Abstracts

Rationale: Seizures are abnormal neurophysiological events which cause a characteristic hyper-synchronization of brain networks as they evolve. There is evidence that at least for some types of seizure, related neural activity begins to modulate networks minutes if not hours prior to clinical onset. However, consistent and robust estimation of seizure precursors from electroencephalograms (EEG) has proved to be a difficult problem. In addition to the inherent variability of EEG signals, seizure heterogeneity and limited specificity of some proposed measures of precursory activity may significantly affect the accuracy of seizure prediction. Information theory provides an attractive framework for estimating potential pre-ictal network coordination using probabilistic measures. Methods: We estimated information theoretic parameters of interaction between networks from pre-ictal and ictal EEGs from 7 patients with multiple focal seizures, with temporal or frontal onset. These parameters included time-dependent mutual information, conditional entropy and net information. We also estimated corresponding measures from baseline EEGs, clinically identified as non-seizure related. A total of 36 seizures were analyzed and at least two baseline segments from each patient at two spectral intervals, below and above 100 Hz, respectively. Results: Information theoretic parameters, including relative and conditional entropy and mutual information estimated from high-frequency EEG, were specifically modulated in the pre-ictal interval. These parameters were statistically identical to those in the ictal interval, but distinct from baseline. In contrast, corresponding lower-frequency parameters varied non-specifically between the three intervals. Mutual information and conditional entropy in the pre-ictal interval monotonically increased, in the same way as during the ictal interval, particularly in frontal seizures, which suggests that brain networks may become increasingly hyper-synchronized prior to seizure onset. Conclusions: Resting brain networks are specifically modulated prior to clinical seizure onset. Information theoretic parameters quantify precursory neurodynamic changes in the high-frequency EEG, and may be thus be used for seizure prediction. In contrast, lower frequency network interactions change non-specifically at baseline and pre-ictal intervals and may, therefore, not be robust measures of seizure-related precursory neural activity. These findings have important implications for next-generation therapeirs in epilepsy, particularly for the optimization of automated seizure prevention and drug-delivery systems.