Applied Psychophysiology and Biofeedback, Vol. 31, No. 1, March 2006 (C ?2006)
Foundation and Practice of Neurofeedback
for the Treatment of Epilepsy
M. Barry Sterman1,3and Tobias Egner2
Published online: 14 April 2006
This review provides an updated overview of the neurophysiological rationale, basic and
clinical research literature, and current methods of practice pertaining to clinical neuro-
feedback. It is based on documented findings, rational theory, and the research and clinical
experience of the authors. While considering general issues of physiology, learning prin-
ciples, and methodology, it focuses on the treatment of epilepsy with sensorimotor rhythm
(SMR) training, arguably the best established clinical application of EEG operant condi-
tioning. The basic research literature provides ample data to support a very detailed model
its efficacy in clinical treatment. Further, while more controlled clinical trials would be
desirable, a respectable literature supports the clinical utility of this alternative treatment
for epilepsy. However, the skilled practice of clinical neurofeedback requires a solid under-
standing of the neurophysiology underlying EEG oscillation, operant learning principles
and mechanisms, as well as an in-depth appreciation of the ins and outs of the various
hardware/software equipment options open to the practitioner. It is suggested that the
best clinical practice includes the systematic mapping of quantitative multi-electrode EEG
measures against a normative database before and after treatment to guide the choice of
treatment strategy and document progress towards EEG normalization. We conclude that
the research literature reviewed in this article justifies the assertion that neurofeedback
treatment of epilepsy/seizure disorders constitutes a well-founded and viable alternative to
KEY WORDS: neurofeedback; neurotherapy; EEG; operant conditioning; epilepsy.
The origins of neurofeedback for the treatment of clinical disorders can be directly
traced to the first systematic demonstration of EEG operant conditioning in general (for
an in-depth review, see Sterman, 1996). In the context of sleep research, Sterman and
associates conducted a series of studies investigating learned suppression of a previously
rewarded cup-press response for food in cats (Roth, Sterman, & Clemente, 1967; Sterman
1Departments of Neurobiology and Biobehavioral Psychiatry, School of Medicine, UCLA.
2Functional MRI Research Center, Columbia University, Columbia.
3Address all correspondence to e-mail: firstname.lastname@example.org.
1090-0586/06/0300-0021/1 C ?2006 Springer Science+Business Media, Inc.
22 Sterman and Egner
& Wyrwicka, 1967; Sterman, Wyrwicka, & Roth, 1969; Wyrwicka & Sterman, 1968).
During learned suppression of this response, the appearance of a particular EEG rhythm
over sensorimotor cortex emerged above non-rhythmic low-voltage background activity.
This rhythm was characterized by a frequency of 12–20 Hz, not unlike EEG sleep spindles,
with a spectral peak around 12–14 Hz, and has been referred to as the “sensorimotor
rhythm” (SMR) (Roth et al., 1967). The investigators decided to study this distinct rhythm
directly, attempting to apply the operant conditioning method to see if cats could be trained
to voluntarily produce SMR, by making a food reward contingent on SMR production.
Cats easily accomplished this feat of EEG self-regulation, and the behavior associated with
SMR production was one of corporal immobility, with SMR bursts regularly preceded by
a drop in muscle tone (Sterman et al., 1969; Wyrwicka & Sterman, 1968).
In a serendipitous twist, Sterman’s laboratory was soon afterwards commissioned to
establishdose–response functions ofahighlyepileptogenic fuelcompound. Whenemploy-
ing the cats that had previously taken part in SMR conditioning as experimental animals,
these cats were found to display significantly elevated epileptic seizure thresholds com-
pared to untrained animals, suggesting that SMR training had somehow inoculated the cats
against experiencing seizures. Subsequently, this research was successfully extrapolated to
humans, where it was repeatedly documented that seizure incidence could be lowered sig-
nificantly (or on rare occasions abolished) by SMR feedback training (see Section Clinical
Findings with Epilepsy).
Owing to its close link to intracranial recordings in animals, the neurogenesis of SMR
somatosensory information (Fig. 1). During conditioned SMR production, nVB firing
patterns shift from fast and non-rhythmic (tonic) discharges to systematic, rhythmic bursts
of discharges (Harper & Sterman, 1972), which in turn are associated with suppression
of somatosensory information passage (Howe & Sterman, 1973) and reduction in muscle
tone. Upon reduction of afferent somatosensory input, the nVB cells hyperpolarize. Instead
of remaining at a stable level of inhibition, however, a gradual depolarization mediated by
a slow calcium influx causes the nVB neurons to discharge a burst of spikes, which are
relayed to sensorimotor cortex and thalamic reticular nucleus (nRT) neurons. Stimulation
of the latter in turn leads to a GABAergic inhibition of VB relay cells, thus returning them
to a hyperpolarized state and initiating a new cycle of slow depolarization. In this way, the
interplay between neuronal populations in nVB, nRt, and sensorimotor cortex results in
rhythmic thalamocortical volleys and consequent cortical EEG oscillations.
