Comparison of Discrete-Trial-Based SMR and SCP Training and … · 2020. 5. 18. · Jacques Duysens...

Comparison of Discrete-Trial-Based SMR and SCPTraining and the Interrelationship Between SCP

and SMR Networks:Implications for Brain–Computer

Interfaces and Neurofeedback

Michiel Kleinnijenhuis, MScMartijn Arns, MSc

Desiree Spronk, MScRien Breteler, PhD

Jacques Duysens, MD, PhD

ABSTRACT. Background. Operant conditioning of one’s slow cortical potential (SCP) or sen-sorimotor rhythm (SMR) can be used to control epilepsy or to manipulate external devices, asapplied in BCI (Brain-Computer Interface). A commonly accepted view that both SCP andSMR are reflections of central arousal suggests a functional relationship between SCP andSMR networks.

Method. The operant conditioning of SCP or SMR was tested with a single electroencephalo-graphic (EEG) channel wireless biofeedback system. A series of trainings taught 19 participantsto control SCP or SMR over vertex during 20 neurofeedback sessions. Each session consisted of96 trials to decrease cortical arousal (SCP positivity=SMR enhancement) and 64 trials toincrease cortical arousal (SCP negativity=SMR suppression). In each trial, participants wererequired to exceed an individual threshold level of the feedback parameter relative to a 500-msec

Michiel Kleinnijenhuis is affiliated with the Department of Medical Physics and Biophysics, Radboud Univer-sity Nijmegen, Nijmegen, The Netherlands.

Martijn Arns is affiliated with Brainclinics Diagnostics B.V. and Brainclinics Treatment B.V., Nijmegen,The Netherlands.

Desiree Spronk is affiliated with Brainclinics Diagnostics B.V., Nijmegen, The Netherlands.Rien Breteler is affiliated with EEG Resource Institute B.V. and Department of Clinical Psychology,

Radboud University Nijmegen, Nijmegen, The Netherlands.Jacques Duysens is affiliated with the Department of Medical Physics and Biophysics, Radboud University

Nijmegen, Nijmegen, The Netherlands; SMK-RDE, Nijmegen, The Netherlands; and Research Center forMovement Control and Neuroplasticity, Department of Biomedical Kinesiology, Katholieke UniversiteitLeuven, Leuven, Belgium.

Address correspondence to: Martijn Arns, MSc, Brainclinics Diagnostics B.V., Bijleveldsingel 34, 6524 ADNijmegen, The Netherlands (E-mail: [email protected]).

This study was funded by the BIAL foundation (Grant Number 163=04) and supported by BrainclinicsDiagnostics B.V. The authors sincerely thank Noemı S�aanchez N�aacher for her extensive contribution to thetraining of the participants, Vivian Weerdesteyn for her counsel on the statistics, and Erica Schot-Heesen forher support in the initial phase of the study.

Journal of Neurotherapy, Vol. 11(4) 2007Available online at http://jn.haworthpress.com

# 2007 by The Haworth Press. All rights reserved.doi: 10.1080/10874200802162808 19

prefeedback baseline and to hold this level for 2 sec (SCP) or 0.5 sec (SMR) to obtainreinforcement.

Results. Ten of the 19 participants achieved control over their EEG. In the SCP-trained group, 4 of 9 participants increased the differentiation between their SCP responseson positivity-required versus negativity-required trials. SMR suppression and enhancementwas achieved by 3 and 4 of the 10 SMR-trained participants. The SMR-trained respondersdid not show differentiation in their SMR responses, but did show a differentiation in theirSCP response—while trained on SMR.

Conclusions. The results showed the proposed method was successful to teach control of SCPor SMR. Bidirectional control was very difficult to achieve with the present SMR training pro-cedure. SCP positivity and SMR enhancement were easier to learn. The results suggest thatSMR training modulates excitability thresholds in the striatal-thalamocortical motor loop,whereas changes in the loop’s excitability thresholds by SCP training do not affect the thalamicbursting that underlies the SMR.

KEYWORDS. Brain–computer interface, discrete training, epilepsy, neurofeedback,sensorimotor rhythm, slow cortical potential

The efficacy of neurofeedback in the controlof seizures in persons with uncontrolled epi-lepsy has been well documented over the pastdecades. The most promising approaches toseizure control include neurofeedback ofthe sensorimotor rhythm (SMR; Sterman,2000; Sterman & Friar, 1972; Tozzo, Elfner,& May, 1988) and the slow cortical potential(SCP; Kotchoubey et al., 2001; Rockstrohet al., 1993). SMR neurofeedback exercisesthe patient’s regulatory mechanisms in thethalamocortical loop that produces the SMR(Sterman, 2000). Similarly, the proposedmechanism of SCP neurofeedback seizurereduction is an improved regulation of corticalexcitability, because positive SCPs reflect adecrease in cortical excitability (Birbaumer,Elbert, Canavan & Rockstroh, 1990).

A second application of neurofeedbackmethodology is in the field of brain–computer interfacing (BCI). BCI is a tech-nique that uses signals extracted from thebrain to control devices (computers) withoutmuscular activity or overt speech. Such con-trol can be beneficial for patients with severemotor disabilities. For example, Birbaumerand colleagues (Birbaumer et al. 1999)developed a spelling device (Thought Trans-lation Device [TTD]) for patients sufferingfrom amyotrophic lateral sclerosis (ALS).The TTD uses self-regulation of the SCP asa means of binary decision strategy. By

either increasing or decreasing the SCP rela-tive to a pretrial baseline the patient canchoose between selections of letters of thealphabet until the desired letter is selected.Numerous variations on the TTD trainingprotocol have been tested (Birbaumer et al.,1999; Kubler, et al., 1999; Neumann et al.,2004). Several other studies showed that itis possible to operate a BCI with rhythmicactivity over the sensorimotor cortex as well.Typically, the mu or beta rhythms are used(Kubler et al., 2005; Pfurtscheller, Flotzinger,Pregenzer, Wolpaw, & McFarland, 1996;Wolpaw, Birbaumer, McFarland, Pfurtschel-ler, & Vaughan, 2002).

Because both the SCP and the SMR canbe regarded as measures of cortical arousal(Nagai, Goldstein, Critchley, & Fenwick,2004; Sterman, 1982), perhaps they areexpressions of the same underlying neuro-physiological network. The generatingmechanisms of both the SCP (Birbaumeret al., 1990) and SMR (Sterman, 2000) arerelatively well understood. Further, theinvasive animal studies with SCP (Birbaumeret al., 1990) and SMR (Bazhenov, Timofeev,Steriade, & Sejnovski, 1999) and humanfunctional magnetic resonance imaging stu-dies with SCP (Hinterberger et al., 2003)and SMR (Beauregard & Levasque, 2006)showed that they share cortical and subcorti-cal structures and rely heavily on regulation

20 JOURNAL OF NEUROTHERAPY

of activity in the striatal-thalamocorticalmotor loop (Birbaumer, 2006). Therefore, itcan be hypothesized that the self-regulationof one of the parameters affects the other(i.e., invoked SCP shifts induce SMR changesand=or SMR changes shift the excitabilitythresholds of the cortical pyramidal neuronsreflected by the SCP). This proposed func-tional relationship could be further exam-ined with coregistration of SCP and SMRwhile training one or the other in a neuro-feedback training protocol.

