Elsevier

Clinical Neurophysiology

Volume 116, Issue 10, October 2005, Pages 2480-2489
Clinical Neurophysiology

Detecting temporal lobe seizures from scalp EEG recordings: A comparison of various features

https://doi.org/10.1016/j.clinph.2005.06.017Get rights and content

Abstract

Objective

Sixteen different features are evaluated in their potential ability to detect seizures from scalp EEG recordings containing temporal lobe (TL) seizures. Features include spectral measures, non-linear methods (e.g. zero-crossings), phase synchronization and the recently introduced Brain Symmetry Index (BSI). Besides an individual comparison, several combinations of features are evaluated as well in their potential ability to detect TL seizures.

Methods

Sixteen long-term scalp EEG recordings, containing TL seizures from patients suffering from temporal lobe epilepsy (TLE), were analyzed. For each EEG, all 16 features were determined for successive 10 s epochs of the recording. All epochs were labeled by experts for the presence or absence of seizure activity. In addition, triplet combinations of various features were evaluated using pattern recognition tools. Final performance was evaluated by the sensitivity and specificity (False Alarm Rate (FAR)), using ROC curves.

Results

In those TL seizures characterized by unilateral epileptiform discharges, the BSI was the best single feature. Except for one low-voltage EEG with many artifacts, the sensitivity found ranged from 0.55 to 0.90 at a FAR of ∼1/h. Using three features increased the sensitivity to 0.77–0.97. In patients with bilateral electroencephalographic changes, the single best feature most often found was a measure for the number of minima and maxima (mmax) in the recording, yielding sensitivities of ∼0.30–0.96 at FAR ∼1/h. Using three features increased the sensitivity to 0.38–0.99, at the same FAR. In various recordings, it was even possible to obtain sensitivities of 0.70–0.95 at a FAR=0.

Conclusions

The Brain Symmetry Index is the most relevant individual feature to detect electroencephalographic seizure activity in TLE with unilateral epileptiform discharges. In patients with bilateral discharges, mmax performs best. Using a triplet of features significantly improves the performance of the detector.

Significance

Improved seizure detection can improve patient care in both the epilepsy monitoring unit and the intensive care unit.

Introduction

Although various bodily functions are measured continuously and on a routine basis in the critical care or stroke unit, the brain is typically still monitored by means of routine bedside clinical testing. Even under the best circumstances, this is a discontinuous and subjective process. Furthermore, when patients are sedated, in coma, or medically paralyzed, this method becomes quite uninformative.

There is increasing awareness that several derangements of brain function may occur in critically ill patients, that are difficult, if not impossible, to detect by clinical examination (Jordan, 1999, Vespa et al., 1999, van Putten, 2003a). Furthermore, in various circumstances changes in brain function may occur several hours before clinical signs become apparent. Examples include progressive brain ischaemia, as may occur in particular stroke patients (van Putten and Tavy, 2004b), or in patients suffering from vasospasm after a subarachnoid hemorrhage.

Recently, Jordan, 1999, Vespa et al., 1999 discussed the influence that continuous EEG monitoring (cEEG) in Intensive Care Units (ICUs) has on management decisions. They suggest that in 58% of the cases cEEG had an important influence on the decisions taken within the ICU. Various other studies stress the importance of cEEG in the ICU as well, see e.g. (Vespa et al., 1997, Jordan, 1993, Kay, 1998, Nuwer, 1999, Scheuer, 2002, Claassen et al., 2004, Vespa, 2005).

In the application of cEEG, there is a need for (semi-) automatic EEG analysis (Agarwal et al., 1998, Kull and Emerson, 2005), for instance for the detection of epileptic activity (Gotman, 1999, Jerger et al., 2001, van Putten, 2003a; Faul et al., 2005, Harrison et al., 2005, Vespa, 2005) or ischaemia (van Putten et al., 2004a, van Putten and Tavy, 2004b). This paper discusses various features and their (triplet) combinations to detect seizure activity from scalp EEG recordings. As performance measures, the sensitivity and the false alarm rate (FAR) are used, as discussed in Section 2.