While attenuation of efferent motor and afferent somatosensory activity can initiate
SMR, the oscillatory activity is also largely influenced by non-specific cholinergic and
monoaminergic neuromodulation, which can affect excitability levels both in thalamic
relay nuclei and in the cortical areas receiving the relayed signals. During waking activ-
ity, the neuromodulator influences as well as cortical projections normally keep VB cells
depolarized and thus suppress rhythmic bursting patterns, while during behavioral still-
ness, oscillations at SMR frequency may be observed. As SMR constitutes the dominant
“standby” frequency of the integrated thalamocortical somatosensory and somatomotor
pathways, operant training of SMR is assumed to result in improved control over excitation
in this system. Increased thresholds for excitation in turn are thought to underlie the clinical
benefits of SMR training in epilepsy and other disorders characterized by cortical and/or
thalamocortical hyper-excitability. For instance, SMR training has been shown to be an
Foundation and Practice of Neurofeedback 23
Fig. 1. Sample of recorded sensorimotor EEG (top), nVB activity (RVPL). Posterior cortical EEG (R Post.
Marg Gy), electro-oculogram (EOG), and timing, relay and feeder indicators during SMR training in a cat.
Note correspondence in SMR between cortical EEG and VPL activity, reflecting origin of the SMR EEG
rhythm in thalamus. Note also that eye movement activity ceased during SMR. Additionally, it can be seen
that each rewarded burst of SMR is followed by a slower rhythmic pattern in posterior cortex, labeled as
“post-reinforcement synchronization”, or PRS (from Howe & Sterman, 1972).
effective treatment of attention deficit hyperactivity disorder (ADHD) (Fuchs, Birbaumer,
Lutzenberger, Gruzelier, & Kaiser, 2003; Monastra, Monastra, & George, 2002; Rossiter &
LaVaque, 1995; for a recent review see Monastra et al., 2005) and has furthermore been
documented to result in reduced impulsive response tendencies in healthy volunteers
(Egner & Gruzelier, 2001, 2004).
More recently, fMRI studies in human subjects have shown that the SMR EEG pat-
tern is clearly associated with an increase in metabolic activity in the striatum of the
basal ganglia nuclear complex (Birbaumer, 2005). Further, examining fMRI changes in
children with ADHD who improved significantly in cognitive tests after SMR neurofeed-
back training, Lavesque and Beauregard (2005) have observed a specific and significant
increase in metabolic activity in the striatum. Collectively, these findings support the notion
that the state changes underlying the SMR are associated with functional changes in the
ized anatomically as a system of fiber connections, which form a loop from cerebral cortex
and back to cerebral cortex via thalamic relays (Brodal, 1992). The two major components
of the striatum include the putamen/globus pallidus complex and the caudate nucleus. The
striatum has been attributed a role in managing background motor tone and the planning
phase of movements (Chevalier & Deniau, 1990; DeLong, 1990). The putamen provides
an inhibitory input to the globus pallidus. When the putamen is excited by pre-motor and
sensorimotor cortex the globus pallidus, which functions to inhibit various thalamic relay
nuclei projecting back to motor and pre-motor cortex, is itself inhibited, thus releasing
excitatory input to motor and sensorimotor cortex via the thalamic relays. When input to
24Sterman and Egner
the putamen from the sensorimotor cortex is reduced, as would be expected during SMR
activity, the globus pallidus becomes more active, thereby imposing inhibition upon its
thalamic relays to motor cortex. This inhibition would alter involuntary motor regulation,
reducing muscle tone and the intention to move.
Consistent with an activation of striatal inhibitory mechanisms, the studies mentioned
earlier have documented a reduction in background motor tone, reflex excitability, and
activity in extrapyramidal motor pathways during SMR bursts (Babb & Chase, 1974;
Chase & Harper, 1971; Harper & Sterman, 1972; Sterman & Wyrwicka, 1967; Sterman
et al., 1969). It is also evident that both animals and humans suppress the intention to move
(Sterman, 1996). This convergence of findings suggests that facilitation and/or regulation
of the SMR substrate alters motor output, and sets the stage for reduced proprioceptive
afferent input to thalamus. The important fact here may be that this reorganization of motor
and thalamic status is accompanied by volleys of strong oscillatory discharge to cortex with
each trained SMR response (for example, see Fig. 1), the relevance of which derives from
a different and highly significant area of investigation.
Findings in the study of synaptic mechanisms mediating experience-based neuronal
reorganization, and thus learning, provide an appealing theoretical basis for a potentially
repetitive afferent input to cortical and other forebrain neurons can promote increased
synaptic strength in relevant circuits (see reviews by Abel & Lattal, 2001; Malenka &
Nicoll, 1999; Soderling & Derkach, 2000; Walker, 2005). These changes, produced over
time by protein synthesis and the insertion of new excitatory transmitter channels at post-
synaptic receptor sites, result in a synaptic state called “long-term potentiation”, or LTP.
Under appropriate circumstances LTP increases synaptic sensitivity and the probability of
future activation in affected neuronal circuits.
of strong afferent discharge to sensorimotor cortex. This recurrent discharge arrives on the
and location of these strong, recurrent afferent bursts appears to be particularly significant.
Activation of distal portions of these same pyramidal neurons must accompany this input
due to the requisite attention associated with SMR conditioning. This convergence of
with recent concepts of “coincidence detection” and synaptic plasticity (Froemke, Poo, &
Dan, 2005). That is, the coincidence of strong thalamic afferent input near the pyramidal
cell body and back-depolarization from cognitively based distal excitation of the same
cell, magnifies local depolarization and subsequent LTP. Thus, the functional changes in
sensorimotor circuits mediating the discrete and recurrent onset of SMR activity in the
EEG may be specifically strengthened during feedback training through a progressive
potentiation process, resulting in a lasting decrease in sensorimotor excitability. Such
changes presumably would not occur during spontaneous sensorimotor rhythmic activity
due to the absence of coincident cognitive engagement.