Although in recent years significant pro-gress has been made in the field of electroen-cephalographic- (EEG-) based BCI, thesystems commonly employed are traditionalfull-cap systems, which require significanteffort to apply and maintain. Indeed, someof the commonly used techniques (spatial fil-tering, independent component analysis)require the application of a multitude of elec-trodes. Because these techniques are cumber-some, the question arises whether a singlemeasurement electrode could be sufficientfor BCI systems, which therefore would beeasier to apply, and more practical in use.Recent advances in technology have resultedin the development of portable wireless EEGequipment that can measure a limited num-ber of EEG channels (Arns & van Dorsten,2005). During tele-neurofeedback (patientstrain at home, supervised by their therapistover the Internet), this equipment can be eas-ily applied by end-users with only minimaltraining in application of the electrodes anduse of the software (Breteler, de Ridder,Monsuwe, & Arns, 2006). A BCI shouldbe applicable to, for example, spinal cord-injured patients and therefore setup shouldbe as simple as possible, such that theend-user is able to apply and operate it withminimal assistance.

The regular approach in clinical neuro-feedback is to reward desired behavior in acontinuous task execution setting (e.g., in a5-min run, the advancement of a video isdependent on the feedback variable exceed-ing a certain threshold level). In BCI, how-ever, a discrete trial method is moreappropriate. The nature of the goal that ispursued imposes these different approaches.The aim for clinical neurofeedback is to

manipulate EEG power in a designatedbandwidth geared toward a clinical outcome,whereas in BCI a transient EEG response isappropriate for controlling a device. Despitethe broadly accepted convention to usecontinuous tasks in clinical neurofeedback,a discrete approach might be a more effec-tive learning method, because it is a key fea-ture of traditional operant conditioning(Ferster & Skinner, 1957; Sterman, M. B.,personal communication Oct. 10, 2005).

In this study, we describe a novelapproach for the feedback of physiologicalparameters using a discrete-trial basedapproach where participants are required toshow an increase or decrease in the feedbackparameter. This approach was developedfor the purpose of BCI as well as clinicaldiscrete (EEG) biofeedback training. In theexperiment presented here, we focus on thefeedback of the SMR and SCPs, as bothSMR and SCP control have been proven tobe advantageous for epilepsy patients andit was shown that SCP can be used forBCI purposes. This investigation exploreswhether the 12 to 15 Hz SMR might alsobe an EEG rhythm suitable for BCI. Thespecific questions we address in this studywith healthy participants are (a) are partic-ipants able to voluntarily increase ordecrease their SCP or SMR using discretefeedback, (b) do participants show a changein their trained SCP or SMR responses overthe course of training reflecting improvedskill acquisition, (c) how do the percentagesof successful responses of SCP-trained andSMR-trained participants compare over thetasks of increasing and decreasing the SCPor SMR level, and (d) is there evidence fora functional relationship in the networks thatgenerate the SCP and SMR (i.e., do changesoccur in one while participants are beingtrained on the other)?

METHOD

Participants

There were 19 participants in this study (9male, 10 female). All participants wererecruited according to the normative subject

Scientific Articles 21

profile of the Brain Resource InternationalDatabase (Gordon, 2003) in which exclusioncriteria include a history of neurological,psychiatric, or psychological disturbances;motoric, hearing, or vision impairments;and serious medical conditions. Every par-ticipant gave informed consent prior tothe study. The study was approved by theInstitutional Review Board (METC NoordHolland; number M05–010).

EEG Recordings

Data recording was achieved using per-sonal computers (Pentium IV processors)and BioExplorer software (V1.3). A customopen source design was programmed in theBioExplorer software environment. The useddesign and manual can be downloaded fromhttp://www.brainclinics.com.

The participants’ EEG was recorded fromCz referenced against linked mastoids[(A1þA2)=2] using the wireless two-channelbipolar Brainquiry PET SCP with activeelectrodes. The second channel of the devicewas used for recording vertical eye-movement activity (vEOG). The EOG elec-trodes were placed on the sagittal midline1 cm above and below the outer canthus ofthe right eye. The ground electrode was

placed on AF3. EEG and EOG wererecorded with a sampling frequency of200 Hz. Disposable pre-gelled Ag=AgþCl�

electrodes (Arbo electrodes H124SG, Tyco)were used for EEG recording. Ten-20 elec-trode paste was applied on the Cz recordingsite. All electrode sites were prepared withalcohol and Nuprep abrasive gel.

Procedure

The participants received 20 neurofeed-back training sessions in which they weretrained to self-regulate either their SCP(n ¼ 9) or SMR (n ¼ 10). The experimentspanned a total of 8 weeks with threetraining sessions per week and no more thanone session per day. The sessions were div-ided into four 7-min runs of 40 discrete trialseach. After two runs, participants wereencouraged and informed on their progressduring a 1- to 2-min pause. Trials were sepa-rated by variable intertrial intervals (1.5–3sec) with no task requirement and stimuli.Before the experiment, the participants wereinstructed on the functions of the various ele-ments of the feedback window (Figure 1)and the task requirement. The basic mechan-isms of neurofeedback were explained, butparticipants were not provided with a

FIGURE 1. The time course of a SCP in a D trial and the feedback window.


strategy to control the SCP or SMR. Instead,participants were asked to stay focused ontheir task and try to find a strategy for them-selves by closely observing the feedback andtrying to relate this to internal states. A hintwas provided stating that the feedback param-eter was related to arousal. The partic-ipants were further instructed to minimizemovement of body, head, hands, and eyes.This instruction was repeated if necessary.

During the trials, two conditions weremixed pseudorandomly (i.e., during everysession one of six trial sequences was used).In the sequences, the ‘‘down’’ (D) trials(which required lowering the level of corti-cal arousal: SCP positivity=SMR enhance-ment) and ‘‘up’’ (U) trials (heighteningarousal: SCP negativity=SMR suppression)were randomly mixed with the D conditioncomprising 60% of all trials and the U con-dition comprising 40% of all trials. Theasymmetric distribution of D trials and Utrials was introduced, because previouswork by Hinterberger and colleagues (Hin-terberger, T., personal communicationMay 5, 2005) indicated that increasing corti-cal arousal was easier to learn compared todecreasing cortical arousal. In addition,because reduction of epileptic seizures ismediated by lowering the cortical excit-ability level, emphasizing the D condition

was hypothesized to be more useful andsafer. Furthermore, intertrial intervals (1.5,2.0, 2.5, or 3.0 sec) were mixed randomlyin the sequences. The six trial sequenceswere randomized over sessions, and thisrandomized order was counterbalanced overparticipants by inverting the order in half ofthe participants in each group.

