Content uploaded by Klaus Lehnertz
Author content
All content in this area was uploaded by Klaus Lehnertz on Mar 22, 2019
Content may be subject to copyright.
INVITED REVIEWS
State-of-the-Art of Seizure Prediction
Klaus Lehnertz,*†‡ Florian Mormann,* Hannes Osterhage,*† Andy Mu¨ller,*† Jens Prusseit,*†
Anton Chernihovskyi,*† Mattha¨us Staniek,*† Dieter Krug,*† Stephan Bialonski,*† Christian E. Elger*
Summary: Although there are numerous studies exploring basic
neuronal mechanisms that are likely to be associated with seizures,
to date no definite information is available as to how, when, or why
a seizure occurs in humans. The fact that seizures occur without
warning in the majority of cases is one of the most disabling aspects
of epilepsy. If it were possible to identify preictal precursors from
the EEG of epilepsy patients, therapeutic possibilities and quality of
life could improve dramatically. The last three decades have wit-
nessed a rapid increase in the development of new EEG analysis
techniques that appear to be capable of defining seizure precursors.
Since the 1970s, studies on seizure prediction have advanced from
preliminary descriptions of preictal phenomena and proof of prin-
ciple studies via controlled studies to studies on continuous multiday
recordings. At present, it is unclear whether prospective algorithms
can predict seizures. If prediction algorithms are to be used in
invasive seizure intervention techniques in humans, they must be
proven to perform considerably better than a random predictor. The
authors present an overview of the field of seizure prediction, its
history, accomplishments, recent controversies, and potential for
future development.
Key Words: Epilepsy, Preictal state, Seizure anticipation, Seizure
prediction, Seizure forecasting, Statistical validation, EEG.
(J Clin Neurophysiol 2007;24: 147–153)
Epilepsy is characterized by a recurrent and sudden mal-
function of the brain that is termed seizure. Epileptic
seizures reflect the clinical sign of an excessive and hyper-
synchronous activity of neurons in the cerebral cortex. De-
pending on the extent of involvement of other brain areas
during the course of the seizure, epilepsies can be divided
into two main classes. Generalized seizures involve almost
the entire brain, whereas focal (or partial) seizures originate
from a circumscribed region of the brain (epileptic focus)
and remain restricted to this region. Epileptic seizures may
be accompanied by an impairment or loss of conscious-
ness; psychic, autonomic, or sensory symptoms; or motor
phenomena.
Epilepsy affects more than 50 million individuals
worldwide—about 1% of the world’s population. Two thirds
of affected individuals have seizures that can be sufficiently
controlled by antiepileptic drugs. Another 7% to 8% may
profit from epilepsy surgery. In about 25% of individuals
with epilepsy, however, seizures cannot be controlled by any
available therapy (Annegers, 1996). Given these figures, the
motivation for research into the predictability of seizures is
straightforward. The fact that seizures occur without warning
in the majority of cases is one of the most disabling aspects
of epilepsy. If it were possible to predict seizures with high
sensitivity and specificity, even seconds before their onset,
therapeutic possibilities would change dramatically (cf.
Elger, 2001). One might envision a simple warning system
capable of decreasing both the risk of injury and the feeling
of helplessness that results from seemingly unpredictable
seizures. Side effects from treatment with antiepileptic drugs,
such as sedation and clouded thinking, could be reduced by
on-demand release of a short-acting drug or electrical stim-
ulation during the preictal state. Paired with other suitable
interventions (Morrell, 2006), such applications could reduce
morbidity and mortality as well as greatly improve the quality
of life for people with epilepsy. In addition, identifying a
preictal state could greatly contribute to our understanding of
the pathophysiologic mechanisms that generate seizures.
Considering epilepsies as dynamical diseases of brain
systems, Lopes da Silva and colleagues proposed two differ-
ent scenarios of how a seizure could evolve (Lopes da Silva
et al., 2003a, b). The first is that a seizure could be caused by
a sudden and abrupt transition, in which case it would not be
preceded by detectable dynamical changes in the EEG. Such
a scenario would be conceivable for the initiation of seizures
in primary generalized epilepsy. Alternatively, this transition
could be a gradual change (or a cascade of changes) in
dynamics, which could in theory be detected and could be
more likely in focal epilepsies.
Given the a posteriori knowledge of a seizure occur-
rence, it appears rather tempting to postulate the existence of
a transitional preictal state. However, its unequivocal a priori
definition, and with it the possibility to predict an impending
seizure, is obviously far from being trivial. That there is both
physiologic and clinical support for the idea that at least
certain types of seizures are predictable has been summarized
in a number of previous reviews (Iasemidis, 2003; Lehnertz
et al., 2001, 2003, 2007; Litt and Echauz, 2002; Litt and
*Department of Epileptology, †Helmholtz-Institute for Radiation and Nu-
clear Physics and ‡Interdisciplinary Center for Complex Systems, Uni-
versity of Bonn, Bonn, Germany.
Address correspondence and reprint requests to Dr. Klaus Lehnertz,
Department of Epileptology, Neurophysics Group, Bonn University
Medical Center, Sigmund Freud Str. 25, 53105 Bonn, Germany;
e-mail: Klaus.Lehnertz@ukb.uni-bonn.de.
