Stimulation-based anticipation and control of state transitions in the epileptic brain
Introduction
Prediction and dynamic control (through direct electric brain stimulation) of epileptic seizures have recently become both a clinical necessity and an intellectual challenge. Epilepsy is a neurological condition characterized by intermittent seizures that in most cases strike without clear warning. Being able to predict those transitions on the basis of some measurements, typically EEG signals, would significantly improve the quality of life of patients with epilepsy and may even open perspectives for adequate treatment of some pharmacologically resistant forms of the disease [1]. This objective, however, appears to be harder to reach than originally expected. On the one hand, the successes reported earlier were largely irreproducible: in many cases the success of prediction could not be validated as significantly better than a random guess. Accordingly, the literature regarding new prediction techniques based on sophisticated signal analytic methods that had once flourished (for a comprehensive review, see [2]) has recently diverted to questions regarding statistical validation of the prediction schemes [3], [4], [5]. A new quest for proving the “predictability” of epileptic seizures has emerged. Any predictive power statistically unexplainable by pure chance is cherished as precious evidence that seizures can in principle be predicted.
On the other hand, control of epileptic seizures using electrical stimulation is still in its early stages of development. Existing clinical trials use intracranial electrical stimulation consisting of periodic high-frequency (100–200 Hz) bursts that are administered at regular intervals [6]. Such schemes are referred to as “open loop” paradigms as they do not take into account the state of the neuronal system because they simply interfere with the ongoing activity of the tissue that is being stimulated. More advanced, state-dependent schemes called “closed loop” paradigms are in development [7]. Their success relies on the possibility of either foreseeing an impending seizure or detecting its occurrence early and only then applying electrical stimulation. A special class of these “closed loop” approaches uses properties of the measured EEG signal to determine the exact timing of the stimulation [8] or even to generate an appropriate seizure suppression stimulation waveform [9]. We refer to the latter as state-reactive control.
In this article we introduce and discuss some basic concepts relevant to the same quest. Our starting point is the question: “Why should we be able to predict and/or control epileptic seizures?” More specifically, we question what sort of dynamics governs the transitions from a “normal” to an epileptic state. Different dynamic scenarios imply different prospects for finding a reliable predictor of seizure transition and control strategy. In the next section, we briefly enlist some basically different possibilities, including stochastic and intermittent systems. We further distinguish between parameter-driven transitions, where the transitions are due to shifts in one or more parameters, and multistable dynamic systems, where input fluctuations or noise alone is sufficient to trigger the transition. Simulation examples based on analytical models illustrating the basic scenarios are given in Section 3. By studying the statistics of interictal intervals, we have shown that stochastic models cannot be ruled out in many in vivo models as well as in certain clinical cases. Section 4 is dedicated to understanding seizure predictability as a general statistical concept. We show that even noise-driven, stochastic systems may provide weak predictability of transitions between states, such that the estimation of the risk of such transition may be possible. In Section 5 we introduce the concept of “active observation paradigms”, where the response of the system to an external stimulation is analyzed. We consider that in multi-attractor systems, this method offers the possibility of “sensing” the emergence of new states, paroxysmal included. We refer to the method of external stimulation, using a carrier frequency, combined with the analytical tool, the so-called “relative Phase Clustering Index “(rPCI), that we introduced previously, as a paradigm that may be used for seizure anticipation in patients with temporal lobe epilepsy. In Section 6 we address the challenge of stopping or preventing transitions to the epileptic state by using either continuous or pulsed electrical stimulation. Furthermore, we demonstrate that the transitions from a normal to epileptic attractor can be reversed by an appropriate, state-dependent counterstimulus in the case of multistable system dynamics. Finally, we summarize our concepts and discuss possible implications and further strategies.
Section snippets
Generic dynamic scenarios of autonomous seizure generation
Although a good deal is known about the factors precipitating epileptic transitions in some conditions, no single theory or model can account for such transitions in general. Accordingly, the realm of seizure prediction has so far focused predominantly on phenomenological trial-and-error approaches. In some cases, seizure prediction strategies have been derived from theoretical assumptions involving some elegant mathematical concepts.
