLE had MD, PUL and cerebellar atrophy and centromedian (CM) and amygdala hypertrophy. IGE were characterised by habenula (Hb), lateral (LGN) and medial geniculate nuclei (MGN) atrophy and CM hypertrophy.
Random Forest classification achieved good discrimination (AUC: 0.732-0.985) with overall accuracy of 74.5% (95% CI: 72.8%-76.2%). The most discriminative features were hippocampus (mean decrease in accuracy (MDA): 30.3), PUL(MDA: 17.3), ANT (MDA: 18.9) for TLE. Hb (MDA: 53.7) and CM (MDA: 29.8) for IGE.CM (MDA: 6.2) for FLE.
Conclusions:
Different epilepsy syndromes have shared and distinct patterns of subcortical volume changes. Opposite thalamic and striatal volume changes supports models of separate seizure generating and stopping networks, while syndrome-specific patterns suggest discrete network components that may provide targeted treatment opportunities for ablation or neuromodulation.
Funding: Epilepsy Research Institute UK