Brain, Vol. 125, No. 3, 640-655,
March 2002
© 2002 Guarantors of Brain
Seizure anticipation in human neocortical partial epilepsy
,11 Laboratoire de Neurosciences Cognitives et Imagerie Cérébrale (LENA), CNRS UPR 640, 2 Unité d’Epileptologie and 3 Service de Neurochirurgie, Hôpital de la Pitié-Salpêtrière, Paris
Correspondence to: Professor Michel Baulac, Unité d’Epileptologie, Clinique Paul Castaigne, Hôpital de la Pitié-Salpêtrière, 47 boulevard de l’Hôpital, 75651 Paris cedex 13, France E-mail: michel.baulac{at}psl.ap-hop-paris.fr Deceased May 28, 2001
Received February 22, 2001. Revised July 9, 2001. Second revision October 1, 2001. Accepted October 22, 2001.
 |
Summary |
|---|
 Top Summary IntroductionMethodsResultsDiscussionIn MemoriamReferences |
|---|
The transition of brain activity towards an epileptic seizure is still a poorly understood phenomenon. Dynamic changes in brain activity have been detected several minutes before seizure emergence in populations of patients with mesial temporal lobe epilepsy (MTLE), using non-linear analysis of intracranial EEG recordings. Similar detection of a pre-ictal state has been obtained with standard scalp EEG recordings using a modified non-linear method. Here we applied this strategy to the seizures of patients with neocortical partial epilepsy. Results obtained by non-linear similarity analysis of 41 seizures from 11 patients with refractory partial epilepsy originating from various sites of the neocortex showed that (i) a pre-ictal state was detected in 90% of the patients and in 83% of the seizures whatever their location, with a mean anticipation time of 7.5 min; (ii) similar pre-ictal dynamic changes were detected when non-linear analysis methods were applied to either intracranial or scalp EEG recordings; (iii) the recording sites exhibiting these pre-ictal changes were distributed both within the epileptogenic focus and at remote locations; (iv) most pre-ictal dynamic changes were not correlated with linear changes in the frequency spectrum or with changes in the visually inspected EEG and the patients’ behaviour. Hypotheses on the neuronal mechanisms underlying the pre-ictal period are discussed. The present results, together with those recently obtained in an MTLE population, suggest that changes in pre-ictal dynamics are a general phenomenon associated with seizure emergence in a wide population of patients with partial epilepsy, wherever the epileptogenic focus is located. The possibility of anticipating the onset of seizures has considerable therapeutic implications.
Keywords: neocortical epilepsy; scalp electroencephalogram; intracranial electroencephalogram; non-linear analysis; seizure anticipation
Abbreviations: ANOVA= one-way analysis of variance; MTLE = mesial temporal lobe epilepsy
|
Introduction |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionIn MemoriamReferences |
|---|
For patients with intractable epilepsy, the unpredictability of seizure occurrence underlies an increased risk of sudden unexpected death (Cockerell et al., 1994
) or morbidity (Buck et al., 1997) and is a major factor contributing to a poor quality of life (Devinsky et al., 1995). Anticipation of seizures would permit their consequences to be minimized and also allow therapeutic measures to be taken before seizures occur. Traditional approaches for detecting pre-ictal changes have encountered difficulties. Clinical distinctions between premonitory sensations and the aura are delicate (Rajna et al., 1997), and if they are prodroma they may imply very subtle and variable feelings even in the same patient. Only a few fortuitous (Baumgartner et al., 1998) or invasive (Weinand et al., 1997) studies using functional cerebral imaging have provided evidence for pre-ictal changes of regional cerebral blood flow in patients with mesial temporal lobe epilepsy (MTLE). Traditional analyses of EEG signals have failed to detect specific changes preceding seizures, either by visual inspection of the recordings, even during intracerebral investigation, or using linear methods based on spectral analysis (Rogowski et al., 1981; Katz et al., 1991).
Strategies of non-linear analysis have succeeded in demonstrating pre-ictal changes in the intracerebral EEG in patients with MTLE (Lehnertz and Elger, 1998; Martinerie et al., 1998). Changes in EEG dynamics, extracted by different non-linear measures, were detected several minutes before the onset of the seizure, when patients reported no specific sensations and visual inspection of the recording showed no clear changes in the EEG signal (Martinerie et al., 1998). Non-linear methods were developed initially to describe the dynamics of complex physical systems with non-linear components, implying that their time course does not follow the linearity of the classical deterministic laws, but may instead exhibit non-proportional responses to specific inputs. Self-organizing behaviour and intermittency are other interesting properties of non-linear systems which permit transitions between states in the absence of external triggers. The epileptic process is also chronically intermittent and spontaneous, except for rare reflex epilepsies, and mathematical tools for the characterization of non-linear deterministic systems have been used successfully to determine changes in the state of brain activity before seizures (Schiff, 1998).
Recently, we developed a new measure of non-linear analysis—the similarity index—which improves the robustness of the method, increases calculation speed and reduces sensitivity to EEG artefacts (Le Van Quyen et al., 1999). Applied to intracerebral recordings, this method allows accurate description of the spatiotemporal changes in the EEG before the seizure (Le Van Quyen et al., 2000). We have shown that this method can efficiently detect pre-ictal changes from the scalp EEG in spite of the blurring due to volume conduction and skull filtering (Le Van Quyen et al., 2001).
