Stationarity and Redundancy of Multichannel EEG Data Recorded During Generalized Tonic-Clonic Seizures

Abstract

A prerequisite for applying some signal analysis methods to electroencephalographic (EEG) data is that the data be statistically stationary. We have investigated the stationarity of 21-electrode multivariate EEG data recorded from ten patients during generalized tonic-clonic (GTC) seizures elicited by electroconvulsive therapy (ECT). Stationarity was examined by calculating probability density functions (pdfs) and power spectra over small equal-length non-overlapping time windows and then by studying visually and quantitatively the evolution of these quantities over the duration of the seizures. Our analysis shows that most of the seizures had time intervals of at least a few seconds that were statistically stationary by several criteria and simultaneously for different electrodes, and that some leads were delayed in manifesting the statistical changes associated with seizure onset evident in other leads. The stationarity across electrodes was further examined by studying redundancy of the EEG leads and how that redundancy evolved over the course of the GTC seizures. Using several different measures, we found a substantial redundancy which suggests that fewer than 21 electrodes will likely suffice for extracting dynamical and clinical insights. The redundancy analysis also demonstrates for the first time posterior-to-anterior time delays in the mid-ictal region of GTC seizures, which suggests the existence of propagating waves. The implications of these results are discussed for understanding GTC seizures and ECT treatment.