Changes in EEG power spectra during biofeedback of slow cortical potentials in epilepsy.
ABSTRACT The goal of the study was to explore parallel changes in EEG spectral frequencies during biofeedback of slow cortical potentials (SCPs) in epilepsy patients. Thirty-four patients with intractable focal epilepsy participated in 35 sessions of SCP self-regulation training. The spectral analysis was carried out for the EEG recorded at the same electrode site (Cz) that was used for SCP feedback. The most prominent effect was the increase in the theta 2 power (6.0-7.9 Hz) and the relative power decrement in all other frequency bands (particularly delta 1, alpha 2 and beta 2) in transfer trials (i.e., where patients controlled their SCPs without continuous feedback) compared with feedback trials. In the second half of the training course (i.e., sessions 21-35) larger power values in the delta, theta, and alpha bands were found when patients were required to produce positive versus negative SCP shifts. Both across-subject and across-session (within-subject) correlations between spectral EEG parameters, on the one hand, and SCP data, on the other hand, were low and inconsistent, contrary to high and stable correlations between different spectral variables. This fact, as well as the lack of considerable task-dependent effects during the first part of training, indicates that learned SCP shifts did not directly lead to the specific dynamics of the EEG power spectra. Rather, these dynamics were related to nonspecific changes in patients' brain state.
- SourceAvailable from: Alkinoos Athanasiou[Show abstract] [Hide abstract]
ABSTRACT: While the field of brain computer interfaces (BCIs) has produced impressive results regarding movement and communication restoration in patients with disability, among its less known clinical applications lie the array of treatment-resistant epileptic conditions. The control of BCI systems relies on brain activity control and regulation and it was to be expected that such systems would be tested for the regulation of abnormal brain activation in epilepsy. A few electroencephalographic (EEG) features have been used as BCI modalities for that cause. Such features are the slow cortical potentials (SCPs) and sensorimotor rhythm (SMR) regulation, and have been tested on epileptic patients with promising results. These methods – especially when used as supplementary to classic treatment - have produced superior results and appear to benefit long-term seizure suppression, seizure prevention and improvement of life quality. In certain cases an increase in cognitive functioning and IQ score has been observed.7th European Symposium on Biomedical Engineering, Halkidiki, Greece; 05/2010
Chapter: Biofeedback.[Show abstract] [Hide abstract]
ABSTRACT: Biofeedback is an evidence-based approach to enhancing personal awareness and control over body and mind. Biofeedback combines the values of the complementary and alternative medicine movement with the biotechnology of modern scientific medicine. The basic biofeedback paradigm suggests that whenever we provide a human being with feedback on about a biological process, that feedback enables the individual to increase awareness of the process, and gain conscious control. Biofeedback uses electronic instruments to monitor and feed back information on about physiological responses1,2.Textbook of complementary and alternative medicine, 2nd edited by C.-S. Yuan, E. J. Bieber, B. A. Bauer, 01/2006: chapter Biofeedback.; Informa Healthcare..
- [Show abstract] [Hide abstract]
ABSTRACT: Neurofeedback training procedures designed to alter a person's brain activity have been in use for nearly four decades now and represent one of the earliest applications of brain computer interfaces (BCI). The majority of studies using neurofeedback technology relies on recordings of the electroencephalogram (EEG) and applies neurofeedback in clinical contexts, exploring its potential as treatment for psychopathological syndromes. This clinical focus significantly affects the technology behind neurofeedback BCIs. For example, in contrast to other BCI applications, neurofeedback BCIs usually rely on EEG-derived features with only a minimum of additional processing steps being employed. Here, we highlight the peculiarities of EEG-based neurofeedback BCIs and consider their relevance for software implementations. Having reviewed already existing packages for the implementation of BCIs, we introduce our own solution which specifically considers the relevance of multi-subject handling for experimental and clinical trials, for example by implementing ready-to-use solutions for pseudo-/sham-neurofeedback.International journal of psychophysiology: official journal of the International Organization of Psychophysiology 09/2013; · 3.05 Impact Factor
Applied Psychophysiology and Biofeedback, Vol. 24, No. 4, 1999
Changes in EEG Power Spectra During Biofeedback
of Slow Cortical Potentials in Epilepsy
Boris Kotchoubey,1,3 Simone Busch,1 Ute Strehl,1 and Niels Birbaumer
The goal of the study was to explore parallel changes in EEG spectral frequencies during
biofeedback of slow cortical potentials (SCPs) in epilepsy patients. Thirty-four patients
with intractable focal epilepsy participated in 35 sessions of SCP self-regulation training.