Here we attempt to take into account different dynamic
Metaphoric models of epileptic dynamics
There is a longstanding debate on detailed versus lumped modeling in biology in general and in neurophysiology in particular. In relatively simple physical systems such as ideal gases and crystals, most of the relevant properties can be described in the “thermodynamic limit” by replacing the microscopic degrees of freedom with few macroscopic-order parameters. Complex biological systems, in contrast, do not seem to have such a straightforward scale of simplification. Critical information can be
Prediction and predictability
The central question in all the aforementioned dynamic scenarios and model considerations is the possibility of predicting the transitions by measuring a suitable feature(s) of the system. We have already noted that in the case of a simple intermittent system, knowing the spectrum of Lyapunov exponents can indicate an impending transition to the chaotic phase. But what if the “stochastic” scenario governs the transition dynamics through either a parameter “random-walk” or a fluctuation-induced
Active observation paradigms in detection of salient states
Knowing how to validate the power of any proposed seizure predictor still leaves unsolved the problem of finding good candidates for such a predictor. Assuming that seizures are generated by random parameter fluctuations we can infer that all that we can “predict” is the closeness of the system to an epileptic transition. If, alternatively, the system generates seizure activity due to salient states of a multistable system, the risk of an impending seizure will be unobservable until the actual
State-reactive control of model seizures in bistable systems
One of the most attractive points of the multistable scenario of seizure generation is that it allows for an acute intervention. We have shown in realistic corticothalamic models that if detected in time, an epileptic seizure can be aborted with even a single stimulation pulse [21]. Here we expand the concept including continuously operating external modules designed to constantly monitor and react to epileptic transitions. We use the same analytical models as described in Section 3.
Our first
Conclusions
In this article we chose to follow a nonempirical approach to seizure prediction and control. Our approach is based on the assumption that epileptic seizures are due to interstate transitions in dynamic systems having multiple attractors. However, this is not the only possible scenario; parameter changes and chaotic transitions (intermittency) might play a role in certain cases of epilepsy. To validate our a priori assumptions we explored statistics of ictal and interictal durations, and we
References (36)
Non-linear time series analysis of intracranial EEG recordings in patients with epilepsy: an overview
Int J Psychophysiol
(1999)- et al.
Neuronal complexity loss in temporal lobe epilepsy: effects of carbamazepine on the dynamics of the epileptogenic focus
Electroencephalogr Clin Neurophysiol
(1997) - et al.
Charting epilepsy by searching for intelligence in network space with the help of evolving autonomous agents
J Physiol Paris
(2004) - et al.
Computational model of thalamo-cortical networks: dynamical control of alpha rhythms in relation to focal attention
Int J Psychophysiol
(2001) - et al.
Dynamics of non-convulsive epileptic phenomena modeled by a bistable neuronal network
Neuroscience
(2004) - et al.
Interpretation of interdependencies in epileptic signals using a macroscopic physiological model of the EEG
Clin Neurophysiol
(2001) - et al.
Electrical brain-stimulation paradigm for estimating the seizure onset site and the time to ictal transition in temporal lobe epilepsy
Clin Neurophysiol
(2005) - et al.
Bidirectional multisite seizure propagation in the intact isolated hippocampus: the multifocality of the seizure ‘‘focus”
Neurobiol Dis
(2006) - et al.
Seizure prediction and the preseizure period
Curr Opin Neurol
(2002) - et al.
Seizure prediction: the long and winding road
Brain
(2007)
Testing the null hypothesis of the nonexistence of a preseizure state
Phys Rev E
The statistics of a practical seizure warning system
J Neural Eng
Predictability analysis for an automated seizure prediction algorithm
J Clin Neurophysiol
Electrical stimulation of the hippocampal epileptic foci for seizure control: a double-blind, long-term follow-up study
Epilepsia
Automated seizure abatement in humans using electrical stimulation
Ann Neurol
Seizure abatement with single dc pulses: is phase resetting at play?
Int J Neural Syst
Tracking and control of neuronal Hodgkin–Huxley dynamics
Phys Rev E
Dynamical diseases of brain systems: different routes to epileptic seizures
IEEE Trans Biomed Eng
Cited by (100)
From phenomenological to biophysical models of seizures
2023, Neurobiology of DiseaseInitiation and termination of epilepsy induced by Lévy noise: A view from the cortical neural mass model
2023, Chaos, Solitons and FractalsEpileptic seizure prediction in intracranial EEG using critical nucleus based on phase transition
2022, Computer Methods and Programs in BiomedicineCitation Excerpt :Electroencephalogram (EEG) is the recording of the electrical activity of the brain, which has been extensively used in diagnosing and detecting epileptic seizures [3–5] due to its high temporal resolution, low cost, and continuous monitoring. Considerable pieces of evidence [1,6–11] have shown that the underlying mechanism of seizure based on the epileptic EEG shares the general principles of other complex non-linear systems. Indeed, a variety of methods such as bifurcation theory [1,8], chaotic dynamics [9,10], fractal concepts [11], and statistical theory [12] have been employed to analyze features of the epileptic EEG, applicable in effective-seizures detecting.
Spatio-temporal evaluation of epileptic intracranial EEG based on entropy and synchronization: A phase transition idea
2022, Biomedical Signal Processing and Control