These studies were mostly performed in patients with MTLE, who constitute the most homogeneous population of adults with partial epilepsy. The population of patients with neocortical partial epilepsies is much less homogeneous because of the diversity of clinical manifestations, underlying causes and electrographic characteristics (Lee et al., 2000). Neocortical partial epilepsy accounts for 
68% of the cases of partial epilepsy in adults (Semah et al., 1998). Patients with medically refractory neocortical epilepsy benefit less from surgical treatment than those with MTLE and the postoperative prognosis is more variable (Engel, 1996; Spencer, 1998). This is due to difficulties in the accurate localization of the epileptogenic focus, to the presence of multiple foci or to the involvement of functional areas. The development of strategies to anticipate seizures could therefore improve the quality of life of these patients considerably. Non-linear analysis of interictal EEG recordings from patients with lesional neocortical epilepsy showed a relationship between the presumed epileptogenic area and the maximum values of a non-linear index (Widman et al., 2000). Neocortical seizures originating from the frontal lobe develop suddenly and propagate rapidly, in contrast to the slower progressive seizure emergence that occurs in MTLE, suggesting that their anticipation might be more difficult. Here we show, for the first time, that non-linear analysis methods can detect pre-ictal changes in EEG dynamics, from either scalp or intracerebral recordings, in patients with various types of neocortical epilepsy.
|
Methods |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionIn MemoriamReferences |
|---|
Patient population
We studied 11 patients presenting medically intractable neocortical partial epilepsy (Table 1). They were selected from patients examined at the Epilepsy Unit of La Pitié-Salpêtrière Hospital in Paris between 1993 and 2000 according to the following criteria: (i) neocortical origin of the epilepsy demonstrated by intracranial EEG recordings (n = 10) or scalp EEG recording (n = 1) in agreement with clinical and imaging data; (ii) spontaneous seizures, excluding seizures preceded by classical precipitating factors (hyperventilation and photic stimulation); (iii) a stable state of vigilance during the recording periods. EEG recordings from patients who were continuously awake (n = 22) or asleep (n = 19), on the basis of videotape recordings, were used for analysis to avoid perturbations due to changes in the clinical state. All patients required long-term continuous scalp EEG recordings and video monitoring using 21 or 27 electrodes, placed according to the extended International 10–20 System. For 10 patients, intracranial recordings were then necessary to localize accurately the structures generating seizure onset. Intracerebral electrodes with multiple contacts explored the lobe suspected of containing the epileptogenic focus and adjacent lobes. Subdural strips were added in some patients to sample the lateral or inferior cortices of the temporal or occipital lobe. Subdural grids with 36 contacts were used in two patients. The post-implantation location of the electrodes was determined by MRI. All patients gave informed consent to participation in this study, which was approved by the Consultative Committee for the Protection of Patients in Biomedical Research of the Pitié-Salpêtrière Hospital, Paris.
|
The video-EEG recordings were performed with a 32-channel BMSI system (Nicolet-BMSI, Madison, Wis., USA). The raw data were digitized at 200 Hz with 12-bit resolution, and were passed to a channel amplifier system with band-pass filter settings of 0.5–99 Hz using an external reference over linked ears. Our study was based on 41 seizures extracted from long-term EEG recordings of 60–120 min duration that included an interval of 50–90 min preceding the seizure. For all but three patients, both scalp and intracranial recordings were available for analysis. Seizure onset was defined as the time of the first clinical symptoms (including aura sensation) or the time of the first significant electrographic changes (burst of spikes, sinusoidal waves or low-voltage fast activity) if they occurred earlier. Additionally, long interictal intracranial EEG periods (n = 8) of 90–450 min without seizure were analysed for five patients in order to study the behaviour of the non-linear indexes far from a seizure in a stable state of vigilance.
Non-linear EEG analysis
Mathematical details of the non-linear method have been described previously (Le Van Quyen et al., 1999). We present here the main steps of the strategy, which was carried out independently for each EEG channel after digital filtering (1–98 Hz for intracerebral recordings and 1–48 Hz for scalp recordings).
Reconstruction of EEG dynamics using time intervals
The dynamics of the EEG signal was evaluated for successive windows of 30 s duration. One novel feature of this method was to reduce the EEG signal to pure phase information. For this purpose, we used the sequence of time intervals between positive-going crossings of the signal through the electrical zero. The EEG dynamics was reconstructed by the delay method with m-dimensional embedding (m is the number of dimensions of the reconstructed dynamics), using m consecutive time intervals obtained by the zero-crossing method (m = 16) (see Fig. 1 in Le Van Quyen et al., 2001). Then, in order to retain the main components of the dynamics and to reduce the noise level, the embedding dimension was reduced to m = 4 using a singular-value decomposition (Albano et al., 1988).
|
Measurement of similarity between dynamics from a reference state and from test EEG windows
The dynamics corresponding to each successive EEG window were compared with that of a reference window. A long reference period was taken from the start of the recording; it preceded the seizure by at least 50 min and had a duration of 300 s. This period contains different features of the interictal activity, possibly including artefacts and spikes, which were taken into account. The reference window started at the same time for all channels. In order to allow comparison of dynamics between this long reference window and shorter test windows (30 s), we identified the most frequent occupations of the phase space flow by random selection of a subset of points in the reference dynamics reconstruction.