Copyright © 2007 by the American Clinical Neurophysiology Society
ISSN: 0736-0258/07/2402-0147
Journal of Clinical Neurophysiology • Volume 24, Number 2, April 2007 147
Lehnertz, 2002) and shall not be recapitulated here. Instead,
we give an update and mention, among others, functional
MRI-based evidence of transitional states preceding seizures
(Federico et al., 2005) and interictal spikes (Ma¨kiranta et al.,
2005); preictal changes in other variables, such as R–R
interval on the ECG (Kerem and Geva, 2005); identifiable
changes associated with high-frequency oscillations that pre-
cede seizure-like activities in in vitro models of epilepsy
(Khosravani et al., 2005) and neocortical seizures in humans
(Worrell et al., 2004); and evidence for premonitory symp-
toms in 6.2% of the 500 patients that filled out a questionnaire
in a recent multicenter study (Schulze–Bonhage et al., 2006).
Interestingly, premonitory symptoms reported in this study oc-
curred at similar periods before seizures as anticipatory changes
that have been identified using EEG analysis methods.
SEIZURE PREDICTION: 1975 TO 2001
The search for the hidden information predictive of an
impending seizure has a long history in EEG analysis, dating
back to 1975. A detailed summary of the field can be found in
a number of previous reviews (see Iasemidis, 2003; Lehnertz et
al., 2000, 2003, 2007; Lehnertz, 2006; Litt and Echauz, 2002;
Litt and Lehnertz, 2002; Mormann et al., 2006a, 2007) and in
two special issues of the Journal of Clinical Neurophysiology
(2001) and of the IEEE Transactions on Biomedical Engi-
neering (2003). For technical details of the applied analysis
techniques, we refer to the cited references and to Elger and
Lehnertz, 2004; Lehnertz, 2006; Mormann et al., 2005, 2007;
and Stam, 2006.
Typically, research into prediction of rare and extreme
events arising in other scientific areas—such as physics,
mathematics, geophysics, meteorology, or economy (see Al-
beverio et al., 2006 for an overview)— has fertilized the field
of seizure prediction and vice versa. Thus, analysis tech-
niques that have been used to predict seizures over the past
three decades typically reflect the state-of-the-art available at
that time.
In the 1970s and 1980s, researchers mainly consid-
ered linear analysis techniques (see Lopes da Silva, 1987)
such as pattern recognition, analytic procedures of spectral
data, or autoregressive modeling of EEG data for predict-
ing epileptic seizures. Findings indicated that EEG changes
characteristic of preictal states could be detectable, at most, a
few seconds before the actual seizure onset. Also during that
time, the relevance of interictal spikes was investigated in a
number of clinical studies. Although some authors re-
ported a decrease or even total cessation of spikes before
seizures, reexamination in a larger sample did not confirm
this phenomenon.
Beginning in the 1990s, the aforementioned linear anal-
ysis techniques were accompanied by methods from the
theory of nonlinear dynamics (see Kantz and Schreiber, 2003,
for an overview) involving measures such as Lyapunov
exponents, dimensions, entropies, or correlation densities.
Use of these nonlinear measures is usually justified by their
ability to allow an improved characterization of the compli-
cated, apparently irregular behavior of the complex nonlinear
dynamical system brain. However, because these measures
are difficult to interpret in terms of their physiologic corre-
late, other techniques have focused on extracting neurophys-
iologic features from the EEG associated with epileptiform
activity in human and animal physiology, such as bursts of
complex epileptiform activity, slowing, chirps, and changes
in signal energy. Further methods focused on defining preic-
tal states include recurrent neural networks, simulated neu-
ronal cell models, and techniques developed in catastrophe
theory or in the field of self-organized criticality. A number
of studies during the 1990s showed characteristic changes
minutes to hours before seizure onset on the EEG and were
interpreted by their authors as defining preictal states of
various durations, some lasting hours.
Almost all of the aforementioned approaches were
based on univariate measures, i.e., related to only a single
recording site, and thus cannot reflect any interactions be-
tween different regions of the brain. The epileptogenic pro-
cess, however, is commonly accepted to be closely associated
with changes in neuronal synchronization in a network of
components that may be spatially distributed. The analysis of
synchronization in the EEG can therefore a priori be regarded
as a promising approach for the investigation of the spatio-
temporal dynamics of ictogenesis. Thus, over the last 5 or 6
years, some researchers have focused on bivariate or, more
generally, multivariate measures that are based on newly
developed physical-mathematical concepts for synchroniza-
tion in nonlinear dynamical systems (see Pikovsky et al.,
2001, for an overview). These analysis techniques permit
assessment of synchronous activity from multiple sites and
include nonlinear interdependence, measures for phase syn-
chronization and cross-correlation, the difference of the larg-
est Lyapunov exponents of two or more channels, nonlinear
causality, and classification approaches based on a fusion of
multiple EEG features from multiple sites.
At the end of the last millennium, there was great
enthusiasm for the ability of a variety of analysis methods
to define a preictal state. By that time work in the area had
also extended to scalp EEG, although the majority of
researchers confined their investigations to intracranial
EEG recordings.
Results obtained indicated that seizures are not random
events, but rather are related to ongoing dynamic processes
that may begin minutes to hours to days beforehand. The
different prediction horizons may reflect different aspects of
ictogenesis as captured by the different approaches, but also
indicated that none of these techniques appears to depict the
process fully. Seizure precursors appeared to wax and wane
in attempts to ignite a clinical event, but the forces both
driving and suppressing seizure generation remained hidden.
Patterns appeared to be patient specific, within a finite range
of pattern types, and it appeared that different approaches
may be required to predict seizures with clinically useful
accuracy in different individuals or in different epilepsy
syndromes. This may be a function of individual physiology
or potentially confounding variables such as electrode place-
ment and the amount and speed of medication taper during
inpatient video-EEG monitoring.