There is abundant indirect evidence that such changes can indeed occur with SMR
operant conditioning. For example, when cats were provided with several weeks of SMR
training and then subjected to an extinction trial where the reward was withheld, there was
a marked increase in the expression of SMR activity during the initial period of extinction
Foundation and Practice of Neurofeedback 25
Fig. 2. Plots showing the output of SMR in the EEG of a cat per 20 s over 6 min periods of
vertical line). Feeder operation on variable interval schedule is indicated by plus marks under
abscissa. Plot at the top shows performance after approximately 1 month of 3/week training
sessions. Plot at the bottom shows identical trial after 3 months of training. Note that despite
performance and extinction is significantly increased after extended training (from Wyrwicka &
(Fig. 2). Such an increase is expected with true operant conditioning. However, when an
identical test was performed after several more months of training both the output of SMR
fact dramatically, increased. Additionally, studies in both animals and humans have found
that sleep recordings obtained after several months of SMR training were characterized
by a lasting increase in sleep spindle density when compared with pre-training recordings
26 Sterman and Egner
(Hauri, 1981; Sterman, Howe, & Macdonald, 1970; Sterman & Macdonald, 1978). Control
feedback conditions had no such effect.
A second EEG oscillation associated with SMR conditioning is seen after reward is
delivered, in the form of “post-reinforcement synchronization” (PRS), as shown in Fig. 1.
In animal studies, the incidence and magnitude of PRS has been found to be directly related
to the desirability of the reward (Sterman & Wyrwicka, 1967) and the rate of learning in
an operant conditioning task (Marczynski, Harris, & Livezey, 1981). Similar patterns have
been seen in humans as well (Sterman, 2005; Sterman, Kaiser, & Veigel, 1996). While
beyond the scope of this review, other neurophysiological findings have linked the PRS to a
We propose that it provides an operational definition for drive reduction (Sterman, 2005).
It is suggested further that, through additional LTP development, this EEG oscillation
following a correct operant response supports stabilization and further consolidation of the
emerging acquisition process.
To summarize, SMR activity reflects synchronized thalamocortical oscillations ini-
tiated by reduced proprioceptive input to ventrobasal thalamus, resulting from decreased
background muscle and reflex tone and suppressed movement. In the context of directed
attention, these oscillations project strong afferent volleys to cortical target neurons, which
result in a cascade of LTP-enhanced motor alterations. These changes are stabilized and
consolidated over time. This “unconscious” learning process is further stabilized through
arousal reduction and related EEG oscillations following reward in the form of PRS. This
enhancement appears to be progressive and sustained, effecting function beyond the neu-
It should be noted that a different neurofeedback approach, based on the measure-
ment of “slow cortical potential” (SCR) shifts (between positive and negative polarity) has
also proved successful in the treatment of epilepsy (e.g., Kotchoubey et al., 1999, 2001;
Rockstroh et al., 1993). Conceptually, SCP and SMR neurofeedback share the same goal
of reducing cortical excitability. As negative slow potentials are reflective of lowered ex-
citation thresholds (through depolarization) in the apical dendrites of cortical pyramidal
neurons, while positive slow potentials represent raised excitation thresholds (for a review,
see Birbaumer, 1997), SCP training in epileptics is aimed at enabling the patient to vol-
untarily produce cortical inhibition (i.e., positive SCPs), and thus interrupt seizure onset.
Interestingly, Birbaumer (2005) also reports that both the SMR pattern in the EEG and
learned increases in positive SCPs were associated with increased metabolic activity in
the striatum of the basal ganglia, suggesting a convergent effect of SMR and SCP training
(Birbaumer, 2005, also personal communication).
commonly employed by clinicians, the current review will focus exclusively on the more
widely used SMR neurofeedback. Further, we shall focus on SMR neurofeedback in the
treatment of epilepsy, primarily because this treatment has now been so well-documented
that the American Academy of Child and Adolescent Psychiatry (AACAP) considers
neurofeeback for seizure disorders to meet criteria for “Clinical Guidelines” of evidence-
based treatments, a recommendation that states that a particular practice should always be
medication treatment for ADHD (see Hirshberg, Chiu, & Frazier, 2005), a treatment that
is abundantly and perhaps excessively utilized today.
Foundation and Practice of Neurofeedback27
CLINICAL FINDINGS WITH EPILEPSY
Since the first single-case study, reported over 30 years ago (Sterman & Friar, 1972), a
fair number of controlled clinical studies, stemming from many different laboratories, have
produced consistent data on the efficacy of SMR training in epileptic patients. It is partic-
ularly noteworthy that these results have been achieved in an extremely difficult subgroup
of epilepsy patients, those with poorly controlled seizures who had proven unresponsive to
pharmacological treatment. We will here provide only a cursory overview of this clinical
(2000), while other recent summaries have also been provided by Monderer, Harrison, and
Haut (2002) and Walker and Kozlowski (2005).
