On the predictability of epileptic seizures

https://doi.org/10.1016/j.clinph.2004.08.025Get rights and content

Abstract

Objective: An important issue in epileptology is the question whether information extracted from the EEG of epilepsy patients can be used for the prediction of seizures. Several studies have claimed evidence for the existence of a pre-seizure state that can be detected using different characterizing measures. In this paper, we evaluate the predictability of seizures by comparing the predictive performance of a variety of univariate and bivariate measures comprising both linear and non-linear approaches.

Methods: We compared 30 measures in terms of their ability to distinguish between the interictal period and the pre-seizure period. After completely analyzing continuous inctracranial multi-channel recordings from five patients lasting over days, we used ROC curves to distinguish between the amplitude distributions of interictal and preictal time profiles calculated for the respective measures. We compared different evaluation schemes including channelwise and seizurewise analysis plus constant and adaptive reference levels. Particular emphasis was placed on statistical validity and significance.

Results: Univariate measures showed statistically significant performance only in a channelwise, seizurewise analysis using an adaptive baseline. Preictal changes for these measures occurred 5–30 min before seizures. Bivariate measures exhibited high performance values reaching statistical significance for a channelwise analysis using a constant baseline. Preictal changes were found at least 240 min before seizures. Linear measures were found to perform similar or better than non-linear measures.

Conclusions: Results provide statistically significant evidence for the existence of a preictal state. Based on our findings, the most promising approach for prospective seizure anticipation could be a combination of bivariate and univariate measures.

Significance: Many measures reported capable of seizure prediction in earlier studies are found to be insignificant in performance, which underlines the need for statistical validation in this field.

Introduction

One of the most disabling aspects in epilepsy is the sudden, unforeseen way in which epileptic seizures strike ‘like a bolt from the blue’. Apart from the risk of serious injury, there is often a severe feeling of helplessness that has a strong impact on the everyday life of a patient. It is undisputed that a method capable of predicting the occurrence of seizures would significantly improve the therapeutic possibilities (Elger, 2001) and thereby the quality of life for epilepsy patients. In addition to portable warning systems, automated on-demand application of short-acting drugs as well as electrical stimulation or other intervention strategies could be envisioned.

A question of particular interest is whether apart from clinical prodromi (which are found only in some of the patients (cf. Rajna et al., 1997)) characteristic and objective features can be extracted from the continuous EEG that are predictive of an impending seizure. Much research has been carried out on this topic, and recent studies have reported certain measures derived from the theory of dynamical systems to be to some extent capable of extracting information from the EEG that allow the detection of a preictal state. For an overview of the literature on seizure prediction see Lehnertz and Litt (this issue).

Despite the many publications reporting evidence for the existence of a pre-seizure state, to date no report of a prospective or quasi-prospective prediction of seizures has been published. A major problem with most of the studies presented up to now is that they do not sufficiently (or not at all) investigate the specificity of the described precursors using interictal EEG as control. In addition, many of these studies rely on the use of a posteriori knowledge, e.g. by selecting certain channels out of a large number of channels, or bear the risk of an in-sample over-training of parameters used to calculate measures for the extraction of predictive information. A particular issue that has been neglected by past studies is the need for statistical validation in order to assess the statistical significance of the predictive performance for a given EEG measure (cf. Andrzejak et al., 2003, Mormann et al., 2003b, Winterhalder et al., 2003, Kreuz et al., 2004). Another problem is that up to now no extensive comparison of the performance of different approaches for seizure prediction has been published. Furthermore, there is little experience with continuous long-term-recordings over days, and no study has been reported on comprising different patients from different centers using different pre-surgical evaluation protocols and acquisition systems.

In this paper, we determine the ability of a number of measures to distinguish between the preictal and interictal period using a statistical approach that does not rely on any additional a posteriori knowledge. Retrospectively analyzing continuous multi-day recordings from a group of five patients acquired at different centers, we compare the performance of different univariate and bivariate measures, comprising both linear and non-linear approaches. Applying different smoothing filters to the measures' time profiles and allowing different durations of the preictal period, we evaluate the potential predictive performance of these measures without any antecedent assumptions in the sense of expecting either a preictal increase or a decrease of a measure. We design a number of different evaluation schemes including the analysis of separate channels and seizures and the use of both a constant and an adaptive baseline and pay special attention to issues such as data quality and artifact control as well as statistical validation using seizure time surrogates as a means to assess the statistical significance of an obtained performance value. Based on our results, we discuss the suitability of different measures for seizure prediction.

