For personal use only. Not to be reproduced without permission of The Lancet.

ARTICLES
Summary
Background New methods derived from non-linear analysis of
intracranial recordings permit the anticipation of an epileptic
seizure several minutes before the seizure. Nevertheless,
anticipation of seizures based on standard scalp-
electroencephalographical (EEG) signals has not been reported
yet. The accessibility to preictal changes from standard EEGs
is essential for expanding the clinical applicability of these
methods.
Methods We analysed 26 scalp-EEG/video recordings, from
60 min before a seizure, in 23 patients with temporal-lobe
epilepsy. For five patients, simultaneous scalp and intracranial
EEG recordings were assessed. Long-term changes before
seizure onset were identified by a measure of non-linear
similarity, which is very robust in spite of large artifacts and
runs in real-time.
Findings In 25 of 26 recordings, measurement of non-linear
changes in EEG signals allowed the anticipation of a seizure
several minutes before it occurred (mean 7 min). These
preictal changes in the scalp EEG correspond well with
concurrent changes in depth recordings.
Interpretation Scalp-EEG recordings retain sufficient dynamical
information which can be used for the analysis of preictal
changes leading to seizures. Seizure anticipation strategies in
real-time can now be envisaged for diverse clinical
applications, such as devices for patient warning, for efficacy
of ictal-single photon emission computed tomography
procedures, and eventual treatment interventions for
preventing seizures.
Lancet 2001; 357: 18388
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 disability
1
and
mortality.
2
Prediction of seizure onset, even in the short-
term, 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.
46
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
systems
7
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
statesinterictal and ictalbut also a preictal transitional
phase of several minutes that is not detected by linear
methods.
911
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 scalp-
electroencephalographical (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 delicate
13
and even questionable
14
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 non-
linear techniques.
1617
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.
Methods
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 hour were
studied. This selection was done with a minimum of a
priori knowledge about the quality of the signal. Recordings
were obtained from a 32 channel Biomedical Monitoring
Systems Inc system (Nicolet-BMSI, Madison, Wisconsin,
USA) with 21 or 27 electrodes, placed according to the
extended International 1020 System.
18
The scalp
potentials were passed to a channel amplifier system with
band-pass filter settings of 0599 Hz by an external
reference over linked ears, and were digitised at 200 Hz.
We obtained 18 long-term scalp recordings of 6090 min
duration including the 3060 min before the clinical
seizure, as defined by either the time of the first symptoms
(including aura sensation) or the first electrographical
Anticipation of epileptic seizures from standard EEG recordings
Michel Le Van Quyen, Jacques Martinerie, Vincent Navarro, Paul Boon, Michel DHav, Claude Adam, Bernard Renault,
Francisco Varela, Michel Baulac
Laboratoire de Neurosciences Cognitives et Imagerie Crbrale
(M Le Van Quyen PhD, J Martinerie PhD, V Navarro MD, C Adam PhD,
B Renault PhD, F Varela PhD, Prof M Baulac MD) and Unit
dEpileptologie (V Navarro, C Adam, Prof M Baulac), CNRS UPR
640, Hpital de la Piti-Salptrire, 75651 Paris, Cedex 13,
France; and Department of Neurology, Epilepsy Monitoring Unit and
Laboratory of Clinical Neurophysiology, University of Gent, 185 De
Pintelaan, B-9000 Gent, Belgium (Prof P Boon PhD, M DHav MS)
Correspondence to: Dr Francisco Varela
(e-mail: fv@ccr.jussieu.fr)
THE LANCET Vol 357 January 20, 2001 183
For personal use only. Not to be reproduced without permission of The Lancet.
changes (burst of spikes, electrodecremental events, or
sinusoidal waves) noted by the epileptologist
electroencephalographer. None of the recordings were
rejected.