In initial studies involving relatively small sample sizes and pre-treatment baseline
measures as a control condition, on average 80% of patients trained at enhancing SMR am-
plitudes were shown to display significant clinical improvements (Kaplan, 1975; Seifert &
(1979) reported that 3 months of SMR training was associated with significantly reduced
seizure incident in five out of seven patients who had previously suffered from very poorly
controlled seizures. These high success rates of SMR training were further confirmed in
investigations that employed more elaborate control conditions, such as non-contingent,
or random feedback (Finley, Smith, & Etherton, 1975; Kuhlman & Allison, 1978; Quy,
nature of SMR training, by documenting symptom reversal when reversing training con-
tingencies (Lubar & Bahler, 1976; Lubar et al., 1981; Sterman & MacDonald, 1978). In
a larger scale study (n=24), Lantz and Sterman (1988) employed a double-blind design,
with age- and seizure-matched patients assigned to either a contingent training sched-
ule of enhancing 11–15 Hz SMR activity (while simultaneously inhibiting slower and
higher frequencies), a non-contingent, “yoked control” training schedule, or a waiting list
control group. Significant reduction of seizure incidence with a median of 61% (range
0–100%) was found in the contingent SMR feedback group only. In the largest study to
baseline in 69 out of 83 patients participating in a mixed relaxation and SMR feedback
In reviewing the data accumulated in these studies, Sterman (2000) found that 82%
of 174 participating patients who were otherwise not controlled had shown significantly
improved seizure control (defined as a minimum of 50% reduction in seizure incidence),
with around 5% of these cases reporting a complete lack of seizures for up to 1 year
positive results, it has to be pointed out that additional replications are stillhighly desirable.
For a very promising treatment targeting such a serious condition as epilepsy, the number
of large-scale clinical trials of neurofeedback training to date is disappointing. A likely
reason for this state of affairs is that neurofeedback research is a very time- and work-
intensive enterprise that has traditionally not received extensive research funding and has,
for obvious reasons, not been pursued by the pharmaceutical industry. Nevertheless, the
neurofeedback training a valuable treatment option, particularly in drug non-responders.
28 Sterman and Egner
Neurofeedback therapy has been greatly aided by advances in computer technology
and software. However, the ease in the development of new software programs for feed-
back functionality and displays, together with the entry into the field of a diverse group
of professionals and semi-professionals, has led to an unfortunate lack of consensus on
methodology and standards of practice. In turn this has contributed to reluctance by the
academic and medical communities to endorse the field.
Based on the history outlined above, the practice of neurofeedback requires a funda-
mental understanding of the principles of relevant neurophysiology, operant conditioning,
neuropathology, and, ultimately, basic clinical skills. Concerning the latter, the unique
opportunity for extended contact with the client provided by this modality favors skilled
clinical insight and guidance. Further, since objective EEG, operant performance, and clin-
ical outcome data are the realm of neurofeedback, documentation and accountability can
and must be its coin. While some who practice neurofeedback pursue different models,
this review and its recommendations will be restricted to a consideration of practice that
respects and adheres to these principles.
Rational standards require that the neurofeedback treatment plan begins with a com-
prehensive quantitative EEG study (qEEG). This involves the collection of appropriate
samples of EEG data from at least 19 standardized sites over the cerebral cortex (placement
and states of task engagement, such as reading, visuo-spatial tracking, memory recall, and
problem-solving. These data are then digitized and subjected to frequency and amplitude
analysis using various versions of spectral transform and database procedures (Etevenon,
1986; Johnston, Gunkelman, & Lunt, 2005; Lorensen & Dickson, 2004). Quantitative and
the distribution of relevant frequencies recorded during these conditions, are then displayed
as tables and graphics and compared statistically with group data from an appropriate
normative database. Deviant local patterns, as well as disturbed interactions among sites,
can be identified through this analysis and used to direct neurofeedback training strategy.
Recent improvements to this methodology have reduced distortions produced by non-EEG
events (artifacts), biological cycles, and certain mathematical and statistical corruptions
(Kaiser & Sterman, 2001, 2005). The importance of accuracy in this analysis cannot be
overemphasized, since these findings will guide the development of neurofeedback treat-
ment strategies and, in turn, determine the quality of its application. Follow-up qEEGs after
a course of neurofeedback training can then provide for an objective assessment of EEG
changes related to treatment outcomes.
An example of a statistically significant deviant frequency pattern from a 37-year-old
shown are from the qEEG analysis program of the Sterman–Kaiser Imaging Laboratory
(SKIL). Plotted is the distribution of mean spectral magnitudes in five frequency bands
across 19 standard sites. Data were derived from three minute EEG recordings with the
eyes open, and are compared statistically with an age-matched normative database. Prior
to treatment (top) statistically significant and clearly abnormal magnitude increases can be
seen in left temporal lobe at both 5–7 and 7–9 Hz frequencies. Adjacent central cortical and
Foundation and Practice of Neurofeedback29
Fig. 3. Brain maps from eyes open data showing mean spectral magnitude distributions of five 3 Hz bands both
before and after neurofeedback training in a patient with partial-complex seizures secondary to brain injury. Color
scale at the left indicates database variance distribution as standard deviation. Pink areas show ≥2 standard
deviations. Prior to the treatment, there was significantly elevated abnormal activity at left centro-temporal and
pre-frontal cortex in 5–7 and 7–9Hz bands. After treatment, approximately 1 year later, an identical study showed
complete normalization. See text for details.
pre-frontal areas were corrupted as well. The neurofeedback treatment strategy employed
and the clinical outcome of this case is discussed below. It is important to note that this
strategy was based on the relevance of the SMR and the guidance of the qEEG findings,
both of which were essential for a truly functional approach to treatment.