The corresponding dynamics of the two EEG windows were then compared using the normalized cross-correlation sum (Manuca and Savit, 1996; Schreiber and Schmitz, 1997) at a radius scale chosen at 30% of the cumulative neighbourhood distribution of the reference set. We used a similarity index, ranging from 0 to 1, to provide a sensitive measure of closeness of dynamics in the two EEG windows; values close to 1 indicate the greatest degree of similarity in the dynamics. The time course of this index shows how the long-term dynamics of the signals from each EEG channel change before seizure onset.
Statistical significance of the pre-ictal deviations of the similarity index
We assessed the statistical significance of pre-ictal changes by quantifying deviations from the reference state. The significance 
of the deviation was defined by the ratio (
– µ)/
(µ and correspond to the mean and standard deviation of similarity variations during the reference period), whose P value was given by the Chebyshev inequality (for any statistical distribution of , P(|| 
k) 
1/k2, where k is the chosen statistical threshold). The temporal evolution of the statistical significance was used to detect the instant at which the similarity index reached a critical level. We characterized this instant as anticipation time when the ratio reached a critical level of k = 5 (P = 0.04 for the bilateral test) and remained at or above this threshold until the seizure.
Traditional EEG analysis
Visual inspection of the EEG recordings
All recordings were examined to define the electrographic onset of the seizure accurately. We also searched for changes in EEG activity of any channel (including modifications of background rhythms or of interictal epileptic activity, or the presence of sustained artefacts) in comparison with activity during the reference period used for the non-linear analysis (Table 2).
|
Spectral analysis
Fast Fourier transforms were performed on successive EEG windows of 10 s duration to analyse the power spectrum of the channels that showed the most significant pre-ictal changes according to the non-linear analysis.
Statistical analysis
Non-parametric tests were used to compare electroclinical data and anticipation times. The significance of differences in anticipation times from scalp and intracranial EEG data or during wakefulness versus sleep was evaluated with the Mann–Whitney U-test. Kruskal–Wallis one-way analysis of variance (ANOVA) was used to compare anticipation times according to the types of clinical seizure. Spearman’s coefficient of rank correlation was determined to correlate anticipation times with the initial frequency or the duration of the epileptic discharges. Two-way ANOVA was used to study whether the location of the epileptogenic focus was correlated with the percentage of perifocal or distant EEG channels from which pre-ictal changes were detected. The criterion for statistical significance was P < 0.05.
|
Results |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionIn MemoriamReferences |
|---|
Representative examples of the spatiotemporal dynamics of the pre-ictal EEG changes
We first present examples of the non-linear analysis of scalp (Fig. ) and intracerebral EEG (Fig. 2) from the same patient (Patient 10), who had a left parieto-occipital focus with no underlying abnormality detected by MRI. Our analyses identified a pre-ictal state preceding distinct seizures in both sets of EEG recordings.
|
The scalp EEG recording contains a seizure that occurred during spontaneous sleep near the end of the night. The background activity showed continuous slow waves (
2.5 Hz), but no changes were apparent from visual inspection of the recording before the seizure (Fig. A). The dynamics of the EEG signal from each of the 21 electrodes were analysed independently by a non-linear measure of similarity between a reference period and successive 30 s segments of the recording. The time course of the similarity indices for all channels showed high deviations from the baseline during the seizure and the post-ictal state (Fig. C). The EEG dynamics then returned to a state similar to the reference. Interestingly, similarity indices for several channels shifted to reduced levels for several minutes before the seizure. In order to study the spatiotemporal dynamics of these changes, the statistical significance of these deviations from the baseline was calculated and dynamic changes up to 5 SD were represented for each of the channels (Fig. D). The statistical map demonstrated widespread pre-ictal changes in dynamics. We defined this period as the pre-ictal state and calculated an anticipation time from the electrode that showed the earliest of the dynamic changes that persisted until the seizure. The Fz channel allowed retrospective anticipation of seizure onset by 15 min. Intracranial EEG recordings were then performed in this patient using five intracerebral electrodes to explore the left parieto-occipital junction and the temporal lobe. Intracranial EEG signals were analysed using the same non-linear method. During the recording shown in Fig. , the patient was watching television or playing electronic games. The time course of the similarity indices (Fig. C) from all 32 channels showed sharp decreases in similarity during the simple partial seizure, but decreases also occurred during two periods before the seizure (from 22 to 41 min, and from 43 to 56 min). The statistical map showed that these deviations were highly significant for several channels (Fig. D). We measured the anticipation time from the second of the dynamic changes (43 to 56 min) because of the transient break between these periods of changed dynamics. The first contact of the precuneus electrode (PC1) detected changes in dynamics 13.5 min before seizure onset. All channels with pre-ictal changes were involved in the initial spread of the seizure. They were either adjacent to the focus or in the neocortical part of the temporal lobe (Fig. E). No significant change in dynamics occurred in the signal from the channel closest to the seizure onset (PC3). Visual inspection of the EEG revealed intermittent spikes for these channels without specific changes before the seizure (Fig. A), but fast Fourier transform analysis of the PC1 channel signal revealed that the frequency spectrum did change (Fig. B).