Simultaneous scalp and intracranial recordings
We also studied seizures of five patients with TLE using
both scalp and intracranial electrodes. The patients were
implanted with mediotemporal intracerebral electrodes by
use of a posteroanterior trajectory. These electrodes
recorded the medial structures of the temporal lobe
(amygdala and hippocampus) and the temporo-occipital
cortical junction. In addition, subdural strips inserted
through burr holes sampled the lateral or inferior-temporal
cortex. Postimplantation location of the electrodes was
verified by magnetic resonance imaging. The EEG-video
recordings were made with a 64-channel Telefactor system
(Conshohocken, PA, USA). The raw data were filtered
with a band-pass of 0170 Hz and digitised at 200 Hz
using an external passive reference. We analysed 8 epochs
of 60 min duration 50 min before the seizure. The patients
were awake (six seizures) or asleep (two seizures) for the
entire preictal hour. Seizure onset was defined by
electrographical criteria as localised, sustained rhythmic
discharges greater than 2 Hz in frequency (burst of spikes,
low voltage, fast, &c) and associated with subsequent
clinical seizure activity.
3
Data analysis
The anticipation method, which has been described in
detail previously,
15
includes four main steps, which are done
independently on each EEG channel to obtain the spatial
distribution of dynamical changes (figure 1).
Reconstruction of EEG dynamics by time intervals
The recording was segmented into consecutive non-
overlapping windows of 30 s duration. For each window,
the signal was transformed to pure-phase information by
taking the sequence of time intervals between positive-
going crossings of a fixed threshold (set here to the zero of
the signal; figure 1A). An important advantage of this zero-
crossing approach is that the time intervals are robust to
noise components of the signal amplitude. Furthermore,
the approach achieves a substantial decrease (by ten orders
of magnitude) in the volume of data, providing the
computational speed for real-time procedures. Finally, the
sequence of time intervals contains the pertinent dynamical
information. Following standard techniques,
19
we converted
time-serial data to a geometric representation (figure 1B)
which describes the EEG dynamics in a high-dimensional
state space (see mathematical appendix on The Lancets
website www.thelancet.com)
Calculation of a similarity index between EEG time windows
The dynamical similarity between any pairs of EEG
windows is quantified by the crosscorrelation integral
between the two dynamics (see mathematical appendix). If
the EEG is stationary, the similarity index yields a value
close to 1. By contrast, if changes in the dynamical state
occur, the similarity index decreases below 1.
Detection of preictal changes by comparison with a
reference state
To give the most general picture of the preseizure changes,
we chose a long reference window (300 s) during a
stationary sequence and distant in time from the seizure.
This reference window contained common features of
interictal activity, including isolated spikes, and was chosen
at the same time for all channels. We computed the
similarity index between this reference window and
consecutive test windows moved over the whole recording
epoch. The timecourse of the similarities, the similarity
profile, provides information about long-term changes
before seizure onset. Figure 1C illustrates a representative
example of this step of our analysis applied to a single scalp
EEG channel (left frontotemporal channel FT9, same
seizure as figure 2) and covering 50 min before a left
medial-lobe seizure. Seizure onset was identified by
localised periodic low-frequency spikes in the left
hippocampus (window 97). The similarity profile, during
the first part of the recording (160 windows), shows a
segment of windows with high similarity, suggesting fairly
uniform dynamical properties. After this first dynamical
state, and before the seizure, there is a transitional preictal
state that begins around 18 min before seizure onset
(windows 6096) and differs dynamically from the
interictal and ictal states. We also did a spectral analysis
over the same period, to see if preictal changes could also
be detected by this linear method. As seen in figure 1C, this
was not the case, supporting the view that non-linear
analysis is essential.
Statistical analysis
We assessed the significance of the preictal changes by
quantifying the deviation of the test window from the
reference state. Let and be the mean and SD of
similarity variations () during the baseline. We define here
the significance of the deviation by the ratio
=()/ whose p-value is given by the Chebyshevs
inequality (for any statistical distribution of : P[||k]
1/k
2
where k is the chosen statistical threshold). The
temporal evolution of the statistical significance is used to
detect when the similarity index reaches a critical value. We
define the anticipation time as when the ratio reaches a
critical value of k=5 (p=004), and remains at or above this
fixed deviation threshold up to the time when the seizure
occurs.