Participants were seated behind a 17-in.TFT monitor displaying the feedback win-dow wearing headphones. In a trial, the par-ticipant had to either increase or decreaseSCP or SMR (12–15 Hz). Figure 1 showsthe time course of a SCP in a D trial andthe feedback window as the participantwould see it at different time points through-out the trial. The start of a trial was indi-cated by a brief tone delivered through theheadphones. The trials were divided in apreparation phase and a feedback phase.In the preparation phase, the trial type wasindicated to the participant by the blinking(see� in Figure 1) of two blue rectangles (Ain Figure 1) in either the upper (U trial) orlower half (D trial) of the feedback window.In the feedback phase, the amount of timethat was left to perform the task wasindicated in the rectangles that were usedto indicate trial type in the preparationphase. The task requirement was to reachan individually determined threshold (C in

FIGURE 2. The detailed overview of the core of real-time data processing.


Figure 1) level of the feedback parameterrelative to the 500-msec prefeedback baselineand to hold this level for at least 2 sec forSCP (Hinterberger, T., personal communi-cation May 5, 2005) or 0.5 sec for SMR(Sterman, M. B., personal communicationOct. 10, 2005).

EEG Filter Settings and Feedback

SMR was filtered from the raw EEG usinga sixth order Butterworth IIR 12 to 15 Hzbandwidth filter. The feedback was basedon an average period of 500 msec. SCP wasfiltered using a 500-ms moving average FIRfilter. A detailed overview of the core ofreal-time data processing (noise check, arti-fact rejection and EOG correction) is pro-vided in Figure 2. SMR data processing didnot include EOG correction, but featuredboth the noise check and artefact detectionidentical to the SCP data processing. On apositive outcome of either the noise or arte-fact check the feedback bar would disappearfor the remainder of the trial and the trialwas excluded from analysis and successpercentages calculation.

Real-time feedback was provided by dis-playing a yellow bar (B in Figure 1) the heightof which was proportional to the level of thefeedback parameter (SCP or SMR amplitude)relative to the average 500-msec prefeedbackbaseline SCP=SMR level. The prefeedbackbaseline level was set at the vertical midline

of the window, and the visible range of thebar was set such that it was proportional(3x) to the—individually different—thresholdvalues, resulting in an identical visual displayfor all participants. Arousal increases (nega-tive SCPs and SMR decreases) were scaledto be displayed in the upper half of the win-dow and arousal decreases (positive SCPshifts and SMR increases) were displayed inthe lower half. In the intertrial-interval andthe preparation phase, no feedback wasprovided.

Reinforcing feedback in the form of two‘‘smiley’’ faces (location D in Figure 1) wasgiven when the participants reached theappropriate threshold. When the task wascompleted successfully, the participant hearda reinforcing sound, delivered through theheadphones. In addition, feedback on per-formance was provided continuously bydisplaying the percentage of successful trialsin a run for U and D trials separately (E inFigure 1). These percentages were updatedafter every trial.

Threshold Procedure

Each participant was provided with indi-vidually determined threshold settings forthe D and U conditions that would startthem at 33% successful trials (the chancelevel that was chosen according to thehypothesis that the number of rewards [onethird] would be sufficient to accomodate

FIGURE 3. Example of individual thresholds setting by linear regression.


learning but low enough to challenge partic-ipants). This standardized the procedureby establishing personal thresholds. Thesepersonal thresholds were determined basedon two pretraining sessions. In the first pre-training session, thresholds were set to a pre-fixed value (–10mV and 10mV for the SCPgroup; –4 mV and 4 mV for the SMR group).Every run in a session is associated with apercentage of successful trials, which isdependent on the threshold value. TheBioExplorer software package features a‘‘playback’’ function: the possibility to rerunthe session data, with (the same or) differentparameters for data processing. Employingthis function, the success percentages of arun in case different threshold settingswould have been used were simulated withfive different threshold settings (�4, 6, 8,10, 12 mV for the SCP group; �2, 3, 4, 5,6 mV for the SMR group), thus rerunningthe session five times for each of the pre-training sessions with different thresholdsettings. The resulting success percentageswere averaged over the runs of the first

pretraining session yielding an average suc-cess percentage for each of the threshold set-tings. Then, a linear regression was carriedout on the averages for D trials and U trialsseparately (see Figure 3). The thresholds forthe second pretraining session were taken asthe level of the feedback parameter withwhich 33% of trials would be successfulaccording to the linear regression of the firstsession.

The playback procedure was repeated forthe four runs of the second session, resultingin another four success percentages for eachthreshold setting, thus a total of eight successpercentages was obtained from the two pre-training sessions for each of the thresholdsettings. The success percentages of the pre-training sessions were averaged over theeight runs, arriving at an average pretrainingsuccess percentage for each threshold setting.Again, a linear regression was conducted,now on the average pretraining success per-centages. The thresholds for the training ses-sions were fixed at the level that predicted33% successful trials.

FIGURE 4. Grand-average SCP responses for D and U trials over the blocks for the SCP responders andnon-responders.


Physiological Responses

The physiological data (EEG and EOG)were processed using BioReview, Matlab,and SPSS software. Invalid trials wereexcluded from analysis. An entire run wasexcluded when more than 10 trials were inva-lid and an entire session was discarded ifmore than two runs were excluded. Theraw signals were filtered offline with thesame filter specifications as in the real-timeimplementation. The EEG was correctedfor eye-movement influences offline accord-ing to the procedure of Gratton, Coles, andDonchin (1983). The 500-msec prefeedbackbaseline was subtracted from every trial forSMR and corrected SCP. Grand-averageSCP and SMR amplitudes were obtainedby averaging D and U trials separately infive blocks of four sessions. To quantifyphysiological responses, the integral betweenthe grand averages of the D=U trials and thebaseline was calculated for the feedbackphase. Integrals PD and PU of the D and Utrials, respectively, are defined as

PD ¼Zb

a

GADdt PU ¼Zb

a

GAU dt

where GAD and GAU are the grand-averagesof the D and U trials, respectively. Theseperformance measures are calculated forboth SCP and SMR in the interval t ¼[a,b] ¼ [0,7] s. Repeated measures analysesof variance (ANOVAs) (2� 5� 2 [Group�Block�Trial Type]) were performed on theintegrals PD and PU for both the SCP andSMR response. The significance level wasset at a ¼ .05.