Detection of pre-ictal changes and anticipation time
Similar analyses were performed in 11 patients with medically refractory partial epilepsy whose presurgical investigations demonstrated a neocortical origin (Table 1). For eight patients, we studied several recordings that contained a seizure, from both the scalp and the intracerebral EEG (Table 2). In 10 patients, pre-ictal changes in EEG dynamics were detected before at least one seizure. Pre-ictal changes were found in 34 (83%) of the 41 seizures analysed. The mean percentage of anticipated seizures per patient was 79 ± 9.5%. The mean anticipation time for these 34 seizures was 7.54 ± 1.15 min.
We then studied factors that might influence the detection of pre-ictal changes and the anticipation time (Table 2). First, we could detect no differences according to the location of the epileptogenic focus in the brain (Table 2). Secondly, pre-ictal changes were detected as efficiently from scalp EEG recordings as from intracerebral EEG recordings (eight out of 10 seizures were anticipated using scalp recordings and 26 out of 31 seizures were anticipated using intracerebral recordings) and the mean anticipation times were not statistically different (6.44 ± 1.71 and 7.88 ± 1.42 min for scalp and intracranial recordings, respectively). Thirdly, we examined how electroclinical data might influence the variability with which pre-ictal changes were detected in the same patient. (i) The arousal state of the patient during the recording did not modify our ability to anticipate seizures (16 out of 19 seizures anticipated during sleep and 18 out of 22 seizures anticipated during wakefulness), and mean anticipation times in the two states were not statistically different (although a tendency was observed: 10 ± 1.99 min during sleep and 5.33 ± 1.08 min during wakefulness; P = 0.06). (ii) The type of seizure did not influence seizure anticipation (eight out of 10 simple partial seizures, 18 out of 22 complex partial seizures and two out of three secondary generalized seizures were anticipated). Differences in anticipation times for the different types of seizure (2.6 ± 0.73, 6.29 ± 1.23, 17.25 ± 2.2 and 15.25 ± 11.75 min for simple, complex partial, indefinite partial and secondarily generalized seizures, respectively) were not statistically significant [Kruskal–Wallis ANOVA; 
2(3) = 2.94, P > 0.05]. (iii) The duration of seizures, defined electrographically, was not correlated with the anticipation time (Spearman’s r = 0.05). However, we did observe a significant correlation between the anticipation time and the initial frequency of the ictal discharge (Spearman’s r = 0.36, P = 0.03), suggesting that the longer the anticipation time the faster the initial ictal frequency.
We could identify several features of the minority of seizures (seven out of 41) which were not anticipated. First, we observed sudden and abrupt declines in the similarity indices at seizure onset, contrasting with the stability of the dynamics before onset (seizures G and H of Patient 1; seizure E of Patient 2). In one seizure (seizure A of Patient 5), similarity indices changed just before the seizure but not sufficiently to reach our statistical threshold. In three cases, the dynamics of scalp and/or intracerebral EEG (seizures A and B of Patient 9; seizure C of Patient 6) changed significantly before the seizure but returned to subthreshold values several minutes before the seizure began.
Spatial distribution of the pre-ictal changes
Calculation of the non-linear similarity index independently for each channel permitted a spatiotemporal analysis of EEG dynamics. For seizures with a pre-ictal change in dynamics, the number of channels implicated in the pre-ictal state varied with a mean of 6.32 ± 0.91. Pre-ictal changes were detected in fewer than three channels for seven seizures and in more than 12 channels for four seizures. We observed several spatiotemporal patterns of pre-ictal changes. We evaluated the spatial distribution of pre-ictal changes by comparing the percentage of channels in the epileptogenic focus (those channels implicated in the seizure onset) that showed pre-ictal changes with the percentage at sites distant from the focus (Table 3 and Fig. 3). Pre-ictal dynamical changes were detected in 25.93 ± 5.07% of channels in the focus and 22.81 ± 3.64% of the remote channels. Interestingly, the spatial distribution of the pre-ictal changes varied with the location of the focus (Fig. ). For seizures originating in the temporal, parietal or occipital lobes, the higher proportion of channels showing pre-ictal changes was higher for channels in the focus (33.10 ± 7.29%) than for those at a distance (15.87 ± 2.85%), whereas the opposite was found for seizures of frontal origin (14.36 ± 4.87 and 34.02 ± 7.51%, respectively). Two-way ANOVA showed that this interaction between the site of the focus and the spatial distribution of the pre-ictal changes was statistically significant [F(1,32) = 12.37, P = 0.001].
|
|
Electroclinical correlation with the pre-ictal changes
We evaluated temporal correlations between changes in EEG dynamics detected by the non-linear similarity index and modifications of three parameters: visual inspection of EEG recordings, the frequency spectrum of the EEG signal (a classical linear analysis) and the behaviour of the patient, as monitored on the video recordings (Table 2).