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 profiles. No consistent changes in
preictal spiking nor other electroencephalographical
features correlated with these dynamical changes. After this
preictal state, the electrographic seizure was accompanied
by a second drop to the lowest similarity. Postictally, the
ARTICLES
184 THE LANCET Vol 357 January 20, 2001
Patient number Brain magnetic Epileptic focus
resonance imaging (EEG-video)
1 to 8 Left hippocampal sclerosis Left temporal
9 to 13 Right hippocampal sclerosis Right temporal
14 Right hippocampal sclerosis and Right temporal
right external temporo-occipital scar
15 Right internal temporo-occipital tumour Right temporal
16 Normal Right hemisphere
(right medial temporal*)
17 Normal Left temporal
18 Right external temporo-occipital scar Right temporo-occipital or
right anterior temporal
19 to 21 Left hippocampal sclerosis Left hippocampus*
22 Bilateral hippocampal sclerosis Right hippocampus*
23 Right hippocampal sclerosis Right hippocampus*
*Intracranial recording.
Electroradiological features of the patients
For personal use only. Not to be reproduced without permission of The Lancet.
ARTICLES
THE LANCET Vol 357 January 20, 2001 185
Figure 1: Quantification of preictal changes
A: The first step is to transform the recording into sequences of time intervals I
n
between successive zero-crossings for pairs of windows that are not
necessarily adjacent in time.
B: The second step is to define a multidimensional reconstruction of the dynamics by a time delay embedding of the intervals, shown here with m=3 for a
schematic representation. The third step consists of the comparison of the two reconstructed dynamics by a measure of similarity.
C: Analysis of 50 min of a scalp EEG channel before a spontaneous temporal lobe seizure; left, examples of consecutive windows of 30 s duration in the
recording, the seizure onset occurs at window 97; right, the similarity profile (upper panel) from a long reference phase chosen here from windows 1 to
10. A sustained preictal state is identified 18 min before seizure onset. The spectrogram (lower panel) display over the same epoch shows no relation to
the similarity profile.
For personal use only. Not to be reproduced without permission of The Lancet.
ARTICLES
186 THE LANCET Vol 357 January 20, 2001
Figure 2: Representative example of our analysis
A: left, for several scalp and intracranial channels in the left temporal region, the timecourse of the similarities from a reference state taken at the
beginning of the recording; right, schematic location of the scalp and intracranial electrodes.
B: Spatiotemporal deviation map of preseizure changes. The deviations from the reference state of each contact are depicted as a function of time in SDs
by a colour scale. The corresponding anticipation times are indicated on the right of each contact. Long lasting decreases around 18 min before seizure
were found at a large number of the recording sites.
For personal use only. Not to be reproduced without permission of The Lancet.
similarity index increased again and tended to reach the
interictal level. There was a large correlation between the
intercranial and the scalp similarity profiles (=788% at lag
zero between T9 and LH1).
To obtain a complete spatiotemporal picture of preictal
dynamics, this analysis was applied to all recorded channels
(figure 2B). Significant decreases in similarity were
uniformly recorded several minutes before the seizure.
These changes were present at almost all recording sites in
both scalp (25 of 27) and intracranial recordings (29 of 32).
For many channels, state transitions before the seizure
seem to wax and wane and are occasionally disrupted, in
particular, shortly before seizure onset. Nevertheless, some
channels (scalp: F7; intracranial: left hippocampus 5, right
hippocampus 1) show sustained deviations above the
statistical threshold up to the onset of the electrographic
seizure. In the example shown, the preictal changes are
more pronounced and sustained in the left-temporal lobe
(scalp: FT9 to F7, intracranial: left hippocampus) but can
also be seen in contralateral contacts (scalp: T8 to TP10,
intracranial: right hippocampus). The spatial distribution of
preseizure changes is widespread and projects beyond the
epileptogenic focus.
