Characterizing dynamics of absence seizure EEG with spatial-temporal permutation entropy

Abstract Characterizing transient brain dynamics prior to seizures is a main challenge in absence epilepsy study. As brain is a chaos dynamical system, many complexity based methods have been used to track the dynamical changes of absence seizure EEG. However, most of these methods treat multichannel EEG recordings as a set of individual time series, which will inevitably lead to the loss of crucial cross-channel correlation in the epileptic network. Recently, a spatial-temporal permutation entropy method called multivariate multiscale permutation entropy (MMPE) was proposed to measure the complexity of multichannel data. In this study, MMPE was applied to multichannel EEG for characterizing dynamics of absence seizure. It was found that the pre-ictal EEG exhibited a significant lower MMPE value than interictal EEG, and a significant higher MMPE value than the ictal EEG, indicating that the complexity of multichannel EEG decreased in the transition of brain activities. This finding confirmed the existence of a pre-seizure state in absence epilepsy. The identification ability of MMPE was tested against its original univariate complexity measures: permutation entropy (PE) and multiscale permutation entropy (MSPE), and another multivariate multiscale entropy: multivariate multiscale sample entropy (MMSE). After evaluating the performance by four classifiers (Decision Tree, K-Nearest Neighbor, Discriminant Analysis, Support Vector Machine), MMPE can achieve accuracy of 87.2% at least, which is about 15%, 12%, and 10% higher than that of PE, MSPE and MMSE. Hence, this work supports the view that EEG has a detectable change prior to an absence seizure, and MMPE could be considered as a candidate precursor of the impending absence seizures.