Section snippets

EEG recordings

EEGs were recorded with a Brainlab digital EEG system (OSG bvba, Belgium) using a sampling frequency fs=250 Hz. Analog filter settings were 0.16–70 Hz. The EEGs were recorded with Ag/AgCl electrodes placed at the Fp2, Fp1, F8, F7, F4, F3, T4, T3, C4, C3, T7, T5, P4, P3, O1, O2, Fz, Cz and Pz loci of the 10–20 International System. Impedance was kept below 50 kΩ to avoid polarization effects. All EEGs were band-filtered between 0.5 and 30 Hz before further analysis. Bipolar montages were applied

Results

Patient characteristics are presented in Table 2. In approximately half of the recordings, epileptiform discharges were unilateral, in the remaining recordings, bilateral discharges were present.

In Fig. 2 we show an example of two EEG recordings. The top recording shows a strong EEG asymmetry during seizure activity. In the lower recording, this is less pronounced, and bilateral epileptiform discharges are present. The corresponding ROC curves are shown in Fig. 3, illustrating the differences

Discussion and conclusions

In the application of cEEG, there is a need for (semi-) automatic EEG analysis (Agarwal et al., 1998), for instance for the detection of epileptic activity (Gotman, 1999, Jerger et al., 2001, van Putten, 2003a) or ischaemia (van Putten et al., 2004a, van Putten and Tavy, 2004b).

In this study, we present a detailed comparison of various features and the potential contribution of combining various features in their ability to detect seizure activity in scalp EEG recordings from patients suffering

Acknowledgments

We thank Prof W. Paeschen and Dr L. Lagae from the Universitair Ziekenhuis Gasthuisberg, Leuven, Belgium, for kindly providing the EEG data and OSG bvba, Belgium, for developing the software tool for conversion of the EEG Brainlab data to the MatLab environment. We also thank the anonymous referees for their useful suggestions and comments.

References (42)

  • A. Shoeb et al.

    Patient-specific seizure onset detection

    Epilepsy Behav

    (2004)
  • M.J.A.M. van Putten

    Proposed link rates in the human brain

    J Neurosci Methods

    (2003)
  • M.J.A.M. van Putten et al.

    A brain symmetry index (BSI) for online EEG monitoring in carotid endarterectomy

    Clin Neurophysiol

    (2004)
  • P. Vespa et al.

    Early detection of vasospasm after acute subarachnoid hemorrhage using continuous EEG ICU monitoring

    Electroencephalogr Clin Neurophysiol

    (1997)
  • S. Wilson et al.

    Seizure detection: evaluation of the reveal algorithm

    Clin Neurophysiol

    (2004)
  • J. Bhattacharya

    Complexity analysis of spontaneous EEG

    Acta Neurobiol Exp

    (2000)
  • J. Claassen et al.

    Detection of electrographic seizures with continuous EEG monitoring in critically ill patients

    Neurology

    (2004)
  • D. Gabor

    Theory of communication

    J IEE

    (1946)
  • J. Gotman

    Automatic detection of seizures and spikes

    J Clin Neurophysiol

    (1999)
  • J. Gotman

    The treatment of epilepsy: principles and practice

    Ch. computer-assisted data collection and analysis

    (2001)
  • J. Iriarte et al.

    Independent component analysis as a tool to eliminate artifacts in EEG: a quantitative study

    J Clin Neurophysiol

    (2003)
  • Cited by (71)

    • Data fusion for paroxysmal events’ classification from EEG

      2017, Journal of Neuroscience Methods
      Citation Excerpt :

      The selection of features to be extracted was based on the existing literature in an attempt to include features that have been widely used for the analysis of EEG signals. In order to provide an insight into the extracted features we refer to the corresponding studies (Greene et al., 2007; Pippa et al., 2015; van Putten et al., 2005; Petrantonakis and Hadjileontiadis, 2010; Valderrama et al., 2010; Le Van Quyen et al., 1999). Statistical features such as minimum, maximum, mean, variance, standard deviation, percentiles, interquartile range, mean absolute deviation, range, skewness, kyrtosis and energy are used to capture the variations in the amplitude of the EEG signals that accompany the electroencephalographic seizure activity (van Putten et al., 2005; Petrantonakis and Hadjileontiadis, 2010; Valderrama et al., 2010).

    View all citing articles on Scopus
    View full text