Teaching the Brain to Calm Seizures

Neurofeedback for epilepsy did not begin in a clinic but in a sleep lab with a room full of cats. In their 2006 review "Foundation and Practice of Neurofeedback for the Treatment of Epilepsy," Sterman and Egner bring together decades of animal and human work to show how training the brain’s own rhythms can reduce seizures in people whose epilepsy has resisted standard medication. This is from the archives rather than brand‑new research, but it remains one of the most important conceptual foundations for how we still work with seizure disorders in neurofeedback today.

Broadly speaking, biofeedback uses real‑time information from the body (like heart rate, breathing, or skin conductance) to help people learn to change their own physiology. Neurofeedback is a specific form of biofeedback that does this with brain activity, typically measured with EEG, allowing people to practice increasing or decreasing particular brain rhythms while receiving moment‑to‑moment feedback.

In the case of epilepsy, Sterman’s early experiments showed that cats can be trained to increase a specific sensorimotor rhythm (SMR, roughly 12–15 Hz) over their sensorimotor cortex, and that this learned rhythm is associated with bodily stillness and reduced motor excitability. Later, those same cats turned out to be more resistant to chemically induced seizures. That observation opened the door to a radically different approach to epilepsy: instead of only suppressing seizures with medication, what if we could raise the brain’s seizure threshold by training its own stabilising rhythms?

Sterman and Egner’s review traces how this idea developed into a clinical method, how SMR‑based neurofeedback is thought to work in the brain, and how modern practice integrates quantitative EEG mapping, careful operant conditioning design, and collaboration with medical care. It is both a historical tour and a practical roadmap for using neurofeedback as a serious therapeutic option for seizure disorders.

Methods

From cats to cortex: how SMR training works

The review starts with the basic physiology of SMR. In Sterman’s early animal work, hungry cats were trained to suppress a previously rewarded lever‑press for food. During the still, non‑moving periods that followed this learned inhibition, a distinct 12–20 Hz rhythm appeared over sensorimotor cortex, with a spectral peak around 12–14 Hz. This was named the sensorimotor rhythm.

Recordings in these cats showed that SMR is not just a surface phenomenon. It arises from rhythmic bursting in the ventrobasal nuclei (nVB) of the thalamus, which relay somatosensory information to sensorimotor cortex. During SMR, nVB cells shift from fast, irregular firing to rhythmic bursts, reducing the flow of somatosensory input and lowering muscle tone. These thalamic oscillations then drive sensorimotor cortex and interact with the thalamic reticular nucleus to sustain a stable thalamocortical rhythm.

Crucially, SMR tends to appear when the body is still and motor intention is suppressed. Operant conditioning was then applied directly to this rhythm: cats received food whenever their SMR amplitude exceeded a threshold. They quickly learned to increase SMR, adopting a quiet, motionless posture with reduced muscle tone. Over extended training, SMR became more robust, and extinction tests showed the classic patterns of true operant learning.

Long‑term plasticity and seizure thresholds

Sterman and Egner argue that SMR training taps into mechanisms of long‑term potentiation (LTP). The thalamocortical volleys underlying SMR deliver strong, repeated input to pyramidal neurons in sensorimotor cortex. Because the animal is actively attending to the task, there is also top‑down activation of these same neurons. The timing and convergence of these inputs create ideal conditions for LTP: synapses in circuits that generate and stabilise SMR become stronger, and the system shifts toward a less excitable, more stable baseline. Over time, this may lower the likelihood that random fluctuations will escalate into runaway excitation and seizures.

A second oscillation, post‑reinforcement synchronization (PRS), emerges after rewards: a slower, high‑amplitude burst pattern over posterior cortex that appears to reflect transient arousal reduction after successful responses. PRS seems to scale with reward value and learning rate and may further consolidate the newly strengthened patterns.