Lehnertz et al. Journal of Clinical Neurophysiology • Volume 24, Number 2, April 2007
Copyright © 2007 by the American Clinical Neurophysiology Society148
SEIZURE PREDICTION: 2002 TO 2006
Despite over two decades of excellent work in the field,
results from different research groups revealed considerable
contradiction. Convincing evidence demonstrating unequiv-
ocal seizure prediction in blinded, prospective, randomized
clinical trials, with appropriate statistical validation, remained
elusive. Central to the problem was the challenge of devel-
oping algorithms to detect unknown patterns associated with
seizure generation, a process that remains poorly understood.
Much of the EEG data analyzed in studies up to that time
were highly selected and restricted with regard to seizure
type, patient state, signal-to-noise ratio, duration of record-
ings, artifacts, etc. In addition, there were no standardized
methods or nomenclature for marking continuous EEG data,
no accepted methods for assessing algorithm performance,
and no agreement on acceptable test data. Even clear defini-
tions of exactly what constitutes seizure onset, seizure pre-
diction, anticipation, or forecasting and the definition of ictal
events either clinically or by EEG, were nebulous.
To address these issues, the International Seizure Pre-
diction Group (ISPG) was formed in 2000 to provide an
informal structure for the major groups working in this area
to share data and ideas. The ISPG was established with the
specific goal of moving the field of seizure prediction forward
from “proof of principle” experiments into validated, well-
understood methods that could be applied to both basic
science and clinical applications. The first international work-
shop of this group was held in Bonn, Germany in 2002 and
aimed at assessing the state of the field at that time by having
each major group apply its methods to predict seizures from
a shared set of continuous intracranial EEG data (Lehnertz
and Litt, 2005). Findings obtained from applying a large
number of analysis techniques are summarized in eight peer-
reviewed articles published together in the journal Clinical
Neurophysiology (D’Alessandro et al., 2005; Esteller et al.,
2005; Harrison et al., 2005a; Iasemidis et al., 2005; Jerger et
al., 2005; Jouny et al., 2005; Le Van Quyen et al., 2005;
Mormann et al., 2005; see also Ebersole, 2005). The results of
all these investigations were inconsistent and at times con-
tradictory despite substantial efforts to provide uniform data
in terms of disease type, conditions, and recordings. Four
studies had negative results, three studies had positive results
(predicting seizures for different time horizons), and one had
both, depending on which techniques were employed. Al-
though investigations with positive findings identified a state
of increased seizure likelihood lasting up to several hours,
seizures could not be predicted with exact timing. The group
agreed that, at present, none of the EEG analysis techniques
was sufficient for broad clinical application, and that there
were major practical problems to overcome. Nevertheless,
much was learned from the exercise, particularly with regard
to the need for standardization of analyses, data requirements,
performance criteria, and nomenclature. Some of the results
were encouraging, whereas other results illustrated that cer-
tain approaches are unlikely to be worthwhile. In line with
this inconclusiveness are recent controversies about the rel-
evance of nonlinear approaches for the prediction of epileptic
seizures (McSharry et al., 2003a,b; Maiwald et al., 2004;
Mormann et al., 2005) and studies raising doubts about the
reproducibility of previously reported claims (Aschenbrenner–
Scheibe et al., 2003; De Clerq et al., 2003; Harrison et al.,
2005a, 2005b; Lai et al., 2003, 2004; Lehnertz et al., 2003;
Maiwald et al., 2004; Winterhalder et al., 2003). Although
there is evidence from several methods for identifiable pre-
cursors preceding partial onset seizures, one should keep in
mind that this evidence is based on retrospective analyses of
mostly intracranial EEG data recorded during evaluation for
resective surgery. Up to now, no study has been published
that demonstrates unequivocal seizure prediction in blinded,
prospective, randomized clinical trials. Reasons for this be-
come apparent when considering the major methodological
steps involved in seizure prediction algorithms along with
problems posed by respective study design.
EEG Recording
Taking into account that epilepsy is a heterogeneous
disorder, what constitutes a good EEG dataset for testing the
performance of a seizure prediction algorithm? More impor-
tantly, what constitutes an optimum spatial and temporal
sampling of the ictogenic process? In many previous studies,
analyses were confined to one or several electrodes, relying
on a posteriori knowledge about location and extent of the
ictal onset zone. At best, data from sites distant from the ictal
onset zone were included in these studies only for comparison
with more “normal” regions. Others have developed optimi-
zation schemes that allow one to select certain electrodes out
of a large number of electrodes (Chaowalitwongse et al.,
2005; Jerger et al., 2005; Le Van Quyen et al., 2005), relying
on a posteriori knowledge about the dynamics of the icto-
genic process. Due to the availability of more powerful
computers, it is only recently that EEG data from all electrode
sites has entered seizure prediction studies. Interestingly,
some studies reported that the sites selected as best for
prediction were not in close vicinity to the epileptic focus but
could be located in remote or even contralateral brain struc-
tures (Mormann et al., 2003a, 2005; D’Alessandro et al.,
2005; Kalitzin et al., 2005; Le Van Quyen et al., 2005;
Mormann et al., 2003a; Navarro et al., 2005). This seemingly
counterintuitive finding may indicate the importance of brain
regions outside of the ictal onset zone but within the “epi-
leptic network” in generating clinical seizures (see also Wen-
dling et al., 2002, 2003). This is also in accordance with
findings showing that the synchronization of specific popu-
lations in relation to the epileptic focus may be of crucial
importance to determine whether a seizure is likely to occur
and to spread (Le Van Quyen et al., 2005; Mormann et al.,
2003a, 2003b, 2005). However, it may also indicate a rather
nonspecific phenomenon whose temporal proximity to sei-
zure onset was just by chance. Obviously, correct site selec-
tion would be a practical problem in instances where elec-
trode contacts are limited. Without a better understanding
about seizure generation in the epileptic network, there is no
way to know what might be an optimal spatial sampling for
seizure prediction studies.