The hardware and software used in the application of neurofeedback training strategy
should provide for the collection and evaluation of quality EEG signals needed both for
qEEG data acquisition and valid operant conditioning. A variety of equipment options is
available for qEEG recording, with a broad range of pricing, depending on component
quality and software functionality. It is important to note, however, that the more expensive
systems have been developed for research and medical applications, and were not designed
to address the assessment issues relevant to neurofeedback. This is, of course, more a
function of software than hardware.
30 Sterman and Egner
Most systems provide for the collection of valid EEG data but the interested pro-
fessional should do their homework in understanding and inquiring about such important
issues as amplifier noise, dynamic range, sampling rate, and filter settings.
The objective of the qEEG for neurofeedback treatment is to provide for the detection
and frequency/topography characterization of relevant EEG pathology. Analysis software
should thus give an accurate indication of localized frequency and topographic deviations,
and disturbances in the functional coordination of these variables among brain regions,
as indicated by metrics such as coherence or comodulation (Sterman & Kaiser, 2001;
Thatcher, 1992). It is important, however, to point out that the power transform (squaring
of magnitudes) is never used for neurofeedback treatment. Accordingly, some programs
avoid using this transform in the qEEG analysis, expressing data only as magnitude. This
correction also prevents the excessive skewing which results from squaring of EEG values,
and thus increases statistical validity (Kaiser, 2000; Sterman, Mann, Kaiser, & Suyenobu,
1994). Additionally, accurate specification of both normal and abnormal frequencies re-
quires analysis based on single frequency bins, and avoid the traditional bands such as
“theta” and “alpha”. These traditional “clinical bands” overlook significant individual dif-
ferences, and often distort the accuracy of relevant frequency specification (Kaiser, 2001;
Klimesch, Schimke, & Pfurtscheller, 1993).
Software used for neurofeedback training should be based on the need to apply mean-
ingful operant conditioning procedures to training objectives. A well established scientific
literature dictates the functional characteristics required for effective operant conditioning.
For example, feedback training should be configured to provide for discrete trials, a fun-
damental element of both classical and operant conditioning (Brogden, 1951; Ferster &
Skinner, 1957). That is, each rewarded response is an independent event, followed by at
least a brief pause prior to the next effort. Additionally, empirical data have demonstrated
that the response and reward must be truly contingent for optimal learning to occur, with
reward immediately following response (Felsinger, Gladstone, Yamaguchi, & Hull, 1947;
Grice, 1948). Further, events associated with both response and reward must not contain
content that can block or overshadow the desired EEG response (Pearce & Hall, 1978;
Williams, 1999). For example, a visual or auditory stimulus just preceding reward but not
related to the desired EEG response may acquire much of the reinforcement effect of the
reward. These requirements may not be met by all developers of neurofeedback programs,
and deserve close scrutiny by the potential customer. Finally, and perhaps of equal impor-
tance to the technical issues discussed above, software programs and training strategies
should stress exercise rather than entertainment, at least most of the time.
By definition, neurofeedback treatment for seizure disorders is today directed by the
pattern of EEG pathology detected through qEEG analysis, and by an appreciation for the
in which seizure pathology may manifest in the EEG. Also, the EEG can be significantly
affected by the anticonvulsant medications taken by the patient.
Atypical slow or fast EEG patterns may be observed. In some cases, these are accom-
panied by such transients as spike-and-wave discharge, sharp waves, or poorly organized,
high-amplitude events termed paroxysms, allof which should be noted but deleted fromthe
qEEG analysis in order to focus training on the more stable background EEG. Knowledge
of these EEG characteristics and complexities is essential for the appropriate application
of neurofeedback treatment. However, despite corruption produced by medications, the
literature in this field indicates that feedback strategies directed to the suppression of either
Foundation and Practice of Neurofeedback31
background and/or transient abnormal patterns, together with the enhancement of central
cortical SMR activity, results in the most effective therapeutic outcomes (for review see
Often the EEG abnormality disclosed by the qEEG is rather specific. For example, in
the case described above the most relevant pathology observed was essentially restricted
to the left centro-temporal area (Fig. 3, top), where the cranial impact was delivered. This
injury resulted in localized cortical hyper-excitability, with cognitive disturbances and the
financial institution, and was unable to resume work at the time he entered neurodfeedback
treatment. He experienced memory difficulties and emotional lability due to this injury, and
felt slowed by the anticonvulsant medications prescribed. He wanted to get his life back
Data from the eyes open qEEG were selected for neurofeedback guidance, since this
most approximated the training condition (Fig. 3). However, it should be pointed out that a
similar pattern of abnormality was seen in task states as well. The most affected frequency
range was between 6 and 8 Hz, which was significantly elevated. Thus, neurofeedback
treatment involved the suppression of 6–8 Hz activity from the left anterior temporal site
T3, while simultaneously increasing 12–15 Hz activity at the adjacent medial central site
C3. Rewards were obtained only when both conditions were met for at least a quarter of a
One-hour treatment sessions were provided twice per week for the first 6 weeks and
then once per week for the next 30 weeks. The equipment and procedures used conformed
to all of the principles discussed above. The display presented to the patient included two
adjacent vertical bars, one dark blue and the other light blue in color. He was required
to suppress the light blue bar (6–8 Hz activity at T3) below an established threshold line
set at 20% less than his baseline mean value, while raising the dark blue bar (12–15 Hz
activity at C3) to a threshold line at least 20% above baseline. If both of these objectives
were achieved simultaneously a counter in the top-center of the screen advanced one digit
and the unit sounded a pleasant tone. The system paused for 2 s and the task was repeated.