From animal work to human protocols

In humans, SMR neurofeedback typically uses scalp EEG recorded from central regions over sensorimotor cortex (for example, C3, Cz, C4 in the 10–20 system). Training often targets enhancement of 12–15 Hz activity while simultaneously inhibiting slower theta (roughly 4–8 Hz) and sometimes very fast high beta activity. Feedback is given in discrete trials: when the desired SMR increase and unwanted frequency suppression are both achieved for a brief window (such as 0.25 seconds), a reward is delivered (a tone, visual change, counter increment), followed by a short pause before the next trial. This mirrors the operant structure used in the original animal studies.

Sterman and Egner also discuss an alternative approach using slow cortical potentials (SCPs) – shifts in the baseline EEG voltage that reflect changes in cortical excitability. Training patients to produce positive SCPs (higher thresholds, more inhibition) has also reduced seizures. However, the review focuses primarily on SMR training, as it is more widely used in clinical practice.

qEEG‑guided neurofeedback

By the time of this paper, Sterman had moved firmly toward quantitative EEG (qEEG) as a basis for individualising treatment. A standard protocol involves recording multi‑channel EEG (at least 19 sites) during eyes‑closed rest, eyes‑open rest, and task states. These data are then transformed into spectral magnitudes and compared with age‑matched normative databases. Deviations in specific frequency ranges at particular sites, as well as abnormal patterns of connectivity (for example, coherence or comodulation), are identified to guide training.

Importantly, Sterman and Egner recommend basing neurofeedback on raw magnitude spectra rather than squared power, and using narrow frequency bins rather than broad bands like “theta” or “alpha” to avoid statistical distortions and to respect individual variability. The treatment plan is then constructed to suppress deviant activity (for example, focal excess slow activity) and enhance stabilising rhythms such as SMR over relevant regions.

A detailed case example illustrates how this looks in practice: in a 37‑year‑old man with partial‑complex seizures after a head injury, qEEG showed abnormally high 6–8 Hz activity in the left temporal and adjacent central/prefrontal regions. Training was designed to inhibit 6–8 Hz at T3 (left anterior temporal) while enhancing 12–15 Hz at neighbouring C3 (left central). Sessions lasted about an hour, initially twice per week for 6 weeks, then weekly for 30 weeks, with thresholds adjusted over time to shape performance.

Results

Evidence from animal studies

The animal work provides unusually strong mechanistic support for a behavioural treatment. Cats trained to enhance SMR not only showed clear EEG learning curves and behavioural correlates (stillness, reduced muscle tone, reduced reflex excitability), but also lasting changes in sleep architecture: increased spindle density in sleep recordings weeks after training ended, suggesting a durable reorganisation of thalamocortical dynamics.

The most striking finding, however, came from toxicology experiments. When SMR‑trained cats were later exposed to a powerful convulsant fuel compound, they showed markedly higher seizure thresholds than untrained controls. In other words, the same SMR training that produced stillness and altered thalamocortical rhythms had effectively “hardened” the system against seizure induction.

Clinical outcomes in epilepsy

Human studies, reviewed in detail by Sterman and Egner, show encouraging and fairly consistent results, especially considering that participants were typically patients with difficult, medication‑resistant epilepsy.

Early open and partially controlled studies found that roughly 80% of patients trained to enhance SMR (often 12–15 Hz at central sites) achieved a clinically meaningful reduction in seizure frequency, typically defined as at least a 50% decrease relative to baseline. In some cases, seizures were eliminated entirely for extended follow‑up periods.

More rigorous designs added random feedback controls, yoked controls (where patients received feedback based on another person’s EEG), and ABA crossover designs. These showed that:

• Contingent SMR feedback, but not non‑contingent or random feedback, led to significant seizure reductions.

• When training contingencies were reversed (for example, reinforcing frequencies associated with instability), clinical gains could partially reverse, underlining the frequency‑specific, causal nature of the training.

