Slow Component Analysis Based Interictal-Preictal EEG Prediction

An accurate and compact epileptic seizure prediction system would be significant to intractable patients as it can significantly improve the quality of their lives. This paper presents a novel approach for prediction of seizures in epileptic patients based on scalp electroencephalography (EEG) recordings. In this study, we explored meaningful underlying representation to derive latent predictive characteristics of EEG signals, at the same time make it possible to design a lightweight, compact solution for seizure prediction. This paper proposes a novel approach that explores the underlying representation of the signals by introducing Slow Component Analysis (SCA) for the differentiation. In addition, extracting several epileptiform-band signals from the original EEGs prior to the slowness analysis helps find the intrinsic change of the brain states. The performance of the proposed methodology was evaluated with the publicly available CHB-MIT dataset. This method achieved an accuracy of 94.41%, a sensitivity of 94.27%, a specificity of 95.61%. The experimental evaluation indicated that the proposed lightweight solution has acceptably satisfactory prediction performance.