Brain Dynamics and Modeling in Epilepsy:

Prediction and Control Studies

Leonidas Iasemidis, Shivkumar Sabesan, Niranjan Chakravarthy,

Awadhesh Prasad and Kostas Tsakalis

Abstract Epilepsy is a major neurological disorder characterized by intermittent

paroxysmal neuronal electrical activity, that may remain localized or spread, and
severely disrupt the brain’s normal operation. Epileptic seizures are typical mani-
festations of such pathology. It is in the last 20 years that prediction and control
of epileptic seizures has been the subject of intensive interdisciplinary research. In
this communication, we investigate epilepsy from the point of view of pathology
of the dynamics of the electrical activity of the brain. In this framework, we revisit
two critical aspects of the dynamics of epileptic seizures – the seizure predictability
and seizure resetting – that may prove to be the keys for improved seizure pre-
diction and seizure control schemes. We use human EEG data and the concepts
of spatial synchronization of chaos, phase and energy to first show that seizures
could be predictable in the order of tens of minutes prior to their onset. We then
present additional statistical evidence that the pathology of the brain dynamics prior
to seizures is reset mostly upon seizures’ occurrence, a phenomenon we have called
seizure resetting. Finally, using a biologically-plausible neural population mathe-
matical model that can exhibit seizure-like behavior, we provide evidence for the
effectiveness of a recently devised seizure control scheme we have called “feed-
back decoupling”. This scheme also provides an interesting dynamical model for
ictogenesis (generation of seizures).

1 Introduction

Epileptic seizures are manifestations of epilepsy, a neurological dynamical disor-

der second only to stroke. Of the world’s 50 million people with epilepsy, about
1/3 has seizures that are not controlled by anti-convulsant medication. One of the
most disabling aspects of epilepsy is the seemingly unpredictable nature of seizures.
If seizures cannot be controlled, the patient experiences major limitations in fam-
ily, social, educational, and vocational activities. These limitations have profound

L. Iasemidis (B)
The Harrington Department of Bioengineering and Electrical Engineering,
Arizona State University, Tempe, Arizona, USA

S.K. Dana et al. (eds.), Complex Dynamics in Physiological Systems: 185

From Heart to Brain, Understanding Complex Systems,
DOI 10.1007/978-1-4020-9143-8 12,  C Springer Science+Business Media B.V. 2009
186 L. Iasemidis et al.

effects on the patient’s quality of life, as well as on his or her family [1–3]. In
addition, status epilepticus, a life-threatening condition where seizures occur con-
tinuously, is treated only upon extreme intervention [4]. Until recently, the general
belief in the medical community was that epileptic seizures could not be anticipated.
Seizures were assumed to be abrupt transitions that occurred randomly over time.
However, theories based on reports from clinical practice and scientific intuition,
like the “reservoir theory” postulated by Lennox [5], existed and pointed out to
the direction of seizure predictability. Various feelings of auras, that is, patients’
reports of sensations of an upcoming seizure, also exist in the medical literature.
Penfield [6] was the first to note changes in the cerebral blood flow prior to seizures.
Deterministically predictable occurrences of seizures (reflex seizures) in a small
minority (about 3 to 5%) of epileptic patients have been reported as a result of
various sensory stimuli [7, 8]. These theories and facts have implied that seizures
might be predictable.
The ability to predict epileptic seizures well prior to their occurrences may lead to
novel diagnostic tools and treatment of epilepsy. Evaluation of anti-epileptic drugs
and protocols, in terms of duration of patients’ seizure susceptibility periods and/or
preictal (before a seizure) periods detected by seizure prediction algorithms, may
lead to the design of new, more effective and with less side effects drugs for early
disruption of the epileptic brain’s route towards a seizure. Electromagnetic stim-
ulation and/or administration of anti-epileptic drugs (AEDs) at the beginning of
the preictal period may disrupt the observed dynamical entrainment of normal brain
sites with the epileptogenic focus (the area that first exhibits the electrographic onset
of ictal activity), and lead to a significant reduction of epileptic seizures. Aside from
their immediate clinical applications to epilepsy, successful seizure prediction and
control algorithms could be useful for investigations into a wide variety of other
complex, nonstationary and spatio-temporal biological and physical systems that
undergo intermittent transitions.
The 80s saw the emergence of new signal processing methodologies, based on the
mathematical theory of nonlinear dynamics, for the study of spontaneous formation
of organized spatial, temporal or spatiotemporal patterns in physical, chemical and
biological systems [9, 10, 12, 13]. These methodologies quantify the signal structure
from the perspective of dynamical invariants (e.g. dimensionality of the attractor
through correlation dimension, or divergence of trajectories through the largest Lya-
punov exponent), and are a drastic departure from the signal processing techniques
based on the linear model (e.g. Fourier analysis). In 1988, a small group at the
University of Michigan, Ann Arbor, led by Iasemidis, Sackellares and Williams,
reported the first application of nonlinear dynamics to clinical epilepsy [14]. That
also led to the first NIH (National Institute of Health) supported clinical inves-
tigation into the nonlinear dynamics of epileptic seizures (“Dynamical studies in
temporal lobe epilepsy”, NIH-NINCDS RO1 NS31451). This group started to an-
alyze continuous, multichannel, preictal (before seizure), ictal (during seizure) and
postictal (after seizure) EEG from epileptic patients with temporal lobe epilepsy
devising new and modifying existing measures from the theory of chaos to quan-
tify the rate of divergence of trajectories (Lyapunov exponent) for the analysis of
Brain Dynamics and Modeling in Epilepsy: Prediction and Control Studies 187