No modification of the EEG was observed before most seizures (n = 28). For the 13 other seizures, we detected concomitant changes of the EEG signal visually (n = 9) and/or on examination of the frequency spectrum (n = 11). These linear changes were highly variable in terms of frequency range and dynamics (disappearance or emergence of one or several frequency bands) (Table 2).
Among the awake patients (n = 22), behavioural changes sometimes occurred simultaneously with the non-linear changes in the EEG (n = 4). They involved responses to medical questions (n = 2), use of the telephone (n = 1) or playing electronic games (n = 1). Apart from the last of these behavioural changes, the correlations may have been coincidental. Changes in EEG dynamics in the game-playing patient began when he started playing, disappeared when he stopped and recommenced when he played again (Fig. C). Nevertheless, analysis of other video-EEG recordings from the same patient permitted us to exclude the use of the electronic game as a systematic factor precipitating the seizures.
Specificity of the method
In order to study the specificity of the changes in dynamics detected before the seizures, we analysed long EEG periods without a seizure. The time course of the similarity indices for all channels showed high stability of the non-linear dynamics when the vigilance state of the patient was stable (Fig. 4). Few statistically significant changes occurred; they were very brief (mean duration 57.3 ± 13.2 s) and largely restricted to a single channel, so that the average number of false positives (defined as significant changes lasting >1 min and concerning more than two channels simultaneously) was 0.31 ± 0.20 per hour.
|
On the other hand, when major changes in vigilance occurred, such as the transition from wakefulness to deep sleep, profound changes in EEG rhythms necessarily modified measurements of the dynamics. If the reference state were to be taken during wakefulness, the similarity index would be expected to show deep and sustained deviations when the patient fell asleep. For this reason, we restricted our study to EEG periods before seizure in a stable state of vigilance.
|
Discussion |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionIn MemoriamReferences |
|---|
Seizure anticipation in partial epilepsy
Our results demonstrate that, in addition to the interictal, ictal and post-ictal states, a pre-ictal state can be detected from the brain dynamics of patients with neocortical partial epilepsy. Analysis of EEG signals by non-linear similarity measures allowed hidden properties to be extracted from the electrical signal from the brain, showing that seizures are generally preceded by progressive changes in dynamics. This method permitted identification of pre-ictal EEG changes in most seizures of patients with intractable partial epilepsy originating in the neocortex (34 out of 41 seizures) as well as the mesial temporal lobe (Le Van Quyen et al., 2000
, 2001). Mean anticipation times were of the same order (7 min) for neocortical epilepsy and MTLE (Le Van Quyen et al., 2001).
Scalp and intracranial EEG analyses
Strikingly, our method possessed a similar efficiency when applied to recordings from either scalp or intracranial EEG. In the present study, we showed that the number of seizures anticipated and the anticipation times did not differ in analyses of scalp and later intracranial EEG data from the same patients. Similar results have been obtained recently from MTLE patients in comparisons of simultaneously recorded scalp and intracranial EEGs (Le Van Quyen et al., 2001). Possibly, the phase information contained in EEG signals, crucial to our similarity index, is transmitted faithfully to surface electrodes, whereas the amplitude of deep signals is considerably attenuated because of the interposition of the skull. The ability to identify a pre-ictal state by a standard, non-invasive route has numerous clinical applications (see below).
Influence of pathological parameters
Our finding that neither the location in the neocortex nor the aetiology of these neocortical epilepsies (developmental versus mature pathology, with or without MRI abnormalities) appears to influence our capacity to detect pre-ictal changes indicates that the dynamic route towards the seizure is generic, even if the neuronal networks are affected differently (Spencer, 1998).
Meaning of the pre-ictal state
Two main hypotheses might explain the occurrence of dynamic changes before a seizure. The first involves a ‘route’ towards the seizure, corresponding to the progressive recruitment of neuronal activities until a critical mass has been reached, leading at this moment to the emergence of the seizure (Martinerie et al., 1998). The second hypothesis involves a ‘facilitating’ state, corresponding to a change in brain activity that serves as a background from which a seizure could easily and suddenly appear (Le Van Quyen et al., 2001). It might consist of an impairment of the balance between neuronal excitability and its inhibitory controls (Lopes da Silva and Pijn, 1995). However, the identity of the neuronal networks involved (corticocortical and/or thalamocortical interactions) and the nature of the cellular modifications (synaptic or electrotonic interactions and/or changes in intrinsic cellular properties) remains to be uncovered.
