Epilepsy surgical outcome prediction based on epileptogenic zone localization with a network competition approach

Abstract

Nearly one third of all epilepsy patients are not able to control their epileptic seizures with standard epilepsy medication. For these patients, surgical treatment is the main alternative, which consists on the resection or ablation of the Epileptogenic Zone. Despite the usage of advanced methods for surgical evaluation like intracranial EEG or PET-MRI, at least half of all intervened patients will end up recurring in the long term. Many computational approaches have been taken in order to aid in the prediction of the outcome of epilepsy surgery but to this day, none of them have managed to obtain a perfect accuracy. In this study, a novel approach is taken in conceptualizing the interaction between the non-epileptogenic and epileptogenic brain networks as a competition to dominate each other. By implementing Game Theory notions in conjunction with a set of Machine Learning algorithms it was possible to observe how the influence of the epileptogenic regions towards the non-epileptogenic ones grew over time until it produced the onset of a seizure. Furthermore, a score was built in order quantitatively assess the probability of obtaining a favorable surgical outcome. The score was able to fully discriminate subjects with good and bad outcome in cohort of 21 patients. Lastly, using an adapted Genetic Algorithm it was possible to find alternative resection sites that would improve the chances of obtaining seizure freedom in 2 out of 7 patients that initially obtained a bad outcome.