Performance

The participants’ performances were ana-lyzed with SPSS software. The percentagesof successful trials (for every run) were aver-aged in five blocks of four sessions. Theaveraged success percentages were enteredin a 2� 5� 2 (Group�Block�Trial Type)repeated measures ANOVA.

RESULTS

Physiological Responses

Analysis of the data of the individual partic-ipants indicated large variability betweenthe participants. The SCP-trained partic-ipants (n ¼ 9) can be divided in two rela-tively homogeneous groups based on theirability to control their SCP (the criterionwas a larger slope of a least-squares linearregression for PD as compared to PU overblocks; see also Figure 6 for the groupexample). First, there is a group of four partic-ipants who were able to increase the differ-entiation between their SCP responses onD and U trials over the blocks (responders;Figure 4).

The increase in differentiation (shadedarea) was primarily caused by SCP increasesin the D trials (4 participants). One partici-pant was able to also progressively decreasehis SCP in the U trials over the blocks.Second, 4 participants showed essentiallythe same response in all sessions, and 1participant even showed a decrease indifferentiation.

In the group of the SMR-trained partic-ipants (n ¼ 10), the number of responders(n ¼ 6) was higher than in the SCP-trainedgroup (Figure 5). Three of 10 participantswere able to progressively decrease theirSMR in the U trials (U trial responders),and 4 participants showed consistentincreases in SMR amplitude in the D trialsover the blocks (D trial responders). Oneparticipant was able to differentiate hisresponse between D and U trials.

Further, Figure 5 shows that, in the first 2sec of the feedback phase, the U trial respon-ders (left column) exhibit an increasinglysuppressed SMR response in the U trials(dashed trace) over blocks. However,roughly the same decreasing trend is seenin the D trials (solid trace), where thisresponse is inappropriate. Moreover, theSMR in the D trials is even more suppressedcompared to baseline than it is in the Utrials. An opposite—but similar—pattern isseen for the D trial responders (right col-umn). These participants also showed a sup-pression of SMR compared to baseline in the


first block in both trial types but graduallychanged their response in the feedback phaseto a SMR enhancement. Again, they showed

this in the D trials (the required response) aswell as in the U trials (the inappropriateresponse).

FIGURE 5. Grand-average SMR responses for D and U trials over the blocks for the SMR U trial respondersand D trial responders.

FIGURE 6. The group-average of PD and PU of the SCP (Diamonds, Solid Lines) and the SMR-trained group(Squares, Dashed Lines) for the SCP response (Left Panel) and the SMR response (Right Panel).


The ability to regulate the SCP and SMRacross session blocks was quantified bycalculating the integrals PD and PU betweenthe baseline and the grand-averages forthe feedback phase per block for both theSCP-trained group and the SMR-trainedgroup.

Figure 6 shows the group-averaged PD

and PU of the SCP (diamonds, solid lines)and the SMR-trained group (squares, dashedlines) for the SCP response (left panel) andthe SMR response (right panel). Of interest,the integrals of the SCP response show a dif-ferentiation for both the SCP-trained andSMR-trained participants, but there was alarge difference in the evolution of this dif-ferentiation. For the SCP-trained partic-ipants, it can be observed that the SCPdifferentiation (the difference between theintegrals of the D and U trials) increases overblocks (left panel; solid lines). In contrast,for the SMR-trained participants the SCPdifferentiation is constant over blocks (leftpanel; dashed lines are parallel) but shows

a nonspecific increasing trend (left panel;equal positive slope of the dashed lines).Unlike for the SCP-trained participants, thedifference between D and U trials does notgrow larger over blocks for the SMR-trainedparticipants. Thus, whereas the SCP-trainedparticipants learn to increasingly differentiatetheir SCP response on D and U trials overblocks, the SMR-trained participants onlyshow a positive shift in their SCP level, equalfor both D and U trails.

The integrals of the SMR response areundifferentiated (i.e., there is no net areabetween the grand averages of the D and Utrials) for both the SMR-trained group (rightpanel; dashed lines) and SCP-trained group(right panel; solid lines). Therefore, theSMR-trained participants did not learn todifferentiate their SMR response betweenD and U trials and the SCP-training didnot elicit consistent changes in the SMR inthe SCP-trained participants. In the SMR-trained group, both integrals PD and PU

show the same trend towards positivity, that

FIGURE 7. Success percentages for the mean across participants and best-performing participant overblocks.


is, from a suppression of SMR in the firstblock to the absence of a SMR response inthe last block.

The integrals of SCP response and SMRresponse were both entered in a 2� 5� 2(Group�Block�Trial Type) repeated mea-sures ANOVA. For the SCP response, asignificant effect of Trial Type was found,F(1, 17) ¼ 20.050, p < .001. This indicatesthat participants differentiated in theirresponse to U and D trails. It could not beshown that this differentiation was largerfor the SCP-trained group as compared tothe SMR-trained group (nonsignificant TrialType�Group interaction), F(1, 17) ¼ 2.984,p ¼ .102. Furthermore, a significant maineffect of block was found, F(4, 14) ¼ 4.666,p ¼ .013, and demonstrates a more positivegeneral SCP level relative to baseline overthe blocks, which was the same in the SCP-trained group and the SMR-trained group(nonsignificant Block�Group interaction),F(4, 14) ¼ 1.007, p ¼ .437, and occurred inboth D and U trials (nonsignificant TrialType�Block interaction), F(4, 14) ¼ 1.198,p ¼ .355. Post hoc contrasts indicated aquadratic increase over blocks, F(1, 17) ¼12.246, p ¼ .003, which showed that theincrease in SCP level becomes smaller quad-ratically over blocks. Consequently, theincrease in SCP positivity is much largerfrom Block 1 to Block 2 as compared tofrom Block 4 to Block 5. The three-wayinteraction Trial Type�Block�Group wasalso significant, F(4, 14) ¼ 3.708, p ¼ .029.Post hoc contrasts revealed that this interac-tion occurred in Block 2, where the SCP dif-ferentiation in the SCP-trained group isreduced compared to the other blocks andthe SCP differentiation in the SMR-trainedgroup is increased compared to the otherblocks. Therefore, the three-way interactiondid not provide compelling evidence for thehypothesis that the SCP differentiationshowed a larger increase over blocks in theSCP-trained participants compared to theSMR-trained participants. The ANOVA onthe SMR response yielded no significanteffects. This result indicates that, in thegroup average, the SMR response in thefeedback phase did not deviate from the pre-trial baseline in either the SMR-trained or

SCP-trained participants and that no pro-gress was made across blocks.

Performance

The performance of the participants wasevaluated separately. The task for the partic-ipants was to increase the percentage of suc-cessful trials. The criterion for a successfultrial was to exceed a personalized thresholdlevel of the trained parameter for 2 sec(SCP) or 0.5 sec (SMR). The success percent-age (the percentage of trials within a run inwhich the criterion was reached) was fedback continuously to the participants andupdated after every trial. The mean successpercentages of the five blocks are shown inFigure 7 for the D and U trials of the SCP-trained and SMR-trained groups separately.Although three of the four graphs show anincreasing trend over experimental blocks,the increases in success percentages weremodest.

