Abstracts

RATIONALE: Previous studies have demonstrated the usefulness of a number of nonlinear and linear time series analysis techniques to gather information about features of brain electrical activity predictive of an impending seizure. We investigated the relative merit of these techniques in order to estimate the extent of redundant or complementary information yielded for an anticipation of seizures. METHODS: Intracranial multichannel EEG recorded pre-ictally in patients, in whom epilepsy surgery led to complete seizure control, were probed for indicators predictive of imminent seizure activity. Long-lasting interictal EEG recordings, randomly selected to include different states of vigilance served as control. Time series analysis techniques comprised power-spectral estimates as well as measures for spatio-temporal synchronization, complexity, nonlinearity, and determinism. RESULTS: Measures yielded different degrees of performance in recognizing a pre-seizure state. For different measures, failures were generally found in different patients. In addition, time-scales on which predictive features were observed varied greatly among measures. CONCLUSIONS: Results suggest that most measures reflect different aspects of the underlying dynamics of the epileptogenic process that lead to an increased probability of an ictal state. Thus, a synopsis of different characterizing measures seems to be most appropriate for anticipating epileptic seizures. (Supported by the Deutsche Forschungsgemeinschaft)