In a landmark double‑blind study with 24 participants, patients were assigned to one of three groups: contingent SMR training (enhancing 11–15 Hz while suppressing slower and faster activity), yoked non‑contingent feedback, or a waiting‑list control. Only the contingent SMR group showed a significant and clinically substantial reduction in seizure frequency, with a median reduction of about 61% and a wide range up to complete seizure control.

Across the studies surveyed, Sterman summarises that 82% of 174 otherwise uncontrolled patients achieved at least a 50% reduction in seizure incidence. Around 5% became seizure‑free for up to a year after training cessation. These are not miracle‑cure numbers, but for a population defined by treatment resistance, they are highly meaningful.

Case example outcomes

The detailed case example illustrates both the promise and the realism of SMR neurofeedback. Over 42 sessions, the patient’s seizures declined from 2–4 per week to fewer than 2 per month, with some extended seizure‑free periods. A stressful legal event temporarily increased seizure frequency, highlighting the ongoing role of stress and sleep in seizure vulnerability, but reduction resumed afterwards. Medications were reduced, he returned to work, and follow‑up qEEG about a year into treatment showed complete normalization of the previously abnormal 6–8 Hz activity over the left centro‑temporal and prefrontal regions.

Limitations and critical points

Sterman and Egner are clear that more large‑scale, well‑funded clinical trials would be desirable. Neurofeedback research is labour‑intensive, under‑funded, and not supported by pharmaceutical industry resources, which has limited the number of very large, tightly controlled studies. Methodological variability across clinics and equipment also remains a challenge and has likely contributed to scepticism in some academic circles.

However, considering the severity and chronicity of the epilepsy subgroups studied, the consistency of results across methods and labs, and the strong neurophysiological rationale, the review concludes that SMR‑based neurofeedback for seizure disorders is a well‑founded and viable clinical option, particularly for patients who have not responded to medication.

Discussion

From mechanistic rationale to clinical reality

Taken together, Sterman and Egner’s data sketch a coherent story: SMR is a thalamocortical “standby” rhythm associated with motor stillness and reduced sensory inflow. Through operant conditioning, people can learn to increase this rhythm, engaging thalamic, cortical, and basal ganglia circuits in a way that promotes stability and reduces hyper‑excitability. Over time, this reshapes neural networks via synaptic plasticity, raising seizure thresholds so that the brain is less likely to tip into runaway synchronous firing.

Clinically, this means neurofeedback is not trying to “block” seizures in the moment the way a rescue medication would. Instead, it trains the system toward a different baseline: calmer, less reactive, and more resilient. That also explains why gains are often durable and why they can coexist with normal waking consciousness and functioning—the trained state is not sedation, but a more finely tuned sensorimotor network.

What this means for people living with epilepsy

For someone with epilepsy, particularly drug‑resistant epilepsy, the picture that emerges is cautiously hopeful. SMR neurofeedback requires time and commitment—tens of sessions over many months, focused attention during training, and ongoing management of sleep, stress, and medications. It is not a quick fix, and it does not guarantee seizure freedom.

But it offers something many people with chronic seizures have never been offered: a way to actively train their own brain networks to become more stable. Instead of relying solely on external agents that dampen excitability globally (and often produce cognitive or emotional side effects), neurofeedback invites the nervous system to practise a specific, physiologically grounded pattern of self‑regulation. Even when seizures do not disappear, reductions in frequency and severity, along with improvements in attention, mood, or sleep that often accompany SMR training, can significantly improve quality of life.

Implications for collaborative care

For neurologists, psychiatrists, and other referring professionals, this review argues that neurofeedback deserves a place in the broader treatment toolbox, especially for complex‑partial seizures and other difficult‑to‑treat syndromes. Its side effect profile is favourable, it works through mechanisms distinct from pharmacotherapy, and it can be combined with medication management rather than positioned as a competitor. In practice, Sterman emphasises close collaboration with neurologists, particularly when medications are being tapered as seizure control improves.

