Predicting seizure spread andneurosurgical outcomes in epilepsy bycombining neuroimaging, machinelearning, and computer modelling

Abstract

Background: Epilepsy is a neurological disorder of abnormal brain network in which seizures originate and spread via patient-specific spatial and temporal pathways. Disrupting these epileptic networks can enable seizure control; therefore, it is crucial to map, quantify, and understand these networks. This thesis aims to quantify the whole-brain structural network abnormalities of patients with focal and generalised epilepsy along with patientspecific network disruptions caused by epilepsy surgery. Method: We developed a novel patient-specific metric to quantify structural network abnormality at every brain region by standardising whole-brain structural networks of patients with healthy structural networks. To quantify local changes in the white-matter structure, we applied quantitative neuroimaging techniques and a computational model for making predictions on mechanisms of epilepsy development. We combined the network-based measures in robust cross-validated machine learning models to predict neurosurgical outcomes and seizure spread. Results: In drug-resistant focal epilepsy patients, structural network abnormality associated with post-surgical seizure recurrence and patient history of focal to bilateral tonic-clonic seizures. Combined with routinely acquired clinical variables, we predicted the patientspecific probability of seizure recurrence after surgery. In patients with idiopathic generalised epilepsy, we found localised abnormalities in major white-matter fascicles. The thalamocortical computer model of spike-wave seizures implicated the role of cortico-reticular connections in mechanism of epileptogenesis. Significance: This thesis highlighted the heterogeneity between patients that may be making them susceptible to a varied response for the same treatment. We offer practical tools to quantify these heterogeneities to complement clinical decision-making for effective patient stratification and tailored treatments.