Many previous studies have lacked reference to the
interictal state in terms of insufficient control data or baseline
epochs, and thereby focused merely on sensitivity without
Journal of Clinical Neurophysiology • Volume 24, Number 2, April 2007 State-of-the-Art of Seizure Prediction
Copyright © 2007 by the American Clinical Neurophysiology Society 149
considering specificity of the applied techniques. Maiwald
and colleagues reviewed 14 seizure prediction studies pub-
lished between 1998 and 2003 and concluded that in only half
of these studies was the performance of the applied seizure
prediction technique tested against interictal control data
(Maiwald et al., 2004). To thoroughly evaluate potential
predictors and potentially confounding variables, reference to
continuous, prolonged, multichannel EEG data must be re-
garded as indispensable. This requires large, high-quality,
meticulously collected, and annotated data archives that are
well characterized and represent the heterogeneity of patterns
and patients found in human epilepsy. The mass data-storage
capacity that is now available at many epilepsy centers
enables one to store the complete data acquired during pre-
surgical monitoring without the necessity of selecting sample
recordings. EEG data accompanying clinical information
must be as complete as possible to account for other factors
that may modulate seizure generation. Relating mathematical
approaches to clinical, video, and neurophysiologic data is a
massive undertaking and has begun only recently (Navarro et
al., 2005). There are now attempts to identify and mark the
broad range of physiologic and pathophysiologic changes that
occur in these measures interictally and at seizure onset in a
standardized and clear-cut manner. Given imperfect methods
for determining exact clinical and EEG onset times for
seizures, reference to EEG onset of seizures as opposed to
clinical seizure onset is preferred (Lehnertz and Litt, 2005).
EEG Analysis
At present, it is not clear which of the currently avail-
able characterizing measures, if any, is best suited for pro-
spective prediction, although with their different time hori-
zons and processing methods they may constitute different
ways of viewing the same process. There are indications of a
superior performance for approaches characterizing relations
between different brain regions (Le Van Quyen et al., 2005;
Mormann et al., 2005; Osterhage et al., 2007), and, as some
studies have demonstrated, it appears that some combina-
tion of measures will probably be required to carry out
reliable seizure prediction tailored to individual patients
(D’Alessandro et al., 2005; Lehnertz et al., 2001). One should
keep in mind that estimation of the performance of an
algorithm typically requires optimizing numerous computa-
tional parameters. This optimization typically requires a pos-
teriori knowledge, which leads to a significant risk of in-
sample overtraining for a single measure and even more
strongly for a combination of different measures. The appli-
cation of a huge variety of measures to the EEG might yield
seemingly good results just by chance (particularly on a
limited database) if appropriate methods for dealing with sta-
tistical issues of multiplicity are not implemented. The resulting
explosion in computational degrees of freedom underlines the
need for control tests and independent validation.
Performance Assessment
Many studies have neglected to evaluate the perfor-
mance of seizure prediction techniques, which may account
for the many contradictory findings. This is a major issue in
the field at present (Mormann et al., 2007), and has been
addressed in depth at the Second International Workshop of
the ISPG held in Bethesda, Maryland, in 2006. There are
currently two categories of statistical methods for assessing
the performance of seizure prediction techniques: analytical
approaches (Schelter et al., 2006b; Winterhalder et al., 2003)
and bootstrap approaches (Andrzejak et al., 2003; Kreuz et
al., 2004). Both categories go beyond the mere estimation of
sensitivity and specificity (e.g., using receiver-operating char-
acteristics or ROC). Because both categories make use of
sensitivity and specificity estimates, given a predefined pre-
diction horizon, they provide a framework to assess the
performance of a given seizure prediction technique and to
compare different techniques. The analytical approach pro-
posed by Winterhalder et al. (2003) makes use of the so-
called seizure prediction characteristic and for a given speci-
ficity compares the sensitivity of a method under investigation
with the performance of an unspecific, naı¨ve (random or
periodic) predictor. An extension of this approach has re-
cently been proposed by Schelter et al. (2006b) that takes into
account the application of prediction methods to multiple
time series (EEG channels) and seizures.
Using the method of surrogates, the bootstrap ap-
proaches test the null hypothesis of the nonexistence of a
preictal state, and the nominal size of the test is determined
by the number of surrogates. In contrast to the aforemen-
tioned analytical approaches, these methods do not assume
specific static or dynamical models of the preictal state
development. The concept of seizure time surrogates, as intro-
duced by Andrzejak et al. (2003), is based on the generation of
artificial seizure onset times by randomly shuffling the orig-
inal interseizure intervals. Using these surrogate seizure onset
times instead of the original onset times, the EEG data are
then subjected to the same analysis algorithms and prediction
statistics that were used for the original onset times, including
in particular any in-sample optimization carried out in the
process. The null hypothesis, namely that a given algorithm
cannot detect a preictal state with a performance above
chance level, can only be rejected if the performance of the
algorithm for the original seizure times is better than the
performance for a number of independent realizations of
the surrogate seizure times. The advantage of this type of
statistical validation is that it can be applied to any type
of analysis, algorithmic or statistical. Kreuz et al. (2004)
proposed a modification of this surrogate test, termed mea-
sure profile surrogates, that is based on a constrained ran-
domization of the time profile of the characterizing measure
while the seizure onset times are kept fixed. Both surrogate
methods have conceptual advantages and disadvantages, and
the concept of measure profile surrogates has a greater com-
putational burden. It is important to keep in mind, however,
that the fact that the null hypothesis cannot be rejected does
not prove its correctness. Rather, there may be alternative
explanations for this result.