Figure 3 shows the seizure anticipation times we
recorded from the 23 patients with TLE. Definition of the
dynamical properties of the scalp EEG allowed, in 25 of 26
cases, anticipation of a seizure several minutes before it
occurred (mean 416 s [SD 356]). When multiple seizures
were recorded in single patients, the corresponding
anticipation times were highly variable. This finding is
consistent with our previous results
9,11
and suggests that the
preictal state is a complex process which varies within
individual patients. For the group of patients with both
scalp and intracranial recordings, there was a significant
correlation between the timecourse of scalp and intracranial
preictal changes (figure 3). In most cases, anticipation times
from scalp and intracranial recordings were similar.
Preictal state changes were in the temporal lobe
ipsilateral to the epileptic focus in most cases. The
contralateral temporal lobe was often the site of changes,
although these changes were less precocious and
pronounced. In some patients the changes appeared in
parasagittal and frontal areas and were not spatially
congruent with the epileptic area. The detected changes
frequently showed a widespread topography that projected
beyond the limits of the epileptogenic temporal lobe.
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 changes in the
physiological state of the brain.
5
Our finding that the changes in scalp electrical activity
were similar to those detected from intracranial recordings
is surprising, given that the scalp EEG is an attenuated and
blurred vision of direct intracranial recording because of the
distance between the brain and scalp, with the skull as an
interposing medium. However, the relations between
activity recorded with intracranial and extracranial
electrodes is more complex than a simple decrease in the
signal-to-noise ratio because of cortical convolutions,
anatomical anisotropies, and the orientation, shape, and
extension of the underlying generators.
20
Also, potentials
produced by neocortical changes and recorded from scalp
electrodes might also be driven by events in deeper cerebral
structures. For instance, changes in the hippocampal
activity could cause secondary activation of several
neocortical areas, producing synchronised local field
potentials. Further studies are required to identify the
extent of generators giving rise to the global dynamics
ultimately detected on the scalp, particularly the use of
more extensive intracranial sampling or source localisation
methods.
Keeping in mind these uncertainties, we have
nevertheless shown that scalp recordings retain sufficient
information for use as a preictal marker. A remarkable
property of the similarity estimation is the robustness
against noise compared with other non-linear methods.
9,10
Furthermore, our anticipation times determined from scalp
electrodes, even for data with poor signal-to-noise ratios,
indicate an extensive spatial distribution of the preictal
condition, necessarily involving multiple sources from both
the epileptogenic region and distant non-epileptogenic
regions. This finding is confirmed by our analysis of
simultaneous intracranial recordings and consistent with
our previous results.
11
These findings are further
concordant with the notion that TLE is not a perfect model
of a purely focal process, but often exhibits widespread
functional, cognitive, electrical, and structural
abnormalities, with frequent contralateral temporal-lobe
involvement.
21
It could be argued that our findings cannot be explained
by ictal discharges present at depth but not recognised at
the scalp, which is unlikely for two reasons: the time
difference between intracranial seizure onset and the first
clinical or scalp-EEG signs should have been of several
minutes; simultaneous surface and intracranial recordings
ARTICLES
THE LANCET Vol 357 January 20, 2001 187
P
a
t
i
e
n
t

n
u
m
b
e
r
23
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
86%
75%
94%
56%
80%
66%
81%
73%
20
21
22
23
20 16 13
Anticipation time (min)
10 6 3
Correlation
Surface
Intracranial
Seizure
Figure 3: Anticipation times for all the patients and seizures
under investigation
Defined as the earliest anticipation recorded across all the channels.
Correlation coefficients are expressed as percentages.
For personal use only. Not to be reproduced without permission of The Lancet.
show that the timecourses of preictal change occur in
parallelscalp preictal changes takes place before the
intracranial-seizure onset.
The presence of widespread preictal changes raises the
question of their relation to the process of seizure
generation. As previously suggested,
9,10
one hypothesis is
that preictal changes are due to subthreshold epileptic
recruitment. Large scale drivings starting from the
epileptogenic region can be involved in this process, as
recorded during interictal activity.
