Epilepsy II

Welcome to Novela’s series on Neurostimulation. We will explore applications to pathologies like epilepsy, Parkinson's disease and neuropsychiatric disorders. Our blog posts are educational, therefore we used simpler constructs when possible rather than the precise scientific terminology.

Author Jose Velazquez is a senior scientific advisory board member to Novela Neurotechnologies.

The previous blog was devoted to present general approaches to neurostimulation to treat epilepsy (epileptiform activity). Now we will focus on Deep Brain Stimulation (DBS) to stop seizures.

As mentioned in the last blog, it is important to realize, that DBS is a therapy to decrease seizures, but not to cure epilepsy. Epilepsy is the name given to the condition characterized by recurrent seizures. Seizures are just one manifestation of the abnormal epileptic brain, although the most noticeable. For a description of the diverse epilepsies, see Perez Velazquez and Wennberg, 2004.

The advantages of high level phenomena: neural synchronization

Let us recall another very basic aspect of nervous system function which was explained in the first blog: the functioning of brain cell networks relies on mutual activations of the constituent cells. (See paragraph ALTERING THE ACTIVITY OF NERVE CELLS)

The principal feature of the cellular activity in the nervous system is that of collective activity that results from cells activating one another in chains. These are mainly neurons but, to some extent, glial cells too regulate neuronal activity.

It is like mass action. Many neurons in one network fire action potentials with high degree of synchrony. That activates a connected set of neurons in a downstream network. Thus we see that synchronization of action potential firing is fundamental for proper brain function. Basic notions of brains activity and the importance of synchrony are expounded in some chapters in Perez Velazquez & Frantseva (2011).

The fact that synchrony of cellular activity is crucial gives us a chance to alter it by direct electrical stimulation of brain tissue. This is the basis of DBS. The fact that seizures represent a network phenomenon implies that it may not be crucial to place the stimulating electrodes in the focus (the brain region where seizures start is known as thefocus). Instead, it may be enough to place them in another area that is active during the ictus i.e. seizure. Here the electrical stimuli can break the chain and make the ictus halt. Nevertheless, normally it is the focus the brain region that is targeted by the neurosurgeon implanting the electrodes or removing tissue.

It is also an advantage that disparate microscopic events will lead to similar macroscopic patterns. In this case the “patterns” are the seizures. This may be a disadvantage for pharmacological treatments because the distinct microscopic events (molecules) that cause seizures in different patients will have to be treated with the adequate pharmacological therapy. However, for DBS this is irrelevant. Regardless of the molecular/genetic abnormalities, the important thing is, that there are highly synchronous events which underlie the ictus, and that the electrical pulses can suppress such abnormal high synchrony. (Those highly synchronous events are mainly firing of neuronal action potentials but, concomitant with this, there are also other related neurophysiological phenomena that display synchronicity.)

The dynamical dream in epilepsy therapy

The most widely used DBS protocol is the basically continuous intermittent stimulation. This was explained in the first blog “Brief overview of deep brain stimulation” (paragraph HOW CAN DEEP BRAIN STIMULATION HELP?). In the blog post it was mentioned that this is arather brute-force, trial-and-error approach that is blind to the intrinsic brain dynamics. This ignorance is a bit dangerous because lacking a deep understanding of brain dynamics, and how it changes after DBS, may result in pernicious side-effects.

These DBS methods, that operate almost continuously, are generally termed open-loop protocols. On the other hand, the “dynamical dream” in this field is the implementation of minimal perturbations. This means that a precisely timed brief stimulation with low frequencies and intensities would be used to stop the transition to the ictal event (Perez Velazquez and Wennberg, 2004).

Some understanding of the neural dynamics leading to seizures is needed to achieve that dynamical dream. Applying a short duration stimulus (say, a current pulse of a few seconds instead of minutes or hours) requires knowing somewhat accurately, when an ictal event will appear.

The reason that a stimulation is more efficient, if given just before the ictus, is because ictal events are considered dynamical bifurcations. As physics teaches us, systems are very sensitive to perturbations just at the moment ahead of the bifurcation. Therefore, if you want to alter brain activity with minimal effort, you should try to apply the perturbation – DBS, or any other means at hand – just ahead of the bifurcation point. In the blog post Epilepsy 1 it was mentioned that some patients know when they are about to have a seizure and have learnt an action that will stop it.

Dynamical System Theory

Allow me a brief digression into dynamical system theory. In simple words, a dynamical bifurcation is a qualitative change in dynamics of a phenomenon. Some research indicate that, in the nervous system, fluctuations in synchrony occur via dynamical bifurcations. Indeed, the existence of bifurcations in brain activity during epilepsy has been obtained in vivo (Perez Velazquez et al., 2003).

These dynamical bifurcations create patterns of organised neuronal activity. It is this organised activity that is the fundamental for a proper, healthy brain information processing. In seizures you find high cellular synchrony with little variability in the configurations of connections among diverse brain regions. Hence it is not good for sensorimotor processing, and therefore, loss of awareness is common during seizures.

