Classification of Pre-ictal and Interictal Periods Based on EEG Frequency Features in Epilepsy

Abstract

A seizure prediction system has the potential to significantly help patients with epilepsy. For a seizure forecasting system to work effectively, computational algorithms must reliably identify periods with high probability of seizure occurrence. We herein report results of a classification approach based on machine learning of EEG features in the frequency domain and aimed at differentiating between pre-ictal (close to seizure onsets) and interictal (far away from seizures onset) periods in long-term intracranial EEG recordings from the brain of 5 epileptic dogs. Evaluation of performance by the area under the ROC curve ranged from 0.84 to 0.96 in four dogs, while for the fifth dog was considerably less (0.55), resulting to a global value of 0.87 across dogs. These results offer supporting evidence that seizures may be predictable with a proper analysis of the EEG.