EEG in epilepsy. The central concept was that seizures represented transitions of the
epileptic brain from its “normal” less ordered (chaotic) state to an abnormal, more
ordered state, and back to a “normal” state along the lines of chaos-to-order-to-chaos
transitions [15–24].
The dynamical modeling hypothesis changed some long-held beliefs about seizures.
Iasemidis et al. reported the first evidence that the transition to epileptic seizures
may be consistent with a deterministic process [15, 20], and that ictal electroen-
cephalogram (EEG) can be better modeled as an output of a nonlinear than a linear
system [17]. The existence of long-term preictal periods (order of minutes) was
shown using nonlinear dynamical analysis of subdural arrays [16], and raised the
feasibility of seizure prediction algorithms by monitoring the temporal evolution
of the short-term Lyapunov exponents (STLmax ) [22–24]. The possibility of focus
localization and seizure detection was also reported with the same technique in
1990 and 1994 respectively [18–21]. Since these initial results, several research
groups in the world started to work in the area of seizure prediction (see [25] for
a review). Elger and Lehnertz investigated the spatio-temporal dynamics of the
epileptic focus in 1994 [26,27], while Scott and Schiff directed attention to the time
structure of inter-ictal spikes [28]. Lopes da Silva et al. who have been developing
neurophysiology-driven dynamical models for EEG activity since the late 70s, quan-
tified state bifurcations in epileptogenesis [29]. Iasemidis and Sackellares perfected
their STLmax technique with the use of optimization techniques and the critical mass
hypothesis to predict seizures [30–36]. The fundamental issue that surfaced through
this group’s investigations in seizure predictions was the importance to locate and
use only the channels that carry information about an impending seizure. Changes
in the spatio-temporal structure of the EEG can in principle also be quantified by
sophisticated linear methods involving wavelet decompositions, coherence, and pat-
tern recognition methods such as artificial neural networks, and fuzzy approaches.
Several research groups are employing this approach towards the detection of the
preictal period [37, 38]. However, no prospective results on seizure prediction (i.e.
seizure prediction from long-term continuous EEG data using information only from
past seizure occurrences) have been reported in the literature with any of the above
measures yet, with the exception of Iasemidis et al.’s framework of analysis. The
important conclusion from all these different techniques is an accumulation of ev-
idence that there are measurable differences in the EEG prior to seizure onset that
can be utilized for epileptic seizure prediction.
Iasemidis’s group has reported a progressive preictal increase of spatiotemporal
entrainment/synchronization between critical sites of the brain as the precursor of
epileptic seizures. The algorithm used was based on the convergence of short-term
maximum Lyapunov exponents (STLmax ) among critical electrode sites selected
adaptively. This observation has been successfully implemented in the prospective
prediction of epileptic seizures [39]. Global optimization techniques were applied
for selecting the critical groups of electrode sites to observe preictal entrainment.
Seizure anticipation times of about 71.7 min with a false prediction rate of 0.12 per
h were reported. To further relate these findings to the mechanism of epileptoge-
nesis, these investigators found that majority of seizures in patients with temporal
188 L. Iasemidis et al.