Importantly, the review highlights the variability of EEG pathology in epilepsy. Focal slowing, abnormal oscillations, paroxysmal transients, and medication effects can all shape the qEEG. This makes a one‑size‑fits‑all protocol risky. Instead, using multi‑site qEEG to identify individual patterns and then designing targeted protocols—such as inhibiting focal slow activity at a temporal site while enhancing SMR at a nearby central site—allows neurofeedback to address the specific functional disturbances underlying a person’s seizures.

Lessons for neurofeedback practitioners

For clinicians who already practise neurofeedback, Sterman and Egner’s paper is also a reminder of methodological discipline. Effective training in their model requires:

• High‑quality EEG acquisition, with careful attention to amplifier characteristics, sampling, and artefact handling.

• qEEG analysis that respects statistical constraints, uses narrow frequency bins, and compares against robust normative databases.

• Strict operant design with truly contingent, immediate feedback delivered in discrete trials, and minimal “entertainment” content that could overshadow the EEG contingencies.

They also stress that neurofeedback for epilepsy is not just a technical procedure. Extended, regular sessions create space for therapeutic relationship, motivation building, and careful monitoring of clinical course. Because the central response being trained is internal and not directly perceivable, the willingness of the person to engage with the task, stay curious, and persist through the early “I don’t feel anything” phases becomes critical.

A broader interpretive lens

Beyond epilepsy, Sterman and Egner’s model of SMR training as thalamocortical stabilization has influenced how we think about neurofeedback in other conditions characterised by hyper‑excitability or impaired inhibition: ADHD, impulsivity, anxiety, sleep disorders, and some traumatic brain injuries. The idea that rhythmic thalamocortical volleys, when trained in a context of active attention and reward, can reshape sensorimotor and basal ganglia circuits gives a powerful framework for understanding why SMR protocols often improve attention, impulse control, and motor restlessness.

At a deeper level, this work invites us to see neurofeedback not as a mysterious “brain hack”, but as a very specific application of well‑known learning principles to brain rhythms that have clear anatomical and physiological substrates. Operant conditioning, LTP, basal ganglia loops, cortical excitation thresholds—these are mainstream neuroscience concepts. SMR neurofeedback simply builds a clinical bridge from those concepts to the everyday struggles of people living with seizures.

Brendan’s perspective

Reading Sterman and Egner’s review, I always have the sense of watching neurofeedback grow up in fast‑forward. We start with cats quietly not pressing levers and end with a 37‑year‑old man who gets to go back to his job because someone trained his sensorimotor cortex to be less "twitchy". That arc captures why epilepsy work sits so close to the heart of the neurofeedback field.

In practice, SMR‑based work with seizures has taught me three big lessons about protocol design and clinical process.

First, the target is dynamic stability, not flatness. On paper, “reduce cortical excitability” can sound like we are trying to sand down every bump in the EEG. In the chair, it is much more subtle. The SMR band (roughly 12–15 Hz) at central sites like C3 and C4 acts like a stabilising anchor for sensorimotor networks. When I design protocols for someone with a seizure history, I am almost always thinking about how to strengthen that anchor while reducing noisy, unstable activity around it.

That often means an inhibit/enhance structure: for example, at C3, suppressing individual narrow slow bins (say, 5–7 or 6–8 Hz) that are elevated on their qEEG, plus sometimes very fast high‑beta ranges that reflect tension or cortical overdrive, while reinforcing clean 12–15 Hz. If there is a clearly focal lesion or slowing, like Sterman’s case at T3, I will pair a temporal inhibit (e.g., 6–8 Hz at T3 or T4) with central SMR enhancement, so we are both quieting the irritative zone and strengthening the overall network.

Second, individualisation matters more than recipe fidelity. Sterman was already warning, back in 2006, against generic “theta/beta” bands that ignore individual differences. That applies even more now. Two people with temporal lobe epilepsy can have very different qEEG patterns: one with focal 6–8 Hz excess over one temporal lobe, another with more diffuse slowing and bursts of paroxysmal activity riding on top.