Using these statistical methods for assessing the perfor-
mance of seizure prediction techniques, a number of recent
studies (Aschenbrenner–Scheibe et al. 2003; Maiwald et al.,
2004; Mormann et al., 2005; Schelter et al., 2006b; Winterhalder
et al., 2003) have shown that measures previously considered
Lehnertz et al. Journal of Clinical Neurophysiology • Volume 24, Number 2, April 2007
Copyright © 2007 by the American Clinical Neurophysiology Society150
suitable for prediction perform no better than a random
predictor. Evidence has accumulated, however, that certain
measures, particularly measures quantifying relations be-
tween recording sites, appear to perform significantly better
than random prediction, even if rigorous statistical validation
is applied. Whether their prospective performance however is
sufficient for a clinical application.
Summarizing this section, we conclude that the state-
of-the-art of the field does not allow a definite answer to the
question of whether seizures can be predicted using prospec-
tive algorithms, despite the availability of intracranial, con-
tinuous, multiday, multichannel EEG recordings, a multitude
of refined EEG analysis techniques, and the more rigorous
methodological design in recent studies using statistical
methods for performance assessment. Although a few studies
have used prediction algorithms in a quasi-prospective man-
ner, these studies did not include a statistical validation
(D’Alessandro et al., 2005; Iasemidis et al., 2003, 2005) or do
not allow conclusions on statistical validity because the exact
same analysis procedure was not applied to the seizure time
surrogates as to the original onset times (Chaovalitwongse et
al., 2005; see also Mormann et al., 2006b and Winterhalder et
al., 2006).
The Second International Workshop of the ISPG re-
flected the current ambiguity. Although some groups focused
on the more fundamental issues of seizure prediction such as
development of new methods or evaluation of potential pre-
dictors and potentially confounding variables, and on the role
of high-frequency oscillations in ictogenesis, other groups
already considered closed-loop intervention systems for sei-
zure control. However, to justify prospective clinical trials
involving invasive seizure intervention techniques such as
electrical brain stimulation in patients based on seizure pre-
diction, any of the prediction algorithms devised to date must
be proven to perform better than chance in a quasi-prospec-
tive setting on out-of-sample testing data.
FUTURE DEVELOPMENTS
To further improve the rather young research field of
seizure prediction, we consider the following developments
as necessary. There is great need for establishing convincing
evidence for the existence of a preictal state, and an appro-
priate model for its behavior in human epilepsy. Given the
heterogeneity of epilepsy, one model may not work in all
epilepsy syndromes. Because previous studies have mainly
concentrated on the predictability of focal seizures, it remains
to be established whether similar transitional preictal phe-
nomena can be observed in other types of epilepsies, both in
adults and children. These studies may help to single out
candidate epilepsies or, at least, candidate seizure types for
which prediction-based intervention techniques can be con-
sidered as an alternative therapy option. However, these
studies require large, high-quality, meticulously collected and
annotated data archives that are well characterized and rep-
resent the heterogeneity of patterns and patients found in
human epilepsy. This work is already underway as a collab-
orative effort through the ISPG.
More work on developing and understanding spontane-
ously seizing animal models of epilepsy for use in prediction
research will allow access to deep brain structures and other
locations in the epileptic network that cannot be explored in
human studies due to safety concerns. Characterizing and
verifying the clinical presentation, reliability, and reproduc-
ibility of these models will be vital to interpreting results
involving these animals. Developing more refined measure-
ment techniques would allow to sample at multiple temporal
and spatial scales, from single neurons to ensembles, local
and global networks, and to investigate the physiologic sub-
strate for larger-scale quantitative observations.
Future seizure prediction studies have to be designed
with great care, and with an acute awareness that as we delve
deeper into seizure generation we may be confronted with
other, yet unforeseen problems. Thus, understanding of the
interictal period and all of its confounding variables, of the
when and where of potential precursors, and of mechanisms
underlying ictogenesis has to be further improved. This may
also help to refine already existing characterizing measures
used in the seizure prediction algorithms to increase predic-
tive performance and to develop new analysis concepts and
measures.
Given the recent evidence (Mormann et al., 2005) that
particularly measures quantifying relations between record-
ing sites appear to perform significantly better than random
prediction, even if rigorous statistical validation is applied
(thereby outperforming both linear and nonlinear univariate
approaches), we consider research into the development of
multivariate techniques (Allefeld and Kurths, 2003; Bialonski
and Lehnertz, 2006; Chen et al., 2004; Mu¨ller et al., 2005;
Rudrauf et al., 2006; Schelter et al., 2006a, 2006c; Schiff et
al., 2005) that allow analysis of data from multiple sites
simultaneously as highly promising. New hardware and soft-
ware platforms (Mu¨ller et al., 2006) will likely provide more
computational power, such as through very-large-scale inte-
gration implementation of algorithms (Chernihovskyi and
Lehnertz, 2007; Krug et al., 2006; Kunz and Tetzlaff, 2003;
Sowa et al., 2005), or through reducing the complexity of
algorithms (Staniek and Lehnertz, 2007). One might also
consider totally different EEG analysis techniques such as
biologically inspired computing (see Chernihovskyi et al.,
2005 for an overview).