22
A second hypothesis is
that the preictal state might not be part of the ictal process
itself, but reflects a facilitating state of the brain, which
allows a susceptible region to generate sustained epileptic
discharges. From this perspective, two interrelated
processes with different timescales can be distinguished:
long-lasting and non-specific generalised changes that
could create favourable conditions for the appearance of
seizures; and focal epileptogenic processes in regions prone
to hypersynchronous behaviour in response to these
generalised changes. A possible candidate for this long-term
facilitation process could be a decrease in remote inhibitory
control of the epileptogenic zone.
23
To further understand the pathophysiological basis of
preictal changes, it would be of interest to know if other
physiological indices also show preictal modification.
Invasive techniques to measure regional cerebral blood flow
in a continuous fashion
24
or fortuitous non-invasive single
photon emission computed tomography observations
25
have
shown preictal modifications around 10 min before
temporal-lobe seizures. Interestingly, large alterations in
regional cerebral blood flow have been reported in both
epileptogenic and contralateral non-epileptogenic
neocortices, consistent with a widespread perturbation of
brain perfusion before medial-temporal-lobe seizure onset.
From a clinical point of view, the ability to anticipate
seizure with the scalp EEG may have considerable practical
implications for the large population of patients with
uncontrolled epilepsy. The computational cost of our
method is quite small and it can be implemented in real-
time with a personal computer, facilitating ambulatory
application. Our method holds the potential for clinical
application, warning patients at home or in a hospital
setting. A careful study of the specificity and sensitivity of
this method over long recording periods and larger patient
populations needs to be done. If proven reliable, such an
application would lower the medical consequences of
seizures and improve the quality of life of people with
epilepsy by decreasing the risk of injury, and the sense of
helplessness fostered by the unpredictability of the disease.
The scalp EEG could also facilitate EEG/video monitoring
of surgery candidates by allowing early injection of the peri-
ictal single photon computed tomography emission tracer,
which would lower the cost of the procedure. In addition,
identifying preictal changes would also allow treatment
intervention such as administration of a short-acting
anticonvulsant drug,
26
electrical stimulation,
27
or cognitive
intervention with neurophysiological or behavioural
countermeasures.
Contributors
Michel Le Van Quyen, Jacques Martinerie, and Francisco Varela initiated
the study and did the signal analysis. Vincent Navarro collected the data
with the assistance of Claude Adam and provided substantial input into the
interpretation of the results. Bernard Renault provided assistance during
the early stages of the project. Paul Boon and Michel DHav provided the
data of simultaneous scalp and intracranial recordings. Michel Baulac
supervised the clinical part of the study. Michel Le Van Quyen and
Francisco Varela wrote the paper, with input from Vincent Navarro and
Michel Baulac.
Acknowledgments
This work was supported by the Fondation Franaise pour la Recherche sur
lEpilepsie, Fondation pour la Recherche Mdicale, the Institut National de
la Sant et de la Recherche Mdicale (INSERM), Institut Electricit Sant
de France, Laboratoire Sanofi-Synthelabo.
We are grateful to Robert B Duckrow for his help in the preparation of the
manuscript.
References
1 Devinsky O, Vickrey BG, Cramer J, et al. Development of the quality of
life in epilepsy inventory. Epilepsia 1995; 36: 1089104.
2 Cockerell OC, Johnson AL, Sander JW, Hart YM, Goodridge DMG,
Shorvon SD. Mortality from epilepsy: results from a prospective
population-based study. Lancet 1994; 344: 91821.
3 Engel J Jr. Seizure and epilepsy. Philadelphia, USA: F A Davis
Company, 1989.
4 Rogowski Z, Gath I, Bental E. On the prediction of epileptic seizures.
Biol Cybern 1981; 42: 915.
5 Katz A, Marks DA, McCarthy G, Spencer SS. Does interictal spiking
rate change prior to seizure? Electroencephalogr Clin Neurophysiol 1991;
79: 15356.