The spectral analysis was carried out for the EEG recorded at the same electrode site
(Cz) that was used for SCP feedback. The most prominent effect was the increase in the
02 power (6.0—7.9 Hz) and the relative power decrement in all other frequency bands
(particularly S1, a2, and f}2) in transfer trials (i.e., where patients controlled their SCPs
without continuous feedback) compared with feedback trials. In the second half of the
training course (i.e., sessions 21—35) larger power values in the S, 9, and a bands were found
when patients were required to produce positive versus negative SCP shifts. Both across-
subject and across-session (within-subject) correlations between spectral EEG parameters,
on the one hand, and SCP data, on the other hand, were low and inconsistent, contrary to
high and stable correlations between different spectral variables. This fact, as well as the
lack of considerable task-dependent effects during the first part of training, indicates that
learned SCP shifts did not directly lead to the specific dynamics of the EEG power spectra.
Rather, these dynamics were related to nonspecific changes in patients' brain state.
Two biofeedback approaches have been developed for intractable epilepsy, one of
them based on rhythmical components of the EEG and the other on slow cortical poten-
tials (SCPs). The proponents of the former approach (Lubar, 1984; Lubar et al., 1981;
Sterman, 1986) train their patients to enhance the sensorimotor rhythm (SMR) and/or to
suppress slow rhythmic activities (e.g., in the theta band). In SCP training (Birbaumer
et al., 1991; Kotchoubey et al., 1996; Rockstroh et al., 1993), patients learn to control
slow (i.e., lasting from several hundreds of milliseconds to several seconds) shifts of their
cortical potentials; they are instructed either to increase or to decrease a negative SCP shift
1 Institute of Medical Psychology and Behavioral Neurobiology, University of Tubingen, Gartenstrasse 29, 72074
2 Department of General Psychology, University of Padova, Padova, Italy.
3To whom correspondence should be addressed.
KEY WORDS: EEG biofeedback; epilepsy; slow cortical potentials; spectral analysis.
1090-0586/99/1200-0213$16.00/0 © 1999 Plenum Publishing Corporation
appearing between the onset and the offset of a trial. In the former approach, it remains
unclear how far the SMR increase, rather than the decrease in slow rhythmic activity,
underlies the observed changes in seizure frequency. In the latter approach, recent data
(Kotchoubey et al., 1996,1999) indicate that it is the ability to suppress cortical negativity,
rather than the ability to differentiate between the two tasks, that is related to the following
No research has been conducted to date to compare directly the rate of clinical changes
(e.g., seizure reduction or neuropsychological improvement) attained with these two kinds
of biofeedback therapy, probably because of the high costs of such a study. A comparison of
published research is difficult due to numerous differences between studies with respect to
such factors as sample size, criteria for selection of patients, number, duration, and frequency
of training sessions, and duration of the follow-up period during which the clinical dynamics
were recorded. An inspection of the corresponding reports indicates that, on the average,
the success rates of SCP and SMR biofeedback appear to be similar, with an improvement
being observed in about two-thirds of the patients (Kotchoubey et al., 1997,1999; Lantz &
Sterman, 1988;Lubar, 1984; Rockstroh et al., 1993; Sterman, 1984).
These considerations may raise a question, whether the two approaches really affect
different cortical mechanisms, i.e., whether the self-controlled EEG components in these
two techniques really differ. It might be, for instance, that while learning to control the
SMR, patients, in fact, modify their SCPs, which can also be recorded over the sensorimotor
cortex and may correlate with rhythmical activity. Or during SCP self-regulation, patients
might suppress their slow spectral EEG components (delta and/or theta). The biofeedback
literature contains plenty of examples where subjects operantly control secondary functions,
x2, x3, x4, . . . , while learning to regulate x\ (e.g., Lacroix, 1986; Plotkin, 1976, 1977).
