Abstracts

A rapidly growing number of studies deals with the prediction of epileptic seizures. For this purpose, various techniques derived from linear and nonlinear time series analysis have been applied to the EEG of epilepsy patients. In none of these works, however, the performance of the seizure prediction statistics is tested against a null hypothesis, an otherwise ubiquitous concept in science. In consequence, the evaluation of the reported performance values is problematic. Here we propose the technique of seizure time surrogates based on a Monte Carlo simulation to remedy this deficit. Our approach allows to formally test the null hypothesis of the non-existence of the pre-ictal state.
To illustrate the technique of seizure time surrogates we analyzed the spatiotemporal distribution of a nonlinear measure ([xi]) that was calculated from the intracranial EEG of patients with mesial temporal lobe epilepsy. Seizure time surrogates were constructed by replacing the original seizure times with times randomly chosen from the interictal intervals. The total number of seizures and the distribution of intervals between consecutive seizures were imposed as constraints on the seizure time surrogates. From the time profiles of [xi], a simple evaluation statistics was calculated for the original seizure times as well as for an ensemble of nineteen seizure time surrogates. The measure [xi] together with the applied evaluation statistics are referred to as seizure prediction statistics, the performance of which is denoted by R. Provided that a pre-ictal state exists and that our prediction statistics is be able to detect it, highest values of R should be obtained for the original seizure times.
We obtained a rather high value of R for the original seizure times. At the first glance, this finding could be regarded as evidence for the existence of a pre-ictal state and the capability of the seizure prediction statistic to detect it. However, the value of R obtained for the original seizure times was well within the distribution of R values obtained for the seizure time surrogates. Hence, our technique allowed to successfully disclose the insignificance of a high value of the performance of a seizure prediction statistics.
The aim of the present study was not to prove or disprove the existence of a pre-ictal state. Neither was our aim to propose a certain seizure prediction statistics. Rather, the aim was to introduce the technique of seizure time surrogates that allows to validate the performance of any given seizure prediction statistics. We have provided an example where this technique helped to disclose tantalizing results of a seizure prediction statistics as misleading. In future applications, we expect seizure time surrogates to be a powerful tool to differentiate statistics unsuited for a detection of the pre-ictal state from more promising approaches.