6 Osorio I, Frei M-G, Wilkinson SB. Real-time automated detection and
quantitative analysis of seizures and short-term prediction of clinical
onset. Epilepsia 1998; 39: 61527.
7 Kantz H, Schreiber T. Nonlinear time series analysis. Cambridge, UK:
Cambridge University Press, 1997.
8 Goldberger AL. Nonlinear dynamics for clinicians: chaos theory, fractals
and complexity at the bedside. Lancet 1996; 347: 131214.
9 Martinerie J, Adam C, Le Van Quyen M, et al. Epileptic seizures can be
anticipated by non-linear analysis. Nat Med 1998; 4: 117376.
10 Lehnertz K, Elger CE. Can epileptic seizures be predicted? Evidence
from nonlinear time series analysis of brain electrical activity. Phys Rev
Lett 1998; 80: 501922.
11 Le Van Quyen M, Adam C, Martinerie J, Baulac M, Clmenceau S,
Varela F. Spatio-temporal characterizations of nonlinear changes in
intracranial activities prior to human temporal lobe seizures. Eur J
Neurosci 2000; 12: 122434.
12 Quesney L. Extracranial EEG evaluation. In: Engel J Jr, ed. Surgical
treatment of the epilepsies. New York: Raven Press, 1987: 12966.
13 Schreiber T, Kantz H. Observing and predicting chaotic signals: is 2%
noise too much? In Kravtsov YA, Kadtke JB, eds. Predictability of
complex dynamical systems. New York, USA: Springer, 1996.
14 Theiler J, Rapp PE. Re-examination of the evidence for low-dimensional
nonlinear structure in the human electroencephalogram.
Electroencephalogr Clin Neurophysiol 1996; 98: 21323.
15 Le Van Quyen M, Martinerie J, Baulac M, Varela F. Anticipating
epileptic seizure in real time by a nonlinear analysis of similarity between
EEG recordings. Neuroreport 1999; 10: 214955.
16 Manuca R, Savit R. Stationarity and nonstationarity in time series
analysis. Physica D 1996; 99: 13461.
17 Schreiber T, Schmitz A. Classification of time series data with nonlinear
similarity measures. Phys Rev Lett 1997; 79: 147578.
18 Nuwer MR, Comi G, Emerson R, et al. IFCN standards for digital
recording of clinical EEG. Electrocephalogr Clin Neurophysiol 1998; 106:
25961.
19 Taken F. Lecture notes in mathematics: dynamical systems and
turbulence. Springer Verlag: Berlin, 1981.
20 Alarcon G, Guy CN, Binnie CD, Walker SR, Elwes RDC, Polkey CE.
Intracerebral propagation of interictal activity in partial epilepsy:
implications for source localization. J Neurol Neurosurg Psychiatry 1994;
57: 23343.
21 Spencer SS. Substrates of localization-related epilepsies: biologic
implications of localizing findings in humans. Epilepsia 1998; 39:
11423.
22 Le Van Quyen M, Martinerie J, Adam C, Varela F. Nonlinear analyses
of interictal EEG map the brain interdependences in human focal
epilepsy. Physica D 1999; 127: 25067.
23 Depaulis A, Vergnes M, Marescaux C. Endogenous control of epilepsy.
Prog Neurobiol 1994; 42: 3352.
24 Weinand ME, Carter LP, El-Saadany WF, Sioutos PJ, Labiner DM,
Oommen KJ. Cerebral blood flow and temporal lobe epileptogenicity.
J Neurosurg 1997; 86: 22632.
25 Baumgartner C, Serles W, Leutmezer F, 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: 97882.
26 Akinbi MS, Welty TE. Benzodiazepines in the home treatment of acute
seizures. Ann Pharmacother 1999; 33: 99102.
27 Schiff S, Jerger K, Chang T, Sauer T, Spano M, Ditto W. Controlling
chaos in the brain. Nature 1994; 370: 61520.
ARTICLES
188 THE LANCET Vol 357 January 20, 2001