To disentangle these relationships, one has to record other EEG components in addition
to the target function. In SMR training, the use of AC EEG amplifiers with a relatively short
time constant makes simultaneous SCP recording impossible, since SCPs require either AC
amplifiers with a long time constant (at least 8 sec) or DC amplifiers (i.e., the time constant
is infinite large). In contrast, relatively fast EEG oscillations can be recorded with a long
or a short time constant, deliberately. Thus it is simpler to record EEG frequencies during
SCP training than vice versa. In the present study, we did not examine what effect training
in 6 reduction and/or SMR increase might have on SCP. Rather, we were recording EEG
rhythms in the range from 0.3 to 30 Hz in patients who learned to regulate their SCPs over
Birbaumer et al. (1990,1999) have shown that negative SCPs are generated in the upper
cortical layers after excitatory thalamocortical, intracortical, or interhemispheric inflow to
the dendritic trees. The slow synchronous depolarization underlying negative SCPs uses
cholinergic and glutamatergic synapses and mobilizes the cell in the deeper layers for firing.
The physiological basis of positive SCPs is less clear. In brain areas with extensive folding
and on the orbitofrontal, temporal, and occipital pole, the positivities are particularly difficult
to interpret due to variable dipole direction. However, if the dipoles under the electrode are
vertically oriented, in most cases a positivity indicates a reduction of excitability (mediated
by GABA-ergic systems) in the upper cortical layers, with a concomitant inhibition of
behavioral output. Based on these physiological considerations, one should expect increased
frequencies and desynchronization during learned cortical negativity and increased slow
synchronized rhythmic patterns during self-produced positivities.
Kotchoubey, Busch, Strehl, and Birbaumer
Thirty-four patients (19 females) with drug-resistant focal epilepsy (10 right temporal,
9 left temporal, 15 with unlocalized or multifocal seizures) participated in the study. None
of the patients had primarily generalized seizures with a sudden and complete loss of
consciousness (i.e., grand mal or petit mal); however, many seizures were secondarily
generalized, that is, a seizure began with focal symptoms (such as cramps in an arm or
leg) followed by generalized convulsions and loss of consciousness. The mean age of
the patients was 34.3 years (SD = 8.44 years), the mean seizure history was 23 years
(SD = 10.6 years). The mean seizure frequency prior to training was 3.46 per week (SD =
6.31 per week; median = 1.21 per week). Patients with psychogenic seizures, psychotic
symptoms, or progredient neurological diseases as well as those with an IQ below 80
(WAIS-R) were excluded. All patients were medicated with one or two antiepileptic drugs,
with the medication regime remaining constant for at least 5 months prior to the beginning
of treatment and up to 12 months after its termination.
The training course consisted of two phases, the first including 20 daily sessions and
the second 15 sessions. The two phases were separated by an 8-week interval. Each session
entailed 144 trials. In feedback trials, lasting for 8 sec, the actual SCP amplitude was
referred to the 1-sec pretrial baseline and presented in the form of a rocket-like object on a
computer screen. Simultaneously a letter A or B signaled the type of task (i.e., producing
positivity versus negativity) that had to be performed during that trial. These letters served
as discriminative stimuli. In transfer trials, only the letter A or B was presented for 8 sec,
and patients had to perform the corresponding task without feedback. The ratio between
positivity and negativity trials was 50/50 during the first phase and 70/30 during the second
phase (i.e., more trials with required positivity was presented), whereas the ratio between
feedback and transfer trials varied between 70/30 and 30/70 according to the patient's
performance. The first 30 trials in each session were always feedback trials.
The EEG was recorded at the vertex (Cz) referred to the two mastoid electrodes linked
over a 15-k£2 shunt, using a high-frequency cutoff filter of 40 Hz and a time constant of
10 sec. The vertical EOG was recorded using two electrodes above and below the left
eye. The data were analyzed in 9-sec epochs beginning 1 sec prior to the presentation of
the discriminative stimulus (and the rocket, in feedback trials) and lasting until all visual
stimuli disappeared. The EEG was on-line EOG corrected (for the correction algorithm see
Kotchoubey et al., 1996) and fed back as rocket movements, with each position of the rocket
on the screen corresponding to the averaged EEG amplitude over a 500-msec epoch. These
epochs slid with a 100-msec shift. Thus during an 8-sec trial, patients observed 80 subsequent
positions of the rocket, which created an illusion of a quasi-continuous movement.