Further improvements may also be achieved through
mathematical modeling of the dynamics of neuronal networks
underlying the transition to seizures. Although the models
currently available are used to investigate epileptic phenom-
ena at different levels of complexity (from cellular to network
levels) within specific situations, they share the common
feature of providing insight into mechanisms involved in the
generation of epileptic activity either at a microscopic (sub-
cellular to cellular) or a macroscopic (multicellular to system)
level (see the special issue on computational approaches in
epilepsy of the Journal of Clinical Neurophysiology: Cherni-
hovskyi et al., 2005; Liley and Bojak, 2005; Suffcyinski et
al., 2005; Traub et al., 2005; Wendling, 2005). Research
along this line may help to integrate the plethora of experi-
mental data available, to improve analysis concepts and
Journal of Clinical Neurophysiology • Volume 24, Number 2, April 2007 State-of-the-Art of Seizure Prediction
Copyright © 2007 by the American Clinical Neurophysiology Society 151
measures, and to test various hypothesis concerning preictal
brain dynamics and its relation to endogenous and exogenous
control parameters.
All of these factors promise increasing progress in the
field of seizure prediction, now tempered by experience and
the knowledge that this is a complex, long-term problem.
ACKNOWLEDGMENT
This study was supported by Deutsche Forschungsge-
meinschaft (SFB TR3) and BONFOR, the intramural re-
search fund of the University of Bonn.
REFERENCES
Albeverio S, Jentsch V, Kantz H. Extreme events in nature and society.
Berlin: Springer, 2006.
Allefeld C, Kurths J. Multivariate phase synchronization analysis of EEG
data. IEICE T Fund Electr. 2003;E86A:2218–2221.
Andrzejak RG, Mormann F, Kreuz T, et al. Testing the null hypothesis of the
non-existence of the pre-seizure state. Phys Rev E. 2003;67:010901.
Annegers JF. The epidemiology in epilepsy. In: Wyllie E (ed.) The treatment
of epilepsy: principle and practice. Baltimore: Williams and Wilkins,
1996:165–172.
Aschenbrenner-Scheibe R, Maiwald T, Winterhalder M, et al. How well can
epileptic seizures be predicted? An evaluation of a nonlinear method.
Brain. 2003;126:2616–2626.
Bialonski S, Lehnertz K. Identifying phase synchronization clusters in
spatially extended dynamical systems. Phys Rev E. 2006;74:051909.
Chaovalitwongse W, Iasemidis LD, Pardalos PM, et al. Performance of a
seizure warning algorithm based on the dynamics of intracranial EEG.
Epilepsy Res. 2005;64:93–113.
Chen YH, Rangarajan G, Feng JF, Ding MZ. Analyzing multiple nonlinear
time series with extended Granger causality. Phys Lett A. 2004;324:
26–35.
Chernihovskyi A, Mormann F, Mu¨ ller M, et al. EEG analysis with nonlinear
excitable media. J Clin Neurophysiol. 2005;22:314–329.
Chernihovskyi A, Lehnertz K. Measuring synchronization with nonlinear
excitable media. Int J Bifurcation Chaos. 2007, in press.
D’Alessandro M, Vachtsevanos G, Esteller R, et al. A multi-feature and
multi-channel univariate selection process for seizure prediction. Clin
Neurophysiol. 2005;116:506–516.
De Clercq W, Lemmerling P, Van Huffel S, Van Paesschen W. Anticipation
of epileptic seizures from standard EEG recordings. Lancet. 2003;361:
971.
Ebersole JS. In search of seizure prediction: a critique. Clin Neurophysiol.
2005;116:489–492.
Elger CE. Future trends in epileptology. Curr Opin Neurol. 2001;14:185–
186.
Elger CE, Lehnertz K. Prediction of seizure occurrence by chaos analysis:
Technique and therapeutic implications. In: Rosenow F, Lu¨ ders HO
(eds.) Presurgical assessment of the epilepsies with clinical neurophys-
iology and functional imaging. Handbook of Clinical Neurophysiology,
vol 3. Amsterdam: Elsevier, 2004:491–500.
Esteller R, Echauz J, D’Alessandro M, et al. Continuous energy variation
during the seizure cycle: towards an on-line accumulated energy. Clin
Neurophysiol. 2005;116:517–526.
Federico P, Abbott DF, Briellmann RS, et al. Functional MRI of the pre-ictal
state. Brain. 2005;128:1811–1817.
Harrison MA, Frei MG, Osorio I. Accumulated energy revisited. Clin
Neurophysiol. 2005a;116:527–531.
Harrison MA, Osorio I, Frei MG, et al. Correlation dimension and integral do
not predict epileptic seizures. Chaos. 2005b;15:033106.
Iasemidis LD. Epileptic seizure prediction and control. IEEE Trans Biomed
Eng. 2003;50:549–558.
Iasemidis LD, Shiau DS, Chaovalitwongse W, et al. Adaptive epileptic
seizure prediction system. IEEE Trans Biomed Eng. 2003;50:616–627.
Iasemidis LD, Shiau DS, Pardalos PM, et al. Long-term prospective on-line
real-time seizure prediction. Clin Neurophysiol. 2005;116:532–544.
Jerger KK, Weinstein SL, Sauer T, Schiff SJ. Multivariate linear discrimi-
nation of seizures. Clin Neurophysiol. 2005;116:545–551.
Jouny CC, Franaszczuk PJ, Bergey GK. Signal complexity and synchrony of
epileptic seizures: is there an identifiable preictal period? Clin Neuro-
physiol. 2005;116:552–558.
Kalitzin S, Velis D, Suffczynski P, et al. Electrical brain-stimulation para-
digm for estimating the seizure onset site and the time to ictal transition
in temporal lobe epilepsy. Clin Neurophysiol. 2005;116:718–728.