lobe epilepsy (TLE) irreversibly reset (disentrain) postictally the observed preic-
tal dynamical entrainment [40, 41]. This supports the hypothesis that seizures do
not occur as long as there is no need to reset the brain. Last, but not least, this
group of researchers have also shown through simulations that, in chaos-to-order-
to-chaos transitions of general models of spatially coupled chaotic oscillators, with
increase/decrease of coupling convergence/divergence of the oscillators’ STLmax re-
sembles the observed preictal entrainment and postictal dynamical disentrainment
of the STLmax of critical brain sites. In addition, the model exhibited hysteresis,
a phenomenon that is also observed in the epileptic transition into and out of
seizures [41]. This dynamical view leads to a characterization of the seizure itself
as a mechanism that the brain has developed to reset the preictal entrainment when
a critical mass of sites, or a mass of specific, critical sites, is recruited.
In order to provide insights into the development of feedback control strategies
for suppression of seizures in the epileptic brain, and motivated by analysis and
results of burst phenomena in adaptive systems [42,43], Tsakalis and Iasemidis [44]
postulated the existence of an internal pathological feedback action in the epileptic
brain that lacks the ability to compensate for excessive increases in the network cou-
pling. This situation eventually leads to seizure-like transitions [44–47]. A precursor
of this scenario is an abnormal increase in synchronization that cannot be regulated
quickly enough by the existing pathological internal feedback mechanism. Using a
control-oriented approach, a functional model of an external feedback stimulation
paradigm was developed. During periods of abnormally high synchrony, this scheme
provides appropriate “desynchronizing feedback” to maintain “normal” synchro-
nization levels between neural populations. This feedback control view of epileptic
seizures and the developed seizure control strategies have been validated using cou-
pled chaotic oscillator models as well as biologically plausible neurophysiologic
models [46, 47].
In summary, from our group’s past and ongoing research, the following three
central results about epileptic seizures have emerged. First, we have shown that
seizures are manifestations of recruitment of brain sites in an abnormal hyper-
synchronization. The onset of such recruitment occurs long before a seizure and pro-
gressively culminates into a seizure. Seizures appear to be bifurcations of a neural
network that involve a progressive coupling of the focus with the normal brain sites
during a preictal period that may last days to tens of minutes. Thus, identification of
such a preictal period may constitute the basis for predicting an impending seizure
well in advance.
Second, postictally, time-irreversible resetting of the observed preictal dynami-
cal recruitment occurs (via a hysteretic loop). Preictal and postictal periods could
be mathematically defined and detected from the EEG. Complete or partial reset-
ting of the preictal entrainment of the epileptic brain after a seizure may affect the
route to the subsequent seizure. This may contribute to the observed nonstationary
nature of the seizure occurrences. Therefore, it is expected that estimation of the
magnitude of resetting at a seizure will improve our understanding of the brain’s
route to subsequent seizures, and may even lead to better seizure prediction and
control.
Brain Dynamics and Modeling in Epilepsy: Prediction and Control Studies 189

Third, through control-oriented modeling, a feedback control view of epileptic

seizures has been postulated, wherein epileptic seizures are hypothesized to be a
result of the inability of the internal feedback/regulatory mechanisms of the brain to
track excessive synchronization changes between the epileptogenic focus and other
brain areas prior to a seizure.
We herein present results on the preictal entrainment and brain resetting in EEG
data recorded from epileptic patients, as well as on the generation and control of
seizure-like behavior in a biologically-plausible mathematical model. Accordingly,
the first goal was to identify the most reliable synchronization measures, across
seizures in the same patient and across patients, that also issue the earliest warnings
of upcoming seizures. Second, in order to further shed light on the mechanisms of
seizures occurrence, the concept of seizure resetting is revisited. Third, a feedback
control scheme for generation and suppression of epileptic seizures is shown, after
we suitably modify a coupled neural population model from the literature to exhibit
“seizures”.
In the next Section 2, we first describe the available EEG data, and then the results
of the application of the different synchronization measures to the predictability of
the recorded epileptic seizures. In Section 3, we elaborate on the idea of brain reset-
ting. Novel measures that detect resetting, as well as the sensitivity and specificity of
resetting at seizure points, are investigated. We describe a feedback, systems-based
modeling of the “preictal” dynamics via simulations on coupled neural populations
in Section 4. In Section 5, we present a closed-loop seizure control strategy for the
epilepsy-prone model in Section 4. We further discuss these results and present our
conclusions in Section 6.

2 Predictability of Epileptic Seizures

In the present section, three of the most frequently utilized measures of dynamical
synchronization/entrainment, namely classical energy (E), phase (Φ) and short-term
Lyapunov exponent (STLmax ), are compared on the basis of their ability to detect
preictal changes. It is noteworthy that these three quantities, in the case of the com-
plex exponential basis signal x(t) = Aeat e j(ωt+φ) , correspond to Aeat , ωt + φ, and
␣ respectively, where ␣ is the real part of the pole of the Laplace transform of x(t)
and is equal to STLmax , phase ⌽ = ωt + φ is the imaginary part of the pole and it
depends on ␻, and energy E depends on A and ␣. Due to the current interest in the
field, and the proposed measures of energy and phase as alternatives to STLmax for
seizure prediction [38, 48, 49], it is important to comparatively evaluate each of the
three measures’ seizure predictability (anticipation) capabilities.
Quadratic integer programming techniques of global optimization were applied
to select the optimal electrode sites per measure and seizure, that is the ones that
exhibit maximum synchronization for every recorded seizure. Results from such an
analysis of 43 seizures, recorded from two patients with temporal lobe epilepsy,
showed that optimal sites selected on the basis of STLmax 10 min before a seizure