Section snippets

Patient characteristics and data acquisition

The analyzed recordings were intracranial multi-day recordings from five patients acquired at different epilepsy centers comprising 46 seizures and a total recording time of 311 h. For detailed information on patient characteristics and data acquisition, refer to Lehnertz and Litt (this issue). Since seizure onset times for the different data sets were supplied by the centers providing the data, we used these times although the criteria applied to determine them may have differed among the

Results

The statistical evaluation based on ROC curves is shown in Fig. 1. In this example, the time profile of a measure for phase synchronization (mean phase coherence based on Hilbert Transform RH) for one channel combination of a patient (data set B) is plotted in the upper row of the figure. A smoothing filter of d=5 min was applied and the duration of the preictal period was set to s=240 min in this example. The corresponding amplitude distributions of RH for both the preictal and the interictal

Discussion

When comparing different measures in terms of their suitability for seizure prediction by testing the ability of these measures to discriminate a presumed preictal state from the interictal period, it is useful to distinguish between univariate and bivariate aproaches.

While univariate measures quantify certain properties of the EEG signal thus possibly reflecting the state of a certain region of the brain, the bivariate measures analyzed in this study quantify the degree of synchronization

Acknowledgements

We are grateful to Peter Grassberger and Martin Kurthen for useful discussion and valuable comments. This work was supported by the Deutsche Forschungsgemeinschaft.

References (61)

  • F. Mormann et al.

    Mean phase coherence as a measure for phase synchronization and its application to the EEG of epilepsy patients

    Physica D

    (2000)
  • F. Mormann et al.

    Epileptic seizures are preceded by a decrease in synchronization

    Epilepsy Res

    (2003)
  • E. Pereda et al.

    Interhemispheric differences in awake and sleep human EEG: a comparison between non-linear and spectral measures

    Neurosci Lett

    (1999)
  • P. Rajna et al.

    Hungarian multicentre epidemiologic study of the warning and initial symptoms (prodrome, aura) of epileptic seizures

    Seizure

    (1997)
  • M. Rosenblum et al.

    Phase synchronization: from theory to data analysis

  • M.T. Rosenstein et al.

    A practical method for calculating the largest Lyapunov exponents from small data sets

    Physica D

    (1993)
  • J. Theiler et al.

    Testing for nonlinearity in time series: the method of surrogate

    Physica D

    (1992)
  • M. Winterhalder et al.

    The seizure prediction characteristic: a general framework to assess and compare seizure prediction methods

    Epilepsy Behav

    (2003)
  • A. Wolf et al.

    Determining Lyapunov exponents from a time series

    Physica D

    (1985)
  • R.G. Andrzejak et al.

    Testing the null hypothesis of the non-existence of the pre-seizure state

    Phys Rev E

    (2003)
  • R Aschenbrenner-Scheibe et al.

    How well can epileptic seizures be predicted? An evaluation of a nonlinear method

    Brain

    (2003)
  • S. Blanco et al.

    Stationarity of the EEG series

    IEEE Eng Med Biol

    (1995)
  • C.E. Elger

    Future trends in epileptology

    Curr Opin Neurol

    (2001)
  • C.E. Elger et al.

    Seizure prediction by non-linear time series analysis of brain electrical activity

    Eur J Neurosci

    (1998)
  • D. Gabor

    Theory of communication

    Proc IEEE London

    (1946)
  • P. Grassberger et al.

    Characterization of strange attractors

    Phys Rev Lett

    (1983)
  • P. Grassberger et al.

    Nonlinear time sequence analysis

    Int J Bifurc Chaos

    (1991)
  • B.-L. Hao

    Elementary symbolic dynamics and chaos in dissipative systems

    (1989)
  • M. Hoke et al.

    Spatiotemporal aspects of synergetic processes in the auditory cortex as reavealed by the magnetoencephalogram

  • C. Huygens

    Horoloquium Oscilatorium

    (1673)
  • Cited by (436)

    • EEG-based seizure prediction via Transformer guided CNN

      2022, Measurement: Journal of the International Measurement Confederation
    • The Choice of Evaluation Metrics in the Prediction of Epileptiform Activity

      2024, Communications in Computer and Information Science
    View all citing articles on Scopus
    View full text