Kantz H, Schreiber T. Nonlinear time series analysis, 2nd ed. Cambridge,
UK: Cambridge University Press, 2003.
Kerem DH, Geva AB. Forecasting epilepsy from the heart rate signal. Med
Biol Eng Comput. 2005;43:230–239.
Khosravani H, Pinnegar CR, Mitchell JR, et al. Increased high-frequency
oscillations precede in vitro low-Mg2⫹seizures. Epilepsia. 2005;46:
1188–1197.
Kreuz T, Andrzejak RG, Mormann F, et al. Measure profile surrogates: A
method to validate the performance of epileptic seizure prediction
algorithms. Phys Rev E. 2004;69:061915.
Krug D, Osterhage H, Elger CE, Lehnertz K. Estimating generalized syn-
chronization in brain electrical activity from epilepsy patients with
Cellular Nonlinear Networks. In: Tavsanoglu V, Arik S (eds.) Proceed-
ings of the 2006 10th IEEE International Workshop on Cellular Neural
Networks and Applications. Piscataway, NJ: IEEE-Press, 68–72.
Kunz R, Tetzlaff R. Spatio-temporal dynamics of brain electrical activity in
epilepsy: Analysis with Cellular Neural Networks (CNNs). J Circ Syst
Comp. 2003;12:825–844.
Lai YC, Harrison MA, Frei MG, Osorio I. Inability of Lyapunov exponents
to predict epileptic seizures. Phys Rev Lett. 2003;91:068102.
Lai YC, Harrison MA, Frei MG, Osorio I. Controlled test for predictive
power of Lyapunov exponents: their inability to predict epileptic sei-
zures. Chaos. 2004;14:630–642.
Le Van Quyen M, Soss J, Navarro V, et al. Preictal state identification by
synchronization changes in long-term intracranial EEG recordings. Clin
Neurophysiol. 2005;116:559–568.
Lehnertz K, Arnhold J, Grassberger P, Elger CE. Chaos in brain? Singapore:
World Scientific, 2000.
Lehnertz K, Andrzejak RG, Arnhold J, et al. Nonlinear EEG analysis in
epilepsy: Its possible use for interictal focus localization, seizure antic-
ipation, and prevention. J Clin Neurophysiol. 2001;18:209–222.
Lehnertz K, Mormann F, Kreuz T, et al. Seizure prediction by nonlinear EEG
analysis. IEEE EMB Mag. 2003;22:57–63.
Lehnertz K, Litt B. The First International Collaborative Workshop on
Seizure Prediction: summary and data description. Clin Neurophysiol.
2005;116:493–505.
Lehnertz K. Epilepsy: extreme events in the human brain. In: Albeverio S,
Jentsch V, Kantz H (eds). Extreme events in nature and society. Berlin:
Springer, 2006:123–143.
Lehnertz K, Le Van Quyen M, Litt B. Seizure prediction. In: Engel J Jr,
Pedley TA (eds.) Epilepsy: a comprehensive textbook, 2nd ed. Phila-
delphia: Lippincott Williams and Wilkins, 2007, in press.
Liley DT, Bojak I. Understanding the transition to seizure by modeling the
epileptiform activity of general anesthetic agents. J Clin Neurophysiol.
2005;22:300–313.
Litt B, Lehnertz K. Seizure prediction and the preseizure period. Curr Opin
Neurol. 2002;15:173–177.
Litt B, Echauz J. Prediction of epileptic seizures. Lancet Neurol. 2002;1:22–30.
Lopes da Silva FH. EEG analysis: theory and practice. In: Niedermayer E,
Lopes da Silva FH (eds). Electroencephalography, Basic Principles,
Clinical Applications and Related Fields. Baltimore: Urban and
Schwarzenberg, 1987:871–897.
Lopes da Silva FH, Blanes W, Kalitzin SN, et al. Epilepsies as dynamical
diseases of brain systems: basic models of the transition between normal
and epileptic activity. Epilepsia. 2003a;44(suppl 12):72–83.
Lopes da Silva FH, Blanes W, Kalitzin SN, et al. Dynamical diseases of brain
systems: different routes to epileptic seizures. IEEE Trans Biomed Eng.
2003b;50:540–548.
Maiwald T, Winterhalder M, Aschenbrenner-Scheibe R, et al. Comparison of
three nonlinear seizure prediction methods by means of the seizure
prediction characteristic. Physica D. 2004;194:357–368.
Ma¨kiranta M, Ruohonen J, Suominen K, et al. BOLD signal increase
preceeds EEG spike activity – a dynamic penicillin induced focal
epilepsy in deep anesthesia. Neuroimage. 2005;27:715–724.
Lehnertz et al. Journal of Clinical Neurophysiology • Volume 24, Number 2, April 2007
Copyright © 2007 by the American Clinical Neurophysiology Society152
McSharry PE, He T, Smith LA, Tarassenko L. Linear and non-linear
methods for automatic seizure detection in scalp electro-encephalogram
recordings. Med Biol Eng Comput. 2002;40:447–461.
McSharry PE, Smith LA, Tarassenko L. Prediction of epileptic seizures: are
nonlinear methods relevant? Nature Med. 2003;9:241–242.
Mormann F, Andrzejak RG, Kreuz T, et al. Automated detection of a
pre-seizure state based on a decrease in synchronization in intracranial
EEG recordings from epilepsy patients. Phys Rev E. 2003a;67:021912.
Mormann F, Andrzejak RG, Elger CE, Lehnertz K. Seizure prediction: the
long and winding road. Brain. 2007;130:314–333.