Prior to each in a series of sequential 3 min sets, the patient specified a numerical goal in
the final count for that set. At appropriate intervals the thresholds were adjusted to increase
the challenge and achieve “shaping” of the desired EEG response pattern.
Other display material can be used for neurofeedback as well. Puzzles, engaging
picture sequences, stop/start video clips, and other action themes have been used. These
alternatives to more simple displays like that described here. However, the issue of exercise
versus entertainment mentioned above is relevant in this regard. The patient should be
motivated to view the task much like a workout at a gym, and the feedback objectives as a
special exercise for the brain. Discrete trials with simple but relevant displays achieve this
The patient described above completed 42 training sessions. His seizure rate, which
had previously averaged 2–4 per week, declined progressively after the first month or so
of treatment to less than 2 per month. Some 4 months into the training he was involved in
a court case with serious personal implications. During these proceedings his seizure rate
again increased. This was to be expected due to the stress and related sleep loss caused
by these circumstances, since seizure disorders are at best managed but only rarely cured.
32 Sterman and Egner
seizure-free periods. His medications were reduced and he was able to return to work and
resume a more or less normal life. A follow-up qEEG obtained approximately 1 year after
the beginning of treatment showed no focal abnormalities, and a complete normalization
of quantitative EEG characteristics (Fig. 3, bottom).
This case was rather straightforward, and was not complicated by co-morbidities, a
long history of failed medications and their side effects, behavioral or economic adapta-
tions, or other disruptive factors. In general, patients seeking this remedy often present
such complications, a reality that necessitates cooperation with the neurologist and other
to, family and personal dynamics. But perhaps the most important variable determining
the success of neurofeedback is the clinician’s ability to instill a motivation to succeed
in the patient. Unlike operant conditioning of behavioral responses, where the organism
has a conscious awareness of the responses leading to rewards, with neurofeedback the
response is a subtle pattern of physiological changes of which the subject has little or no
direct experience. The drive to obtain a symbolic reward results in the reinforcement that
ultimately strengthens this response. This is where neurofeedback differs from traditional
operant conditioning, and why refined methodology, clear and meaningful rewards, proper
preparation of the patient, and appropriate clinical skills are essential.
It is important to point out that most of the epileptic patients who have participated in
neurofeedback research studies and many who seek this treatment today represent unques-
tionable failures of anticonvulsant drug therapies, particularly with complex-partial seizure
disorders. It is particularly noteworthy that positive outcomes have often been obtained in
the context of treating this extremely difficult sub-population of epilepsy patients. We view
it as unfortunate, therefore, that some professionals still criticize neurofeedback treatment
for the lack of more consistent or successful outcomes. On the contrary, evidence has
shown that most of these difficult patients benefit beyond any chance or placebo outcome,
and some do so dramatically. Considering the common side effects and costs associated
with life-long pharmacotherapy, we do not view neurofeedback treatment as a “last resort”
option for drug treatment-resistant cases only, but rather as a generally viable alternative
consideration for any patient suffering from seizures. Furthermore, in contrast to drug-
dependent symptom management, the altered modulation of thalamocortical excitability
through neurofeedback training may raise seizure thresholds sufficiently to greatly improve
the prospects for the long-term, non-dependent management of epilepsy. It must be added,
however, that the skilled application of neurofeedback requires a committed, well-trained,
and motivationally adept professional.
Abel, T., & Lattal, K. M. (2001). Molecular mechanisms of memory acquisition, consolidation and retrieval.
Current Opinion in Neurobiology, 11, 180–187.
Seizure, 1(2), 111–116.
Babb, M. I., & Chase, M. H. (1974). Masseteric and digastric reflex during conditioned sensorimotor rhythm.
Electroencephalography and Clinical Neurophysiology, 36, 357–365.
Birbaumer, N. (1997). Slow cortical potentials: Their origin, meaning, and clinical use. In G. J. M. Boxtel & K. B.
E. von B¨ ocker (Eds.), Brain and behaviour—past, present and future (pp. 25–39). Tilborg: University Press.
Foundation and Practice of Neurofeedback33
Birbaumer, N. (2005). Breaking the silence: Brain–computer interfaces in paralysis. Proceedings of the Annual
Conference, Int. Soc. for Neuronal Reg., 13, 2.
Brodal, P. (1992). The basal ganglia. In The central nervous system: Structure and function (pp. 246–261). New
York: Oxford University Press.
Brogden, W. J. (1951). Animal studies of learning. In S. S. Stevens (Ed.), Handbook of experimental psychology
(pp. 568–612). New York: Wiley.
Chase, M. H., & Harper, R. M. (1971). Somatomotor and visceromotor correlates of operantly conditioned 12–14
c/s sensorimotor cortical activity. Electroencephalography and Clinical Neurophysiology, 31, 85–92.
Chevalier, G., & Deniau, J. M. (1990). Disinhibition as a basic process in the expression of striatal functions.
