Surrogate time series improve the capability of nonlinear measures to characterize the epileptic process.

In our study we calculated a total of eight measures derived from the framework, of nonlinear time series analysis from electroencephalographic recordings of epilepsy patients. As a first set of measures we used absolute values of four different techniques: an estimate of the effective correlation dimension, an estimate of the algorithmic complexity, a nonlinear prediction error, and the course grained flow average. For a second set Of measures we combined the measures of the first set with the method of surrogate data. Retrospectively we analyzed EEG recordings of 38 patients with unilateral mesial temporal lobe epilepsy. All patients suffered from drug-resistant seizures and underwent the pre-surgical diagnosis. By means of a moving window technique and subsequent standard statistical evaluation we investigated to which degree results of the respective measure were capable to determine the side of the seizure generating hemisphere. While absolute values of nonlinear measures had only a weak discriminative power for the spatial distribution of the epileptic process, a correct determination of the focal hemisphere could be obtained in a high percentage of the investigated cases using a combination of these measures with the method of surrogates.