Nevertheless, a number of participants didmanage to increase their success percentageconsiderably, in one or both trial types. Toillustrate this, the best performers in bothtrial types of both groups are included inthe plots of Figure 7. In the SCP-trainedgroup four participants increased by morethan 2.5% per block on D trials. On the Utrials, there were 3 SCP-trained participantswith an increase rate larger than 1.5% perblock with 1 participant achieving even anincrease in success percentage of more than8.5% per block. For the SMR participants,2 participants achieved an increase of morethan 2.5% per block on D trails and 2 par-ticipants increased more than 1.5% perblock on U trials.

To investigate if the participants were ableto increase their success percentage, a2� 5� 2 (Group�Block�Trial Type)repeated measures ANOVA was performedon the success percentages. In spite of a largenumber of participants showing the correctaverage physiological responses it cannotbe concluded from the results of this analysisthat the two groups of participants were ableto improve their performance, as judgedfrom the increase of their success percentage,


as the ANOVA did not demonstrate any sig-nificant effects.

DISCUSSION

A novel approach for single electrodeSMR and SCP training was used on the basisof discrete feedback trials, employing auto-mated online correction for EOG and amethod for the derivation of personalizedthreshold settings. This approach was testedin an experiment that aimed to teach 19 par-ticipants to self-regulate their SCP or SMRamplitude through neurofeedback. Morethan half of the participants (10 of 19) wereable to acquire some control over theirSCP or SMR response, with the resultsshowing distinctive interindividual differ-ences. The SCP responders (n ¼ 4) weremainly successful in SCP positivity trials,and SMR responders (n ¼ 6) were successfulin either SMR uptraining (n ¼ 4) or SMRdowntraining (n ¼ 3), and 1 participant wasable to master both conditions.

Consistent with previous studies, not allparticipants were able to gain control overtheir SCPs. In a study on healthy individualsby Rockstroh, Elbert, Birbaumer, andLutzenberger (1990), less than half (21 of45) of the participants mastered the skill ofSCP self-regulation. This is very comparableto the results we observed in this study (4 of 9participants). Hinterberger et al. (2004)reported successful regulation in 6, 4, and 2participants in three groups of 18 par-ticipants receiving visual, auditory, and com-bined feedback, respectively. In contrast toour study, these studies used full-cap EEGsystems for training their participants. Thevariability in the success of self-regulationof the SCP is also observed in patients withepilepsy (Rockstroh et al., 1993; Strehlet al., 2006) and patients with ALS (Neumann& Birbaumer, 2003).

A decrease in SCP differentiation of theSCP-trained responders was observed inthe last blocks of the experiment. Becausethe acquisition of the skill of self-regulationis motivationally dependent (Kleinman,1981) and oral reports from the participants

in our experiment indicated a decrease inmotivation throughout the second half ofthe experiment, we consider this to be a likelyexplanation for the performance decrease.For future research, a performance dependentmonetary reward (Elbert, Rockstroh, Lutzen-berger, & Birbaumer, 1980) or the coupling ofparticipants to introduce a competitive ele-ment in the experiment (Parente & Parente,2006) might boost the participants’ motiv-ation to maximize their performance.

One of the questions that may beanswered with our experiment is whetherSCP or SMR training has the most potentialfor BCI and epilepsy treatment. In total, 4 of9 SCP-trained participants were successful,whereas in the SMR-trained group, 6 of 10participants responded to the training pro-cedure. SMR suppression or enhancementcould be increased over the course of thetraining by, respectively, 3 and 4 SMR-trained participants. However, these par-ticipants showed the same response to boththe D and U trials. This inability to differen-tiate suggests that it is difficult to switchbetween SMR enhancement and suppressionon a very short time scale. We are unawareof previous research investigating SMRenhancement and suppression switching ona trial-by-trial basis. Sterman and Shouse(1980) were successful in both enhancingand suppressing SMR in single participantsbut not on a trial-by-trial basis. Theyswitched contingencies after 3 months ofrewarding SMR enhancement to rewardingSMR suppression and did not use discretetrials.

Of interest, in the SMR-trained group itwas observed that the SMR enhancementtrials were associated with larger SCP positiv-ity compared to the SMR suppression trials.In contrast, the SCP-trained participantsdid not show equivalent changes in SMRresponse (Figure 6). Our data therefore sup-port the notion that there is a relationbetween the SCP and the SMR. A possiblerelation between SCP and SMR was also pro-posed by Kotchoubey, Busch, Strehl, andBirbaumer (1999). They did not find consis-tent changes in EEG power spectra in theirSCP-trained epilepsy patients but attributedthe observed posttraining changes in delta,


theta, alpha, and beta frequency bands tononspecific changes in the participants’brain state. A similar nonspecific effect canbe concluded from our data, because weobserved a constant SCP differentiation inthe SMR-trained responders over the courseof the experiment, whereas the SMRresponse was either increasing (enhancementtrial responders) or decreasing (suppressiontrial responders). In addition, directionaldifferentiation between the enhancement-required and suppression-required trialswas observed for SCP but not for SMR.We suggest that SMR control sensitizes thestriatal-thalamocortical motor loop, whichcauses modulation of excitability thresholdsin this loop and thereby SCPs. On the otherhand, the modulation of the loop’s excit-ability thresholds by invoked SCPs doesnot appear to cause changes in the burstingbehavior of the thalamic nuclei responsiblefor the generation of the SMR. This couldimply that SMR training should be favoredover SCP training for the purpose of epilepsytreatment, as SMR training would improveboth the control over SMR mechanisms aswell as excitability thresholds. However, theSCP seems to be the more sensitive parameterbecause we observed SCP differentiationbetween arousal-increasing and arousal-decreasing trials, whereas we did not detectSMR differentiation.

In theory, it is possible that the SCP differen-tiation in the SMR-trained participants origi-nates from a source outside the brain (EMG=EOG contamination). However, EMG con-tamination is unlikely because it can beexpected that the SMR response would bemost highly affected by EMG artefacts.Because we observed a very similar SCPresponse (that was constant over blocks) inthe enhancement trial responders and the sup-pression trial responders of the SMR-trainedgroup while the SMR response of these groupsdeveloped in directions opposite to each other,we can exclude the involvement of EMG. It ispossible that the EOG correction procedure isinadequate. However, we find this explanationhighly unlikely, because the data from a thirdgroup that was trained (on Galvanic SkinResponse) using the exact same procedure(Spronk, Arns, Kleinnijenhuis, Breteler, &

van Luijtelaar, in preparation) did not showthis SCP differentiation. If the experimentaldesign would have caused consistent eye move-ments that affected the SCP at Cz and, more-over, would not have been corrected by theEOG correction procedure, we should haveobserved similar SCP differentiation in theSMR-trained participants and in this thirdgroup of Galvanic Skin Response–trainedparticipants.