Trends in Neuroscience, 13, 277–280.
Cott, A., Pavloski, R. P., & Black, A. H. (1979). Reducing epileptic seizures through operant conditioning of
central nervous system activity: Procedural variables. Science, 203, 73–75.
DeLong, M. R. (1990). Primate models of movement disorders of basal ganglia origin. Trends in Neuroscience,
Egner, T., & Gruzelier, J. H. (2001). Learned self-regulation of EEG frequency components affects attention and
event-related brain potentials in humans. NeuroReport, 12(18), 4155–4160.
Egner, T., & Gruzelier, J. H. (2004). EEG biofeedback of low beta band components: Frequency-specific effects
on variables of attention and event-related brain potentials. Clinical Neurophysiology, 115, 131–139.
of brain electrical activity (pp. 113–141). Boston: Butterworth.
Felsinger, J. M., Gladstone, A. L., Yamaguchi, H. G., & Hull, C. L. (1947). Reaction latency (StR) as a function
of the number of reinforcements. Journal of Experimental Psychology, 37, 214–228.
Ferster, C. B., & Skinner, B. F. (1957). Schedules of reinforcement. New York: Appleton-Century-Crofts.
Finley, W. W., Smith, H. A., & Etherton, M. D. (1975). Reduction of seizures and normalization of the EEG in
a severe epileptic following sensorimotor biofeedback training: Preliminary study. Biological Psychiatry, 2,
Froemke, R. C., Poo, M. M., & Dan, Y. (2005). Spike-timing-dependent synaptic plasticity depends on dendritic
location. Nature, 434, 221–225.
Fuchs, T., Birbaumer, N., Lutzenberger, W., Gruzelier, J. H., & Kaiser, J. (2003). Neurofeedback treatment
for attention-deficit/hyperactivity disorder in children: A comparison with methylphenidate. Applied Psy-
chophysiology and Biofeedback, 28, 1–12.
Grice, G. R. (1948). The relation of secondary reinforcement to delayed reward in visual discrimination learning.
Journal of Experimental Psychology, 38, 1–16.
Harper, R. M., & Sterman, M. B. (1972). Subcortical unit activity during a conditioned 12–14 Hz sensorimotor
EEG rhythm in the cat. Federation Proceedings, 31, 404.
Hauri, P. (1981) Treating psychophysiologic insomnia with biofeedback. Archives of General Psychiatry, 38,
Hirshberg, L. M., Chiu, S., & Frazier, J. A. (2005). Emerging brain-based interventions for children and adoles-
cents: Overview and clinical perspective. Child and Adolescent Psychiatric Clinics of North America, 14,
Howe, R. C., & Sterman, M. B. (1972). Cortical-subcortical EEG correlates of suppressed motor behavior during
sleep and waking in the cat. Electroencephalography and Clinical Neurophysiology, 32, 681–695.
Howe, R. C., & Sterman, M. B. (1973). Somatosensory system evoked potentials during waking behaviour and
sleep in the cat. Electroencephalography and Clinical Neurophysiology, 34, 605–618.
Johnstone, J., Gunkelman, J., & Lunt, J. (2005). Clinical database development: Characterization of EEG pheno-
types. Clinical EEG and Neuroscience, 36, 99–107.
Kaiser, D. A. (2000). QEEG: State of the art or state of confusion. Journal of Neurotherapy, 4, 57–75.
Kaiser, D. A., & Sterman, M. B. (2001). Automatic artifact detection, overlapping windows, and state transitions.
Journal of Neurotherapy, 4(3), 85–92.
Kaiser, D. A., & Sterman, M. B. (2005). Correcting sampling bias of tapering windows. International Journal of
Psychophysiology Manuscript submitted for publication.
Kaplan, B. J. (1975). Biofeedback in epileptics: Equivocal relationship of reinforced EEG frequency to seizure
reduction. Epilepsia, 16, 477–485.
Brain Topography, 5, 241–251.
tial shifts and the prediction of the outcome of neurofeedback therapy in epilepsy. Clinical Neurophysiology,
Kotchoubey, B., Strehl, U., Uhlmann, C., et al. (2001). Modification of slow cortical potentials in patients with
refractory epilepsy: A controlled outcome study. Epilepsia, 42, 406–416.
Kuhlman, W. N., & Allison, T. (1978). EEG feedback training in the treatment of epilepsy: Some questions and
some answers. Pavlovian Journal of Biological Science, 12(2), 112–122.
34 Sterman and Egner
Lantz, D., & Sterman, M. B. (1988). Neuropsychological assessment of subjects with uncontrolled epilepsy:
Effects of EEG biofeedback training. Epilepsia, 29(2), 163–171.
Levesque, J., & Beauregard, M. (2005). Effect of neurofeedback training on the neural substrates of selective
attention in children with attention-deficit/hyperactivity disorder: A functional magnetic resonance imaging
study. Neuroscience Letters.
Lorensen, T. D., & Dickson, P. (2004). Quantitative EEG Normative Databases: A comparative investigation.
Journal of Neurotherapy, 8, 53–68.
Lubar, J. F., & Bahler, W. W. (1976). Behavioral management of epileptic seizures following EEG biofeedback
training of the sensorimotor rhythm. Biofeedback and Self Regulation, 7, 77–104.
Lubar, J. F., Shabsin, H. S., Natelson, S. E., et al. (1981). EEG operant conditioning in intractible epileptics.