Mormann F, Kreuz T, Andrzejak RG, et al. Epileptic seizures are preceded
by a decrease in synchronization. Epilepsy Res. 2003b;53:173–185.
Mormann F, Kreuz T, Rieke C, et al. On the predictability of epileptic
seizures. Clin Neurophysiol. 2005;166:569–587.
Mormann F, Elger CE, Lehnertz K. Seizure anticipation: from algorithms to
clinical practice. Curr Opin Neurol. 2006a;19:187–193.
Mormann F, Elger CE, Lehnertz K. Performance of a seizure warning
algorithm based on the dynamics of intracranial EEG. Epilepsy Res.
2006b;71:241–242.
Morrell M. Brain stimulation for epilepsy: can scheduled or responsive
neurostimulation stop seizures? Curr Opin Neurol. 2006;19:164–168.
Mu¨ ller M, Baier G, Galka A, et al. Detection and characterization of changes
of the correlation structure in multivariate time series. Phys Rev E.
2005;71:046116.
Mu¨ ller A, Osterhage H, Sowa R, et al. A distributed computing system for
multivariate time series analyses of multichannel neurophysiological
data. J Neurosci Methods. 2006;152:190–201.
Navarro V, Martinerie J, Le Van Quyen M, et al. Seizure anticipation in
human neocortical partial epilepsy. Brain. 2002;125:640–655.
Navarro V, Martinerie J, Le Van Quyen M, et al. Seizure anticipation: do
mathematical measures correlate with video-EEG evaluation? Epilep-
sia. 2005;46:385–396.
Osterhage H, Mormann F, Staniek M, Lehnertz K. Measuring synchroniza-
tion in the epileptic brain: a comparison of different approaches. Int J
Bifurcation Chaos. 2007, in press.
Pikovsky A, Rosenblum M, Kurths J. Synchronization: a universal concept
in nonlinear sciences. Cambridge, UK: Cambridge University Press,
2001.
Rudrauf D, Douiri A, Kovach C, et al. Frequency flows and the time-
frequency dynamics of multivariate phase synchronization in brain
signals. Neuroimage. 2006;31:209–227.
Schelter B, Winterhalder M, Eichler M, et al. Testing for directed influences
among neural signals using partial directed coherence. J Neurosci
Methods. 2006a;152:210–219.
Schelter B, Winterhalder M, Maiwald T, et al. Testing statistical significance
of multivariate time series analysis techniques for epileptic seizure
prediction. Chaos. 2006b;16:013108.
Schelter B, Winterhalder M, Dahlhaus R, et al. Partial phase synchronization
for multivariate synchronizing systems. Phys Rev Lett. 2006c;96:
208103.
Schiff SJ, Sauer T, Kumar R, Weinstein SL. Neuronal spatiotemporal pattern
discrimination: the dynamical evolution of seizures. Neuroimage. 2005;
28:1043–1055.
Schulze-Bonhage A, Kurth C, Carius A, et al. Seizure anticipation by
patients with focal and generalized epilepsy: a multicentre assessment
of premonitory symptoms. Epilepsy Res. 2006;70:83–88.
Sowa R, Chernihovskyi A, Mormann F, Lehnertz K. Estimating phase
synchronization in dynamical systems using Cellular Nonlinear Net-
works. Phys Rev E. 2005;71:061926.
Special issue on seizure prediction. J Clin Neurophysiol. 2001;18:191–282.
Special issue on epileptic seizure prediction. IEEE Trans Biomed Eng.
2003;50:537–648.
Stam CJ. Nonlinear dynamical analysis of EEG and MEG: review of an
emerging field. Clin Neurophysiol. 2005;116:2266–2301.
Staniek M, Lehnertz K. Parameter selection for permutation entropy mea-
surements. Int J Bifurcation Chaos. 2007, in press.
Suffczynski P, Lopes da Silva FH, Parra J, et al. Epileptic transitions: model
predictions and experimental validation. J Clin Neurophysiol. 2005;22:
288–299.
Wendling F, Bartolomei F, Bellanger JJ, Chauvel P. Epileptic fast activity
can be explained by a model of impaired GABAergic dendritic inhibi-
tion. Eur J Neurosci. 2002;15:1499–1508.
Wendling F, Bartolomei F, Bellanger JJ, et al. Epileptic fast intracerebral
EEG activity: evidence for spatial decorrelation at seizure onset. Brain.
2003;126:1449–1459.
Wendling F. Neurocomputational models in the study of epileptic phenom-
ena. J Clin Neurophysiol. 2005;22:285–287.
Wendling F, Hernandez A, Bellanger JJ, et al. Interictal to ictal transition in
human temporal lobe epilepsy: insights from a computational model of
intracerebral EEG. J Clin Neurophysiol. 2005;22:343–356.
Winterhalder M, Maiwald T, Voss HU, et al. The seizure prediction char-
acteristic: a general framework to assess and compare seizure prediction
methods. Epilepsy Behav. 2003;4:318–325.
Winterhalder M, Schelter B, Schulze-Bonhage A, Timmer J. Performance of
a seizure warning algorithm based on the dynamics of intracranial EEG.
Epilepsy Res. 2006;72:80–81.
Worrell G, Parish L, Cranstoun S, et al. High frequency oscillations and seizure
generation in neocortical epilepsy. Brain. 2004;127:1496–1506.
Journal of Clinical Neurophysiology • Volume 24, Number 2, April 2007 State-of-the-Art of Seizure Prediction
Copyright © 2007 by the American Clinical Neurophysiology Society 153
