Robust Seizure Prediction Based on Multivariate Empirical Mode Decomposition and Maximum Synchronization Modularity

Abstract

Reliable and timely seizure prediction has been increasingly helpful and indispensable for epileptic patients, ensuring safety and improving life quality. Based on electroencephalogram (EEG), a new patient-specific seizure prediction method is proposed in this paper to detect impending seizures automatically and accurately, using a novel indicator called maximum synchronization modularity. As the first step towards this goal, raw EEG signals are decomposed by multivariate empirical mode decomposition (MEMD). Then graph community detection algorithm is applied to characterize the phase synchronization modularity of sub-band EEG signals. Thus, the deep interaction of scalp electrical activity can be effectively revealed. Finally, radial basis function neural network (RBFNN) is used for the classification. The proposed method achieves an average prediction accuracy of 99.06% and an average sensitivity of 100% on CHB-MIT scalp EEG database, outperforming related works based on the same database.