Archives of Neurology, 38, 700–704.
Marczynski, T. J., Harris, C. M., & Livezey, G. T. (1981). The magnitude of post-reinforcement EEG synchro-
nization (PRS) in cats reflects learning ability. Brain Research, 204, 214–219.
enting style on the primary symptoms of attention-deficit/hyperactivity disorder. Applied Psychophysiology
and Biofeedback, 27, 231–249.
Monastra, V. J., Lynn, S., Linden, M., Lubar, J. F., Gruzelier, J., & LaVaque, T. J. (2005). Electroencephalo-
graphic biofeedback in the treatment of attention-deficit/hyperactivity disorder. Applied Psychophysiology
and Biofeedback, 30, 95–114.
Monderer, R. S., Harrison, D. M., & Haut, S. R. (2002). Neurofeedback and epilepsy. Epilepsy and Behavior, 3,
Pearce, J. M., & Hall, G. (1978). Overshadowing the instrumental conditioning of a lever-press response by a
more valid predictor of reinforcement. Journal of Experimental Psychology: Animal Behavior Processes, 4,
Quy, R. J., Hutt, S. J., & Forrest, S. (1979). Sensorimotor rhythm feedback training and epilepsy: Some method-
ological and conceptual issues. Biological Psychology, 9, 129–149.
Rossiter, T. R., & LaVaque, T. J. (1995). A comparison of EEG biofeedback and psychostimulants in treating
attention deficit hyperactivity disorders. Journal of Neurotherapy, 1, 48–59.
Rockstroh, B., Elbert, T., Birbaume, N., et al. (1993). Cortical self-regulation in patients with epilepsies. Epilepsy
Research, 14, 63–72.
Roth, S. R., Sterman, M. B., & Clemente, C. C. (1967). Comparison of EEG correlates of reinforcement, internal
inhibition, and sleep. Electroencephalography and Clinical Neurophysiology, 23, 509–520.
Seifert, A. R., & Lubar, J. F. (1975). Reduction of epileptic seizures through EEG biofeedback training.Biological
Psychology, 3, 157–184.
Soderling,T.R., &Derkach,V. A.(2000).Postsynaptic proteinphosphorylationandLTP.Trends in Neuroscience,
Sterman, M. B. (1996). Physiological origins and functional correlates of EEG rhythmic activities: Implications
for self-regulation. Biofeedback and Self Regulation, 21, 3–33.
conditioning. Clinical Electroencephalography, 31(1), 45–55.
Sterman, M. B. (2005). Principles of neurotherapy. Proceedings of the Annual Conference, Int. Soc. for Neuronal.
Reg. 13, 23.
Sterman, M. B., & Friar, L. (1972). Suppression of seizures in an epileptic following sensorimotor EEG feedback
training. Electroencephalography and Clinical Neurophysiology, 33, 89–95.
Sterman, M. B., & Kaiser, D. A. (2001). Comodulation: A new QEEG analysis metric for assessment of structural
and functional disorders of the CNS. Journal of Neurotherapy, 4(3), 73–83.
Sterman, M. B., Kaiser, D. A., & Veigel, B. (1996). Spectral analysis of event-related EEG responses during
short-term memory performance. Brain Topography, 9(1), 21–30.
Sterman, M. B., & MacDonald, L. R. (1978). Effects of central cortical EEG feedback training on incidence of
poorly controlled seizures. Epilepsia 19, 207–222.
Sterman, M. B., MacDonald, L. R., & Stone, R. K. (1974). Biofeedback training of the sensorimotor EEG rhythm
in man: Effects on epilepsy. Epilepsia 15, 395–417.
Sterman, M. B., Mann, C. A., Kaiser, D. A., & Suyenobu, B. Y. (1994). Multiband topographic EEG analysis of
a simulated visuomotor aviation task. International Journal of Psychophysiology, 16, 49–56.
Sterman, M. B., Howe, R. D., & Macdonald, L. R. (1970). Facilitation of spindle-burst sleep by conditioning of
electroencephalographic activity while awake. Science, 167, 1146–1148.
Sterman, M. B., & Wyrwicka, W. (1967). EEG correlates of sleep: Evidence for separate forebrain substrates.
Brain Research, 6, 143–163.
Sterman, M. B., Wyrwicka, W., & Roth, S. R. (1969). Electrophysiological correlates and neural substrates of
alimentary behavior in the cat. Annals of the New York Academy of Sciences, 157, 723–739.
Foundation and Practice of Neurofeedback35
Thatcher, R. W. (1992). Cyclic cortical reorganization during early childhood. Brain and Cognition, 20, 24–50.
Walker, M. P. (2005). A refined model of sleep and the time course of memory formation. Behavioral and Brain
Sciences, 28, 51–64.
Walker, J. E., & Kozlowski, G. P. (2005). Neurofeedback treatment of epilepsy. Child and Adolescent Psychiatric
Clinics of North America, 14, 163–176.
Williams, B. A. (1999). Associative competition in operant conditioning: Blocking the response–reinforcer asso-
ciation. Psychonomic Bulletin and Review, 6, 618–623.
Wyrwicka, W., & Sterman, M. B. (1968). Instrumental conditioning of sensorimotor cortex EEG spindles in the
waking cat. Physiology and Behavior, 3, 703–707.
Page 16 Download full-text