Time-Delay Neural Networks and Independent Component Analysis for EEG-Based Prediction of Epileptic Seizures Propagation

This research focuses on the development of a machine learning technique based on Time-Delay Neural Networks (TDNN) and Independent Component Analysis (ICA), to analyze EEG signal dynamics related to the initiation and propagation of epileptic seizures. We aim at designing a generative model to simulate EEG time-series after alteration of specific localized channels (electrodes) in order to explore the effects of brain surgery ex-vivo. Introduction Epilepsy is a chronic illness that affects approximately 12% of the world’s population. Up to 40% of epileptic patients are refractory to medical therapy and may be candidates for brain surgery. Intracranial electroencephalography (EEG) is a critical tool in the pre-surgical analysis of patients with medicationrefractory epilepsy. Epileptic seizures correspond to an abnormal oscillatory electric activity visible on EEG. Since EEG is recorded with several localized electrodes across the brain, it can be used to identify the seizure onset zone, which is defined as a small region of cortex whose removal could abolish seizures.