Abstracts

Nonlinear time series analysis allows characterizing dynamical systems in which nonlinearity gives rise to complex and irregular behavior. While several studies indicate that nonlinear methods can extract valuable information from electroencephalographic recordings from epilepsy patients, others doubt their necessity and conjecture that the same information can be obtained using classical linear techniques. To address this issue, we compared these two concepts but included also a combination of nonlinear measures with surrogates, an approach that has been designed to specifically focus on nonlinearity. As a benchmark for the comparison of the different techniques we used the discriminative power to detect the focal hemisphere in unilateral mesial temporal lobe epilepsy. As linear measures we used the relative power in the delta band, the decay time of the autocorrelation function, the skewness of the amplitude distribution, and the Hjorth mobility. As nonlinear measures we used the prediction error, the local flow, an estimate of an effective correlation dimension, and the algorithmic complexity. For each nonlinear measure we defined a corresponding surrogate corrected measure. To this end each nonlinear measure was calculated for the original EEG time series and set of surrogate time series. The surrogate corrected measure was calculated from the difference between the value calculated from the original EEG time series and the mean value obtained for the surrogates. We analyzed intracranial multi-channel EEG recordings (on average130 min per patient) from the seizure-free interval of 29 patients with pharmaco-resistant unilateral mesial temporal lobe epilepsy. For the linear and nonlinear measures we obtained the following numbers of correct lateralizations of the focal hemisphere. Delta power: 24, decay time of the autocorrelation function: 23, skewness: 21, Hjorth mobility: not significant, nonlinear prediction error: 18, the local flow: not significant, correlation dimension: 21, algorithmic complexity: not significant. Hence, the performance of both linear and nonlinear measures was weak if not insignificant. In contrast to this, a high performance was obtained for the surrogate corrected measures. Surrogate corrected local flow: 27 correct lateralizations, surrogate corrected prediction error: 25, surrogate corrected correlation dimension: 26, surrogate corrected algorithmic complexity: 26. Nonlinear methods can be highly relevant for the lateralization of the focal hemisphere in patients with mesial temporal lobe epilepsy, provided that they are combined with surrogates. Hence, focusing on nonlinearity appears as the key to a successful characterization of the spatial distribution of the epileptic process. (Supported by Deutsche Forschungsgemeinschaft)