The diagnosis of seizure onset zone is crucial in epilepsy surgery for patients having pharmacoresistant, that the localization of ictal onset zone (IOZ) is associated with epileptiform activity region, by using intracranial strip and grid electrodes of electrocorticography (ECOG). Thus, these ECOG signals determine areas of the brain which can be surgically removed by visual inspection, especially in partial epilepsy named focal seizure. However, this paper investigates recent automatic IOZ localization, based on recurrent neural network (RNN) to improve the performance of identification of IOZ. Therefore, statistical patterns in time-frequency domains are extracted after applying hybrid empirical mode decomposition (EMD) and variational mode decomposition (VMD) methods. In addition, the evaluation of this work is performed by implementation of several indicators to distinct between focal and non focal ECOG signals. Moreover, the experiment results using Bern– Barcelona dataset indicated the effectiveness of using hybrid EMD-VMD method and RNN, that the proposed model achieve better classification performance reached 100% of accuracy. Hence, by comparison with other studies in the literature review, ECoG recordings provide best localization of IOZ by using this system. Finally, this developed method shows accurate results to help clinical experts in seizure resection, and indicates localization of IOZ which can be removed from brain area.
Read full abstract