Similar to the physiological responses, theability to improve the success percentage(above the 33% chance level) over sessionsvaried considerably across participants inour experiment. On average, the increasein the percentage of successful trials wasonly moderate, but individual participantsshowed considerable improvements in suc-cessful responses. Comparable results havebeen obtained by Neumann and Birbaumer(2003), who investigated correct responserates in a group of five patients with ALSwho were trained on their SCPs. Theirresults indicated significant improvement incorrect response rates in three patients, withonly one patient exhibiting a very high andstable degree of control.

In other BCI research employing mu andbeta rhythms, the correct response rate isgenerally higher than observed in this experi-ment. McFarland, Sarnacki, Vaughan, andWolpaw (2005) reported accuracies rangingfrom 80% to 100% (with a chance level of50%) after 10 sessions in 5 of 7 participants,whereas 2 participants were unable toachieve control over their mu or betarhythm. Pfurtscheller and colleagues used adifferent approach to BCI. They classifiedimaginary movements of the users by detect-ing event-related desynchronization orevent-related synchronization. In one oftheir initial BCI experiments, Pfurtscheller,Neuper, Flotzinger, and Pregenzer (1997)reported on 3 participants who showed largedifferences in EEG rhythms over sensori-motor cortex during left versus right handmovements that could be classified withan accuracy of 70 to 90% (chance level at50%). However, 7 participants failed toshow sufficient EEG differences.

The methodology that was used in theseexperiments was different from our own,


and this could explain the differencesbetween the correct response rates. First,the trained EEG frequencies were different.We sought to investigate the possibility ofoperating a BCI with the 12 to 15 Hzrhythm, whereas McFarland et al. (2005)and Pfurtscheller et al. (1997) used a strategyof finding the exact frequency of mu or betaoscillation in individual participants andcentering the bandpass filter around that fre-quency. Nevertheless, in experiments featur-ing this approach a number of participantsalso failed to respond to training, despitebeing trained on their exact mu or beta fre-quency. Moreover, a relatively large numberof participants in other experiments didrespond well to 12 to 15 Hz training withoutselection of the exact mu or beta frequency(Lantz & Sterman, 1988; Sterman, 1984).For example, Lubar and Lubar (1984)showed improvements in SMR acquisitionwith extended training in all six children thatwere included in their report. However, theydid not specify the inclusion criteria. Theselection of the mu or beta rhythm thereforeis neither a requirement nor a guarantee forachieving control.

Second, only a single electrode was used incontrast to other studies. The electrodelocation for feedback used in this experimentwas Cz. Whereas for SCP neurofeedbackgood results have been obtained withthe Cz placement (Birbaumer et al., 1999;Hinterberger et al., 2004; Rockstroh et al.,1993) in sensorimotor neurofeedback latera-lized electrodes over sensorimotor cortex areconsidered most suitable (Kropotov et al.,2005; Sterman & Friar, 1972). This isespecially important in the procedure ofPfurtscheller et al., because the specificinstruction of motor imagery of, forexample, hand movements calls for theplacement of electrodes over the hand areasof the motor cortex. Furthermore, Czelectrode placement can be problematic formeasurement of the SMR, because of cancel-lation of non-phase-locked synchronizedrhythmic activity from the left and right sen-sorimotor cortex on the sagittal midline(Storm van Leeuwen, & Versteeg, 1978;but see Egner and Gruzelier (2003) andPfurtscheller, Brunner, Schlogl, and Lopes

da Silva (2006) for examples of SMR=murhythm control over Cz). We selected theCz electrodes for both the SCP and SMR-trained groups for standardization purposesbetween the groups, but this could haveimpeded learning in the SMR-trainedparticipants.

Third, in principle it is possible that ourevaluation of success was not optimally suitedfor learning to increase the success per-centage. We adopted a procedure thatstarted the participants at 33% (chance level)correct responses in the first experimentalsession, whereas in the aforementioned stu-dies, chance level was at 50% correctresponses. Related to this, the experimentfeatured an approach of fixed thresholdlevels based on two pretraining sessions.With this method, the thresholds may havebeen set too high. Because the participantswere highly motivated at the start of theexperiment, it is conceivable that learninghas already taken place in the pretrainingsessions. This would be in accordance withthe results of Kotchoubey, Schleichert,Lutzenberger, and Birbaumer (1997), whofound that healthy participants can learn tocontrol their SCP in as few as two sessions.If the participants indeed learned to increaseor decrease their SCP and SMR levels in thepretraining sessions to a near maximal per-formance, this would certainly have led tothreshold settings that did not leave muchroom for improvement. For the SCP group,our data support this hypothesis, becausealready in the first block of the experimenta differentiation was found in the SCPresponse between D and U trials in someparticipants. Furthermore, the fixed thresh-old levels throughout the experiment couldhave diminished the performance. As wasalready argued by Skinner (1975), theshaping of the desired response is a veryimportant element in operant conditioning.In neurofeedback, when the rewardingand discriminative value of the feedbackdecreases (e.g., in the case where a clientresponds adequately more than 90% of thetime), the threshold for feedback is adaptedto a lower level of reward. The client isinformed about this change and sociallyrewarded by the therapist for the production


of the desired brain activity. Thus, a new,stricter criterion for successful responding isintroduced in the context of improved per-formance. The inability of some participantsto increase their success percentages and lowdegree of control over their physiologicalresponses could have been because of theabsence of a shaping procedure.

Fourth, the time window associated withthe response criterion may be an importantelement. Our approach required the partic-ipants to sustain their SCP or SMR ampli-tude above threshold level for a period oftime. The window for sustaining the thresh-old level was set at 0.5 sec for the SMR-trained participants. This criterion is verysuitable for training patients with epilepsyto produce SMR, as they are rewarded onlyfor extended burst of activity. For BCI, how-ever, sustained activity is not necessary butcould serve as a mechanism to reduce falsepositives. The SCP-trained participants hadto hold their SCP above threshold for 2 sec.In the approach of Rockstroh et al. (1990)the SCP response is averaged over the activephase of the trial (3–5 sec) and consideredcorrect if the average response is above athreshold level. Presumably, the correctresponse criterion of holding the SCP=SMRlevel above threshold for an extended periodis more difficult, as only minor shiftsin SCP=SMR amplitude can have a largeimpact on whether the criterion is met.This can possibly explain the difficulty ofimproving the correct response rate in ourexperiment. This can also explain why someparticipants seemed to improve their correctresponse rate to a lesser extent than theirphysiological performance.