There is an extensive literature suggesting that variability in brain activity is associated with good health—not only in neurophysiology, but also in cardiac activity, hormonal concentrations etc… Variability makes you healthy!

Now, armed with the knowledge that the best moment to perturb the brain dynamics leading to the seizure is just before the impending ictus, then, to apply the DBS at that moment we have to know with certain accuracy, when the ictus will materialise.

This is the realm of a very popular field, and some scientists prefer to use one term rather than the other: seizure prediction, or anticipation, or forecast. However, for our purposes—we need to know when the seizure will appear—let’s assume all these words mean the same.

On seizure prediction

Implementing the aforementioned dynamical dream(closed-loop DBS) starts with anticipating an impending ictus. It is the basis of a personalized approach to seizure control. I have termed it dynamiceutical approach (Perez Velazquez, 2017).

Later on, I will describe our own studies. In those studies, we used electrodes manufactured by Novela, on a feedback—or closed-loop—DBS protocol. The “predictor” of seizures was synchronization (Salam et al., 2015).

Trends in synchronization among neuronal networks have been investigated for quite some time as a possible indicator of approaching ictal periods. Particularly, differences in the phase of the recorded brain signals have been scrutinised.

However, a variable number of results can be found in the literature. Some studies have described a decrease in phase synchronizationduring the pre-ictal period, just before the seizure(Mormann et al, 2000, 2003; Le van Quyen et al. 2001, Perez Velazquez et al., 2011).

Meanwhile, other studies reported an enhancement in local synchrony (van Putten, 2003),and still others found no clear trend in synchrony patterns. More importantly, some studies found trends in the spatio-temporal characteristics of synchrony, that were unique for each patient (Schevon et al., 2007). This emphasises that a personalised approach is needed.

But the field of seizure prediction has had a long history, and recently, it has been investigated with an unparalleled enthusiasm, within the field of the dynamics of epilepsy.

Nevertheless, already in the 1970s, McDonnell Douglas Astronautics Co. funded projects on seizure predictability from EEG recordings. These projects were abandoned, due to variable outcomes across patients—afirst indication of what was to come.

The end result of many studies is that, presently, a host of seizure prediction algorithms exist. The number of such algorithms is probably even more than 30 (see figure 1 in Le van Quyen et al. (2001)for the state of affairs at the turn of the century).

Some of these algorithms have been reviewed in many publications, e.g. Litt and Lehnertz, (2002), Ebersole (2005), Alotaiby et al. (2014). These methods all seem to work to some extent. The prediction time windows range from a few minutes to hours.

As aforementioned, the variables used in the creation of these algorithms are quite a few: from variation in synchrony patterns to complexity measures and correlation integrals (Lerner, 1996).

Failed attempts to anticipate seizures have been reported in Aschenbrenner-Scheibe et al. (2003) and Harrison et al. (2005), to cite a few. Such negative results are almost always related to the generality of the validity of the different methods for all epileptic patients.

However, whereas there is no one general algorithm applicable to all patients, the vast majority of algorithms work in specific cases. Therefore, we can say that nearly all seizure-prediction algorithms succeed, at least to some extent – but only in certain cases.

The question is whether a general seizure-prediction rule, that would anticipate seizure occurrence in the immense majority of epileptic cases, can be found. Our contention is that, most likely, not.

As an illustration, let’s look at one variable that was thought could be of value for seizure prediction. That is, possible changes in synchronization before the ictal events. In fact, synchrony does seem to change as the ictus approaches, but it is patient-specific, as reported by Aarabi et al. (2008). The synchrony increases in 63% of seizures during a presumed “pre-ictal” state, and decreases in 31% of the cases studied.

Perez Velazquez et al. (2007) described, based on MEG recordings, phase synchrony decreases before (and after) seizures in restricted cortical areas. In other words, depending on what MEG sensors were taken to be analysed, a change would be appreciable or not.

Finally, Schevon et al. (2007) found that the variation in synchrony was unique to each patient, and Winterhalder et al. (2006) reported that 50% of the patients showed a trend, either increase or decrease in synchrony. These authors judiciously recommend that “the prediction methods […] have to be determined for each patient individually”.

In all these discussions on seizure prediction something seems to be taken for granted: the predictability of ictal events. But, are seizures really predictable? (Bahar, 2006).

Within the framework of brain multistable/metastable phenomena, a fundamental role of noise can be expected. The result of noise-induced transitions in brain activity may lead to unpredictability of brain states.

In addition, the presence of bifurcations and critical states in brain dynamics furnishes another reservation regarding the possible predictability.

Nevertheless, to end this section on seizure anticipation, let us say that it is very probable that for each patient there is an algorithm that will predict upcoming seizures. It is all a matter of investigating each patient’s brain dynamics. That is, a personalised approach.