So I start with the qEEG like a map, but I keep my ear tuned to the clinical story. If a person reports that their seizures cluster after nights of poor sleep and are preceded by a rising sense of internal “buzziness,” I know I will need to pay attention not only to SMR but also to sleep‑related rhythms and anxiety networks. That might mean adding gentle alpha uptraining at occipital or parietal sites for relaxation, or incorporating heart‑rate‑variability biofeedback between neurofeedback blocks to stabilise autonomic tone.

Third, epilepsy protocols remind us that life doesn’t stand still just because we are training the brain. The case Sterman describes, where a stressful legal process bumped seizures back up mid‑treatment, is completely typical. I routinely see seizure frequency track big changes: job loss, relationship upheavals, medication changes, even positive stress like starting school again. If we looked only at seizure counts, we might misinterpret those fluctuations as “neurofeedback stopped working.” When we zoom out, we see a nervous system that is still vulnerable, but now has more tools and a higher threshold than before.

In the chair, that translates into a lot of psychoeducation and expectation management. I’ll often say something like: “Neurofeedback tends to change the slope of the hill, not remove it entirely. Stress can still push you toward a seizure, but we’re trying to make it a steeper climb so it takes more to tip you over.” That framing helps people understand why adding stress management, sleep hygiene, and sometimes psychotherapy is not optional decoration but part of the treatment.

From a technical standpoint, SMR protocols are also a good reminder to privilege clean operant learning over entertainment. It is tempting, especially with younger clients, to make the feedback more and more game‑like. But Sterman’s data suggest that crisp, discrete trials with immediate, contingent feedback are what drive plasticity in the thalamocortical and basal ganglia circuits we care about. So even if we use engaging visuals, I want the EEG contingencies to be transparent, the rewards clearly tied to the brain’s behaviour, and the task framed like mental weight‑training rather than passive media consumption.

Finally, epilepsy work underscores how different clinical practice is from a cleanly controlled study. In research, you want stable meds, tidy inclusion criteria, and minimal comorbidity. In the real world, you get someone with seizures, panic attacks, chronic insomnia, and three medication changes in six months, plus a family who is understandably terrified. When neurofeedback “works” in that setting—fewer seizures, better sleep, less fog—it is rarely because we followed a perfect protocol template. It is because we combined solid physiology (SMR, qEEG guidance, careful inhibits) with flexible, compassionate clinical decision‑making, and because the person in the chair showed up, week after week, to teach their brain a different way to be.

That, to me, is the real message of Sterman and Egner’s paper: epilepsy neurofeedback is not magic and not a miracle cure. It is a disciplined application of neuroscience and learning theory that, when done well, can tilt the odds toward stability for people whose nervous systems have been living too close to the edge.

Conclusion

Sterman and Egner’s review of SMR neurofeedback for epilepsy offers a rare combination of mechanistic insight and clinical practicality. From thalamocortical oscillations in cats to qEEG‑guided protocols in humans, the work traces a clear line: training sensorimotor rhythms can reshape neural circuits in ways that raise seizure thresholds and improve day‑to‑day functioning.

The approach is not effortless—successful treatment usually requires dozens of sessions, careful integration with medical care, and a clinician who understands both EEG physiology and behaviour change. It is also not yet supported by the large number of gold‑standard trials we might wish for. But for people with drug‑resistant seizures, the existing evidence base and decades of clinical experience justify taking SMR‑based neurofeedback seriously as a treatment option, not a last‑ditch experiment.

Perhaps most importantly, this line of work reframes epilepsy management from something that is done to the brain to something that can be practised with the brain. By systematically reinforcing stable, well‑organised rhythms over sensorimotor cortex, neurofeedback invites the nervous system to participate actively in its own protection—a hopeful message for anyone living with recurrent seizures.

References

Sterman, M. B., & Egner, T. (2006). Foundation and practice of neurofeedback for the treatment of epilepsy. Applied Psychophysiology and Biofeedback, 31(1), 21–35. https://doi.org/10.1007/s10484-006-9002-x