EEG Spectra During Biofeedback in Epilepsy
Although the EEG was on-line artifact corrected, the noncorrected data were recorded.
It was then additionally off-line EOG corrected using the algorithm of Gratton et al. (1983).
Further, trials containing zero-line segments or amplitude values higher than 150 mV were
removed. Fast Fourier transformation was carried out over the 8-sec feedback interval using
a Hamming-type window, which resulted in squared amplitude values (i.e., /^V2). These
data were square-root transformed, thus resulting in simple amplitude values (i.e., /LtV)
and then averaged in the frequency domain according to Task (positivity versus negativity)
and Trial Type (feedback versus transfer).4 For SCP analysis, the EEG amplitudes were
averaged in the time domain as well. An example of the averages in both the time and the
frequency domains is shown in Fig. 1.
The resulting average EEG power spectra were subdivided into the S (0.3- to 3.9-Hz),
9 (4.0- to 7.9-Hz), a (8- to 13-Hz), and ft (13.1- to 30-Hz) bands. The maximal power and
its location on the frequency scale were calculated for each band. In addition, mean power
values were obtained for narrower frequency ranges: S1 (0.3-2.0 Hz), 82 (2.1-3.9 Hz), 01
(4.0-5.9 Hz), 02 (6.0-7.9 Hz), al (8.0-10.0 Hz), a2 (10.1-13.0 Hz), p1 (13.1-18.0 Hz),
and ft2 (18.1-30.0 Hz). All power values were taken in relation to the maximum power
value in the particular spectrum, as shown in Fig. 1. The parameters of these frequency
bands were subjected to a repeated-measures ANOVA, separately for the first and second
training phase, with factors Task (2 levels: positivity versus negativity), Trial Type (2 levels:
feedback versus transfer), and Session (20 or 15 levels, for the first and the second phase,
respectively). The degrees of freedom for the last factor were corrected for nonsphericity
using the Greenhouse-Geisser e.
A correlational analysis included both within-subject (across session) and between-
subject product-moment correlations. The former indicated whether parallel changes in
different EEG parameters occurred in the course of SCP training, and the latter indicated
whether patients with larger values in some EEG variable (e.g., negative SCP in transfer
trials) also tended to have larger (or smaller) values in some other variable (e.g., <S1 power
in feedback trials). The correlations were averaged using Fisher's logarithmic function.
The SCP data have been reported elsewhere (Kotchoubey et al., 1996, 1997, 1998)
and are summarized in Table I. In both phases of training, patients were able to produce the
required directional SCP shift (highly significant effects of the factor Task), and better perfor-
mance was achieved with than without feedback (highly significant Task x Trial Type inter-
actions). During the first training phase, the patients' SCP amplitudes became slightly more
positive across sessions, as indicated by a marginally significant effect of the factor Session.
4We tested different kinds of spectral data analysis in addition to that described in the main text. Thus a log
transformation of the spectral data was tested apart from the square-root transformation. Further, analyses of
variance were run for both raw data and several types of normalization. The tendencies found with different
techniques were very similar, but they were best pronounced when square-root transformed and normalized-to-
maximum data were entered into the analysis. For this reason, only these results are reported.
Kotchoubey, Busch, Strehl, and Birbaumer
EEC Spectra During Biofeedback in Epilepsy
Table I. Summary of the Significant ANOVA Effects on SCP During
the First and Second Training Phase
Task x Trial Type
Task x Trial Type
Note. *p < .10; *** p < .01; **** p < .001.
Fig. 1. The EEC averaged across the first training phase (20 sessions)
in the frequency domain (top) and time domain (bottom). Top: One can
see a clear difference between feedback trials and transfer trials. The
power values are presented in relation to the maximal power taken as 1.
Bottom: The effect of Task (positivity versus negativity) can be seen,
which was larger in feedback trials than in transfer trials (Task x Trial
Type interaction; see text). The amplitudes are presented as microvolts;
the negativity is up.