Topological distribution of the pre-ictal changes in dynamics
The spatial dynamics of pre-ictal changes may permit discrimination between these two pathophysiological hypotheses. Pre-ictal changes predominant in channels in the epileptogenic focus might support the ‘route to seizure’ hypothesis, as we observed for the patients with neocortical temporal, parietal and occipital lobe epilepsies. On the other hand, widespread pre-ictal changes remote from the focus are suggestive of the ‘facilitating state’ hypothesis, as in patients with frontal lobe epilepsies. These results, together with those of studies of patients with temporal lobe epilepsy (Le Van Quyen et al., 2001) and studies in which measurements of cerebral blood flow and metabolism showed changes remote from the epileptogenic focus, support the concept of a large-scale ‘epileptic network’ distributed in the brain beyond the epileptogenic area during the pre-ictal state (Weinand et al., 1997; Baumgartner et al., 1998) or the interictal state (Juhasz et al., 2000) in location-related epilepsy.
Clinical and electrical correlations with pre-ictal changes in dynamics
Temporal correlations between changes in classical linear parameters (frequency spectrum) or in the patient’s behaviour, and changes in the non-linear similarity index may also provide pathophysiological indications. In awake patients, behavioural changes were sometimes correlated temporally with changes in EEG dynamics. Analysis of seizure E of Patient 10 tends to support the facilitating state hypothesis: changes in cognitive activities may affect brain dynamics by modifying neuronal networks and thus favour seizure emergence. We have reported previously that non-linear methods can detect subtle modulations of continuous epileptic activity during the performance of various mental tasks (Le Van Quyen et al., 1997). Nevertheless, the phenomenon observed in Patient 10 differs from classical reflex epilepsy in that activation was not related to specific cognitive tasks (higher mental activities or visual stimulation), it did not lead reproducibly to a seizure and the neuropsychological EEG activation concerned idiopathic generalized epilepsies almost exclusively (Matsuoka et al., 2000).
Changes in the EEG frequency spectrum preceded some seizures in sleeping patients. EEG rhythms slowed (by enhancement of the delta frequency and/or disappearance of fast frequencies), suggesting that the pre-ictal state in these cases might correspond to a transition into a deeper sleep stage. In fact, these pre-ictal changes may result from the conjunction of two phenomena: physiological changes in brain dynamics, responsible for a facilitating state, and secondary pathological changes of neuronal networks, underlying the emergence of interictal (Ferrillo et al., 2000) or ictal epileptiform abnormalities specifically at these moments in frontal lobe epilepsy (Crespel et al., 2000). Pathological changes in dynamics were detected by non-linear analysis in a variable number of channels, whereas physiological changes may concern all the channels. This conjunction may also explain the difference (not statistically significant) between anticipation times determined in awake and sleeping patients.
Even if pre-ictal changes in dynamics may sometimes be associated with modifications in classical clinical or electrophysiological parameters and thus could provide interesting correlations, in most seizures no changes were apparent either in the patient’s behaviour or in classical EEG analysis, whereas the non-linear EEG analysis succeeded in detecting pre-ictal changes.
Our comparison between pre-ictal changes in dynamics and ictal electroclinical data showed that anticipation times were significantly correlated with the initial frequency of the seizures, but not with their duration or with seizure type, suggesting that seizure onset may obey the same dynamical laws as those governing the pre-ictal state, whereas other laws govern the unfolding and the end of the seizure.
Methodological improvements
Previous non-linear measures used in the analysis of brain activity—the neuronal complexity loss L* (Lehnertz and Elger, 1998) and the largest Lyapunov exponent (Iasemidis and Sackellares, 1996)—provided quantitative measures of brain dynamics. The use of these indices to describe state changes in EEG dynamics implies a confrontation with brain dynamics that may be non-stationary for long periods, so that successive absolute indices are not easily comparable. Our novel method of non-linear analysis significantly improves the strategy for detecting pre-ictal changes in several ways (Le Van Quyen et al., 1999). First, we calculate a ‘relative index’, which is more adapted to the non-stationarity of long-term electrophysiological recordings (Manuca and Savit, 1996), by measuring similarities between different dynamics that have been reconstructed by non-linear methods. Secondly, at the signal level we now consider only the phase data that contain the most relevant information about the dynamics (Sauer, 1994). This choice reduces both the time required for calculation and the sensitivity to spikes or artefacts, in contrast to the other non-linear measures, which are highly sensitive to variation in signal amplitude.
The similarity index was developed primarily to detect pre-ictal states. Before clinical application, our algorithms must be adapted for real-time operation and the sensitivity and the specificity of this method must be evaluated in further studies. The factors that may influence the specificity of the method comprise the parameters of the algorithm and the statistical thresholds for time anticipation, but also the choice of the reference period. For interictal periods far from a seizure, the similarity index did not vary markedly while the state of vigilance of the patient was stable, in contrast with the deep and sustained deviations detected before most seizures. The rate of false detection (0.31 ± 0.20 per hour) is close to that (1 false positive per hour) of the method for automatic seizure detection (and not anticipation) developed by Gotman (1990), which is currently used in most long-term video-EEG monitoring units. As previously pointed out by others (Katz et al., 1991), changes in the non-linear measure of dynamics, associated with changes in the patient’s state of vigilance, currently limit the use of the method for automated long-term EEG monitoring. For daily analysis of brain dynamics, we plan to reset the reference period according to the state of vigilance of the patient.
