ArticlesAnticipation of epileptic seizures from standard EEG recordings
Introduction
In most patients with epilepsy, seizures occur suddenly and without identifiable external precipitants. The unpredictability of seizure onset is a major threat for people with uncontrolled epilepsy and a cause of disability1 and mortality.2 Prediction of seizure onset, even in the shortterm, would provide time for the application of preventive measures to keep the risk of seizure to a minimum and, ultimately, improve quality of life.
Intracranial recordings of patients who are candidates for surgical treatment offers the most precise access to the emergence of a seizure.3 Ways to anticipate seizure onset before the first intracranial electrical changes have been intensively investigated by conventional linear (ie, frequency) analyses.4, 5, 6 Nevertheless, prediction does not exceed more than a few seconds before visual detection of the seizure. Non-linear analysis is an alternative way to characterise qualitative changes in the dynamics of complex systems7 and could be important in clinical practice.8 Applied to intracranial recordings of patients with temporal-lobe epilepsy (TLE), these methods have shown that the evolution toward a seizure involves not just two states-interictal and ictal-but also a preictal transitional phase of several minutes that is not detected by linear methods.9, 10, 11 This preictal transitional phase could become the basis to anticipate seizures in clinical applications.
To make seizure anticipation practical in real life conditions, and to study types of epilepsy that do not warrant intracranial implantation, we analysed scalpelectroencephalographical (ECG) recordings. Scalp EEG, however, is well known to be subject to signal attenuation, poor spatial resolution, and noise or artifacts,12 which may render delicate13 and even questionable14 the detection of changes with current non-linear measures, originally designed to analyse systems with little noise.
To improve non-linear analysis we have proposed a strategy well suited to track dynamical changes in complex brain signals,11, 15 which measures similarity of EEG dynamics between recordings taken at distant moments in time. This relative measure indicates changes in brain electrical activity with greater accuracy than other nonlinear techniques.16, 17 Furthermore, it has the advantage of being very robust against noise and artifacts, and is fast enough to be carried out in real-time.
In the present study, we have applied this non-linear strategy to analyse scalp-EEG recordings from patients with TLE to assess whether changes in brain dynamics can be detected early enough to anticipate the onset of the clinical seizure. In a subgroup of patients, we validated our results on simultaneous surface and intracranial recordings.
Section snippets
Scalp EEG recordings
We studied a sample of 18 patients with refractory TLE who required continuous EEG recording and video monitoring to localise seizure onsets. Hippocampal sclerosis as defined by magnetic resonance imaging was the most frequent pathological condition associated with TLE (table). Patients were free to move in the recording room.
To avoid changes induced by variation of arousal states,5 only seizures in which the patients were awake (14 seizures) or asleep (four seizures) for the entire preictal
Results
Figure 2 shows a representative example of the results for simultaneous intracranial and scalp recording 50 min before a spontaneous seizure of left-temporal origin. The similarity profiles, relative to a reference state taken at the beginning of the recording, are depicted for selected scalp and depth channels in the left-temporal region (figure 2A). A persistent deviation from the reference state is recorded around 18 min before the seizure in both the scalp and intracranial similarity
Discussion
Our results indicate that preseizure changes in brain dynamics can be detected from recordings of scalp-EEG activity. These changes were characterised with a reference state taken 1 h before the ictus, and were detected several minutes before the earliest clinical or overt EEG manifestations of the seizure. We chose this reference state to analyse only preictal epochs recorded during stable states of wakefulness or during sleep, to avoid changes in the interictal dynamics induced by major
References (27)
- et al.
Mortality from epilepsy: results from a prospective population-based study
Lancet
(1994) - et al.
Does interictal spiking rate change prior to seizure?
Electroencephalogr Clin Neurophysiol
(1991) Nonlinear dynamics for clinicians: chaos theory, fractals and complexity at the bedside
Lancet
(1996)- et al.
Stationarity and nonstationarity in time series analysis
Physica D
(1996) - et al.
IFCN standards for digital recording of clinical EEG
Electrocephalogr Clin Neurophysiol
(1998) - et al.
Endogenous control of epilepsy
Prog Neurobiol
(1994) - et al.
Development of the quality of life in epilepsy inventory
Epilepsia
(1995) Seizure and epilepsy
(1989)- et al.
On the prediction of epileptic seizures
Biol Cybern
(1981) - et al.
Real-time automated detection and quantitative analysis of seizures and short-term prediction of clinical onset
Epilepsia
(1998)