Another consideration related to the trialsetup that could have influenced our resultsnegatively is the continuation of the trialafter the delivery of the reward. In operantconditioning, a rewarded response is fol-lowed by a burst of dominant frequencyactivity called a postreinforcement synchro-nization (PRS) that indicates a strengtheningof the associations between the responseand the reward (Buchwald, Horvath, Wyers,& Wakefield, 1964; Clemente, Sterman, &Wyrwicka, 1964; Pfurtscheller, 1992; Sterman,1996). The occurrence of a PRS is therefore

an important element of learning in neuro-feedback. In our experiment, the PRS mayhave been hampered by possible effects ofnot ending a trial immediately after thereward has been delivered. Instead, onreaching the reward criterion the feedbackon the physiological parameter continuedfor the remainder of the trial. The ongoingfeedback after the reward delivery couldhave encouraged the participants to uncon-sciously stay focused on the feedback andthereby hampered the PRS necessary forconsolidation of the response–reward associ-ation. Furthermore, the continuation of thetrial might have confused the participantsin thinking that they had not reached thetrue goal yet.

CONCLUSIONS

Using this approach, 10 of 19 participantstrained on their SCPs or SMR were able toachieve control over their brain activity.The group of SCP-trained participantsshowed more improvement in the positivity-required condition as compared to thenegativity-required condition. The group ofSMR-trained participants performed betteron SMR enhancement as compared toSMR suppression. For SMR, most partic-ipants did not achieve bidirectional control.The answer to the question which EEGcharacteristic, SCP or SMR, could becontrolled best is inconclusive. A larger per-centage of participants were able to gain uni-directional control over SMR, but theresponding SCP-trained participants showedbidirectional differentiation. Our findingsindicate that for BCI research future studiesshould acknowledge interindividual differ-ences. For example, by finding whethersomeone is better in up or downtrainingSCP or SMR, one could focus the trainingonly on that aspect rather than focusing onbidirectional control. Our results also stressthe need for personalizing training proce-dures for clients in order to achieve reliableresponses with BCIs. Of interest, changes inthe passively recorded parameter occurredin the SMR but not in the SCP-trainedgroup, that is, the SMR-trained group


showed SCP differentiation, but the SCP-trained group did not show equivalent effectsin SMR response. This suggests that excit-ability thresholds in the striatal-thalamocor-tical motor loop are modulated by SMRtraining, whereas shifts in the excitabilitythresholds in this loop induced by SCP train-ing do not affect the thalamic bursting thatunderlies the SMR.

REFERENCES

Arns, M. W., & van Dorsten, J. W. A. (2005). System isfor recording electrical signals occurring in living bodyand has at least two electrodes for obtaining firstanalogue signal corresponding with electrical signaloccurring in living body (Patent NL1024860C).

Bazhenov, M., Timofeev, I., Steriade, M., & Sejnovski,T. J. (1999). Self-sustained rhythmic activity in thethalamic reticular nucleus mediated by depolarizingGABA-A receptor potentials. Nature Neuroscience,2(2), 168–174.

Beauregard, M., & Levasque, J. (2006). Functional mag-netic resonance imaging investigation of the effects ofneurofeedback training on the neural bases of selec-tive attention and response inhibition in children withattention-deficit=hyperactivity disorder. Applied Psy-chophysiology and Biofeedback, 31(1), 3–20.

Birbaumer, N. (2006). Breaking the silence: Brain–computer interfaces (BCI) for communication andmotor control. Psychophysiology, 43, 517–532.

Birbaumer, N., Elbert, T., Canavan, A. G. M., & Rock-stroh, B. (1990). Slow potentials of the cerebral cortexand behavior. Physiological Reviews, 70(1), 3–41.

Birbaumer, N., Flor, H., Ghanayim, N., Hinterberger,T., Iverson, I., Taub, E., et al. (1999). A spellingdevice for the paralyzed. Nature, 398, 297–298.

Breteler, R., de Ridder, S., Monsuwe, A., & Arns, M.(2006). The usability of tele-neurofeedback: Proce-dures and results. Presented at the 2006 Society forApplied Neuroscience Inaugural Abstracts, papersession, Swansea, Sept. 2006.

Buchwald, N. A., Horvath, F. E., Wyers, E. J., &Wakefield, C. (1964). Electroencephalogramrhythms correlated with milk reinforcements incats. Nature, 201, 830–831.

Clemente, C. D., Sterman, M. B., & Wyrwicka, W.(1964). Post-reinforcement EEG synchronizationduring alimentary behaviour. Electroencephalogra-phy and Clinical Neurophysiology, 16, 355–365.

Egner, T., & Gruzelier, J. H. (2003). EEG biofeedback oflow beta band components: frequency-specific effectson variables of attention and event-related brainpotentials. Clinical Neurophysiology, 115, 131–139.

Elbert, T., Rockstroh, B., Lutzenberger, W., &Birbaumer, N. (1980). Biofeedback of slow corticalpotentials. Electroencephalography and ClinicalNeurophysiology, 48, 293–301.

Ferster, C. B., & Skinner, B. F. (1957). Schedules ofreinforcement. New York: Appleton-Century-Crofts.

Gordon, E. (2003). Integrative neuroscience andpsychiatry. Neuropsychopharmacology, 28, 2–8.

Gratton, G., Coles, M. G. H., & Donchin, E. (1983). Anew method for off-line removal of ocular artifact.Electroencephalography and Clinical Neurophysiol-ogy, 55, 468–484.

Hinterberger, T., Neumann, N., Pham, M., Kubler,A., Grether, A., Hofmayer, N., et al. (2004). A mul-timodal brain-based feedback and communicationsystem. Experimental Brain Research. Experimen-telle Hirnforschung. Experimentation Cerebrale,154(4), 521–526.

Hinterberger, T., Veit, R., Strehl, U., Trevorrow, T.,Erb, M., Kotchoubey, B., et al. (2003). Brain areasactivated in fMRI during self-regulation of slowcortical potentials (SCPs). Experimental BrainResearch. Experimentelle Hirnforschung. Exper-imentation Cerebrale, 152, 113–122.

Kleinman, K. M. (1981). The role of reinforcementand motivation in biofeedback performance. Physi-ology & Behavior, 26(5), 921–925.

Kotchoubey, B., Busch, S., Strehl, U., & Birbaumer,N. (1999). Changes in EEG power spectra duringbiofeedback of slow cortical potentials in epilepsy.Applied Psychophysiology and Biofeedback, 24(4),213–233.

Kotchoubey, B., Schleichert, H., Lutzenberger, W., &Birbaumer, N. (1997). A new method for self-regu-lation of slow cortical potentials in a timed para-digm. Applied Psychophysiology and Biofeedback,22(2), 77–93.

Kotchoubey, B., Strehl, U., Uhlmann, C., Holzapfel,S., Konig, M., Froscher, W., et al. (2001). Modifi-cation of slow cortical potentials in patients withrefractory epilepsy: A controlled outcome study.Epilepsia, 42(3), 406–416.