Alternative signal analysis strategies may provide us with more detailed information on the neuronal mechanisms that underlie the pre-ictal state. One promising strategy consists of measuring phase synchrony between the activities of different EEG channels, and preliminary results suggest that changes in synchronization in specific frequency ranges are associated with the pre-ictal state.
Conclusion
We have shown, in patients with medically refractory epilepsy originating in the neocortex, that non-linear changes in the electrical dynamics of the brain precede most seizures by a mean of 7.5 min. Moreover, our non-linear similarity method anticipates seizures as efficiently from the scalp as from intracranial EEG recordings. These findings, together with recent results obtained from a population with temporal lobe epilepsy, suggest that pre-ictal changes in dynamics may be a general mechanism associated with seizure emergence and may be detectable in a wide population of patients with partial epilepsies originating from various sites of the brain. Furthermore, identification of transitions towards pathological states argues that epilepsy belongs to the group of ‘dynamical’ diseases (Mackey and Glass, 1977). Dysregulation of physiological controls may underlie the transition (the pre-ictal state) of the brain system from a highly complex, healthy dynamics (the interictal state) towards a more periodic, pathological dynamics (the seizure). Finally, the ability to detect pre-ictal changes gives to the word ‘epilepsy’ a sense opposite to its Greek root (attacked by surprise).
The possibility of anticipating epileptic seizures offers multiple clinical and therapeutic prospects. Detecting pre-ictal states would considerably facilitate the early ictal injection of single-photon CT tracer among patients in presurgical evaluation. More generally, ambulatory systems might be developed to give a warning of impending seizure to patients who are not candidates for surgery or for whom surgery has failed. Such systems could greatly decrease the risk of injury and improve quality of life by reducing patients’ helplessness in the face of a disease that strikes with unpredictable timing. The time window opened before a seizure should also allow therapeutic intervention to counter the emergence of the seizure.
|
In Memoriam |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionIn MemoriamReferences |
|---|
On May 28, 2001, Francisco J. Varela died. He was the head of the Neurodynamics Group at the laboratory of Cognitive Neurosciences and Brain Imaging (CNRS UPR 640) at the Pitié-Salpêtrière Hospital in Paris. During the later years of his career, he gave considerable impulsion to the group working in the field of seizure anticipation by sharing his exceptional skills and considerable knowledge in neurobiology, neural dynamics, cognitive neuroscience and philosophy.
|
Acknowledgements |
|---|
We wish to thank Richard Miles for critical reading of the manuscript and David Rudrauf for his help in statistical analysis. This work was supported by the Institut National de la Santé et de la Recherche Médicale (INSERM) (V.N.), the Institut Electricité Santé de France (EDF) and Sanofi-Synthelabo.
|
References |
|---|
|
TopSummaryIntroductionMethodsResultsDiscussionIn MemoriamReferences |
|---|
Albano AM, Muench J, Schwartz C, Mees AI, Rapp PE. Singular-value decomposition and the Grassberger–Procaccia algorithm. Phys Rev A 1988; 38: 3017–27.[Medline]
Baumgartner C, Serles W, Leutmezer F, Pataraia E, Aull S, Czech T, et al. Pre-ictal SPECT in temporal lobe epilepsy: regional cerebral blood flow is increased prior to electroencephalography-seizure onset. J Nucl Med 1998; 39: 978–82.
Buck D, Baker GA, Jacoby A, Smith DF, Chadwick DW. Patients’ experiences of injury as a result of epilepsy. Epilepsia 1997; 38: 439–44.[ISI][Medline]
Cockerell OC, Johnson AL, Sander JW, Hart YM, Goodridge DM, Shorvon SD. Mortality from epilepsy: results from a prospective population-based study. Lancet 1994; 344: 918–21.[ISI][Medline]
Crespel A, Coubes P, Baldy-Moulinier M. Sleep influence on seizures and epilepsy effects on sleep in partial frontal and temporal lobe epilepsies. [Review]. Clin Neurophysiol 2000; 111 Suppl 2: S54–9.
Devinsky O, Vickrey BG, Cramer J, Perrine K, Hermann B, Meador K, et al. Development of the quality of life in epilepsy inventory. Epilepsia 1995; 36: 1089–104.[ISI][Medline]
Engel J Jr. Outcome with respect to epileptic seizures. In: Engel J Jr, editor. Surgical treatment of the epilepsies. New York: Raven Press; 1987. p. 553–71.
Engel J. Surgery for seizures. [Review]. New Engl J Med 1996, 334: 647–52.
Ferrillo F, Beelke M, Nobili L. Sleep EEG synchronization mechanisms and activation of interictal epileptic spikes. [Review]. Clin Neurophysiol 2000; 111 Suppl 2: S65–73.
Goldberger AL. Non-linear dynamics for clinicians: chaos theory, fractals, and complexity at the bedside. Lancet 1996; 347: 1312–14.[ISI][Medline]
Gotman J. Automatic seizure detection: improvements and evaluation. Electroencephalogr Clin Neurophysiol 1990; 76: 317–324.[ISI][Medline]
Iasemidis LD, Sackellares JC. Chaos theory and epilepsy. Neuroscientist 1996; 2: 118–26.
