Seizure Prediction from Intracranial EEG Recordings

S EIZURE forecasting systems hold promise for improving the quality of life for patients with epilepsy. One proposed forecasting method relies on continuous intracranial electroencephalography (iEEG) recording to look for telltale feature sets in EEG data that suggest imminent seizure threats. In order for EEG –based seizure forecasting systems to work effectively, computational algorithms must reliably identify periods of increased probability of seizure occurrence. If a reasonably wellperforming algorithm could be implemented to identify those periods, then it is possible to warn patients only before seizures, enabling them to live a close to normal life. Recent research has shown that EEG signals can be classified in four unique categories: interictal (between seizures), preictal (immediately prior to seizure), ictal (during seizure) and postictal (after seizure). In this project, we applied machine learning algorithms to identify potential seizure occurrence using training datasets collected from both dog and human subjects. Specifically, we aimed to differentiate between data collected during preictal and interictal states, and thereby classify periods of increased probability of seizure occurrence in both dog and human subjects with naturally occurring epilepsy.