Juhasz C, Chugani DC, Muzik O, Watson C, Shah J, Shah A, et al. Electroclinical correlates of flumazenil and fluorodeoxyglucose PET abnormalities in lesional epilepsy. Neurology 2000; 55: 825–35.
Katz A, Marks DA, McCarthy G, Spencer SS. Does interictal spiking change prior to seizures? Electroencephalogr Clin Neurophysiol 1991; 79: 153–6.[ISI][Medline]
Lee SA, Spencer DD, Spencer SS. Intracranial EEG seizure-onset patterns in neocortical epilepsy. Epilepsia 2000; 41: 291–307.
Lehnertz K, Elger CE. Can epileptic seizures be predicted? Evidence from nonlinear time series analysis of brain electrical activity. Phys Rev Lett 1998; 80: 5019–22.
Le Van Quyen M, Adam C, Lachaux J-P, Martinerie J, Baulac M, Renault B, et al. Temporal patterns in human epileptic activity are modulated by perceptual discriminations. Neuroreport 1997; 8: 1703–10.[ISI][Medline]
Le Van Quyen M, Martinerie J, Baulac M, Varela F. Anticipating epileptic seizures in real time by a non-linear analysis of similarity between EEG recordings. Neuroreport 1999; 10: 2149–55.[ISI][Medline]
Le Van Quyen M, Adam C, Martinerie J, Baulac M, Clémenceau S, Varela F. Spatio-temporal characterizations of non-linear changes in intracranial activities prior to human temporal lobe seizures. Eur J Neurosci 2000; 12: 2124–34.[ISI][Medline]
Le Van Quyen M, Martinerie J, Navarro V, Boon P, D’Havé M, Adam C, et al. Anticipation of epileptic seizures from standard EEG recordings. Lancet 2001; 357: 183–8.[ISI][Medline]
Lopes da Silva FH, Pijn JP. Epilepsy: network models of generation. In: Arbib MA, editor. The handbook of brain theory and neural networks. Cambridge (MA): MIT Press; 1995. p. 367–9.
Mackey MC, Glass L. Oscillation and chaos in physiological control system. Science 1977; 197: 287–9.
Manuca R, Savit R. Stationarity and nonstationarity in time series analysis. Physica D 1996; 99: 134–61.
Martinerie J, Adam C, Le Van Quyen M, Baulac M, Clémenceau S, Renault B, et al. Epileptic seizures can be anticipated by non-linear analysis. Nat Med 1998; 4: 1173–6.[ISI][Medline]
Matsuoka H, Takahashi T, Sasaki M, Matsumoto K, Yoshida S, Numachi Y, et al. Neuropsychological EEG activation in patients with epilepsy. Brain 2000; 123: 318–30.
Rajna P, Clemens B, Csibri E, Dobos E, Geregely A, Gottschal M, et al. Hungarian multicentre epidemiologic study of the warning and initial symptoms (prodrome, aura) of epileptic seizures. Seizure 1997; 6: 361–8.[ISI][Medline]
Rogowski Z, Gath I, Bental E. On the prediction of epileptic seizures. Biol Cybern 1981; 42: 9–15.[ISI][Medline]
Sauer T. Reconstruction of dynamical systems from interspike intervals. Phys Rev Lett 1994; 72: 3811–14.[ISI][Medline]
Schiff SJ. Forecasting brain storms [letter]. Nat Med 1998; 4: 1117–18.[ISI][Medline]
Schreiber T, Schmitz A. Classification of time series data with nonlinear similarity measures. Phys Rev Lett 1997; 79: 1475–8.
Semah F, Picot MC, Adam C, Broglin D, Arzimanoglou A, Bazin B, et al. Is the underlying cause of epilepsy a major prognostic factor for recurrence? Neurology 1998; 51: 1256–62.
Spencer SS. Substrates of localization-related epilepsies: biologic implications of localizing findings in humans. [Review]. Epilepsia 1998; 39: 114–23.[ISI][Medline]
Weinand ME, Carter LP, el-Saadany WF, Sioutos PJ, Labiner DM, Oommen KJ. Cerebral blood flow and temporal lobe epileptogenicity. J Neurosurg 1997; 86: 226–32.[ISI][Medline]
Widman G, Lehnertz K, Urbach H, Elger CE. Spatial distribution of neuronal complexity loss in neocortical lesional epilepsies. Epilepsia 2000; 41: 811–17.[ISI][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  S. Wong, A. B. Gardner, A. M. Krieger, and B. Litt A Stochastic Framework for Evaluating Seizure Prediction Algorithms Using Hidden Markov Models J Neurophysiol, March 1, 2007; 97(3): 2525 - 2532. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 F. Mormann, R. G. Andrzejak, C. E. Elger, and K. Lehnertz Seizure prediction: the long and winding road Brain, February 1, 2007; 130(2): 314 - 333. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. Aschenbrenner-Scheibe, T. Maiwald, M. Winterhalder, H. U. Voss, J. Timmer, and A. Schulze-Bonhage How well can epileptic seizures be predicted? An evaluation of a nonlinear method Brain, December 1, 2003; 126(12): 2616 - 2626. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||