Kropotov, J. D., Grin-Yatsenko, V. A., Ponomarev,V. A., Chutko, L. S., Yakovenko, E. A., &Nikishena, I. S. (2005). ERPs correlates of EEGrelative beta training in ADHD children.International Journal of Psychophysiology, 55,23–34.

Kubler, A., Kotchoubey, B., Hinterberger, T.,Ghanayim, N., Perelmouter, J., Schauer, M., et al.(1999). The thought translation device: a neurophy-siological approach to communication in totalmotor paralysis. Experimental Brain Research.Experimentelle Hirnforschung. ExperimentationCerebrale, 124, 223–232.


Kubler, A., Nijboer, F., Mellinger, J., Vaughan, T. M.,Pawelzik, H., Schalk, G., et al. (2005). Patients withALS can use sensorimotor rhythms to operate abrain-computer interface. Neurology, 64, 1775–1777.

Lantz, D. L., & Sterman, M. B. (1988). Neuropsycho-logical assessment of subjects with uncontrolled epi-lepsy: Effects of EEG feedback training. Epilepsia,29, 163–171.

Lubar, J. O., & Lubar, J. F. (1984). Electroencephalo-graphic biofeedback of SMR and beta for treat-ment of attention deficit disorders in a clinicalsetting. Biofeedback and Self-Regulation, 9(1), 1–23.

McFarland, D. J., Sarnacki, W. A., Vaughan, T. M., &Wolpaw, J. R. (2005). Brain-computer interface(BCI) operation: Signal and noise during early train-ing sessions. Clinical Neurophysiology, 116, 56–62.

Nagai, Y., Goldstein, L. H., Critchley, H. D., &Fenwick, P. B. C. (2004). Influence of sympatheticautonomic arousal on cortical arousal: Implicationsfor a therapeutic behavioral intervention in epi-lepsy. Epilepsy Research, 58(3), 185–193.

Neumann, N., & Birbaumer, N. (2003). Predictors ofsuccessful self control during brain-computer com-munication. Journal of Neurology, Neurosurgery,and Psychiatry, 74(8), 1117–1121.

Neumann, N., Hinterberger, T., Kaiser, J., Leins, U.,Birbaumer, N., & Kubler, A. (2004). Automaticprocessing of self-regulation of slow cortical poten-tials: Evidence from brain-computer communi-cation in paralyzed patients. ClinicalNeurophysiology, 115, 628–635.

Parente, A., & Parente, R. (2006). Mind-operateddevices: mental control of a computer using bio-feedback. Cyberpsychology & Behavior, 9(1), 1–4.

Pfurtscheller, G. (1992). Event-related synchronization(ERS): An electrophysiological correlate of corticalareas at rest. Electroencephalography and ClinicalNeurophysiology, 83, 62–69.

Pfurtscheller, G., Brunner, C., Schlogl, A., & Lopes daSilva, F. H. (2006). Mu rhythm (de)synchronizationand EEG single-trial classification of differentmotor imagery tasks. NeuroImage, 31, 153–159.

Pfurtscheller, G., Flotzinger, D., Pregenzer, W.,Wolpaw, J. R., & McFarland, D. J. (1996). EEG-based brain–computer interface (BCI): search foroptimal electrode positions and frequency compo-nents. Medical Progress Through Technology, 21,111–121.

Pfurtscheller, G., Neuper, Ch., Flotzinger, D., &Pregenzer, M. (1997). EEG-based discriminationbetween imagination of right and left hand move-ment. Electroencephalography and Clinical Neuro-physiology, 103, 642–651.

Rockstroh, B., Elbert, T., Birbaumer, N., &Lutzenberger, W. (1990). Biofeedback-producedhemispheric asymmetry of slow cortical potentials

and its behavioral effects. International Journal ofPsychophysiology, 9, 151–165.

Rockstroh, B., Elbert, T., Birbaumer, N., Wolf, P.,Duchting-Roth, A., Reker, M., et al. (1993).Cortical self-regulation in patients with epilepsies.Epilepsy Research, 14(1), 63–72.

Skinner, B. F. (1975). The shaping of phylogenicbehaviour. Journal of the Experimental Analysis ofBehavior, 24(1), 117–120.

Spronk, D. B., Arns, M. W., Kleinnijenhuis, M.,Breteler, M. H. M., & van Luijtelaar, G. (in prep-aration). Manuscript in preparation.

Sterman, M. B. (1982). EEG biofeedback in the treat-ment of epilepsy: An overview circa 1980. In L. White& B. Tursky (Eds.), Clinical biofeedback: Efficacy andmechanisms (pp. 311–330). New York: Guilford.

Sterman, M. B. (1984). The role of the sensorimotorEEG activity in the etiology and treatment of gen-eralized motor seizures. In T. Elbert, B. Rockstroh,W. Lutzenberger, & N. Birbaumer (Eds.), Self-regulation of the brain and behavior (pp. 95–106).Berlin: Springer.

Sterman, M. B. (1996). Physiological origins andfunctional correlates of EEG rhythmic activities:Implications for self-regulation. Biofeedback andSelf-Regulation, 21(1), 3–33.

Sterman, M. B. (2000). Basic concepts and clinicalfindings in the treatment of seizure disorders withEEG operant conditioning. Clinical EEG (Electro-encephalography), 31(1), 45–55.

Sterman, M. B., & Friar, L. (1972). Suppression of sei-zures in an epileptic following sensorimotor EEGfeedback training. Electroencephalography andClinical Neurophysiology, 1, 57–86.

Sterman, M. B., & Shouse, M. N. (1980). Quantitativeanalysis of training, sleep EEG, and clinicalresponse to EEG operant conditioning in epileptics.Electroencephalography Clinical Neurophysiology,49(5), 558–576.

Storm van Leeuwen, W., Wieneke, G., Spoelstra, P., &Versteeg, H. (1978). Lack of bilateral coherence ofmu rhythm. Electroencephalography and ClinicalNeurophysiology, 44, 140–146.

Strehl, U., Trevorrow, T., Veit, R., Hinterberger, T.,Kotchoubey, B., Erb, M., et al. (2006). Deacti-vation of brain areas during self-regulation of slowcortical potentials in seizure patients. Applied Psy-chophysiology and Biofeedback, 31(1), 85–94.

Tozzo, C. A., Elfner, L. F., & May, J. G. (1988). EEGBiofeedback and relaxation training in the controlof epileptic seizures. International Journal ofPsychophysiology, 6, 185–194.

Wolpaw, J. R., Birbaumer, N., McFarland, D. J.,Pfurtscheller, G., & Vaughan, T. M. (2002).Brain–computer interfaces for communication andcontrol. Clinical Neurophysiology, 113, 767–791.


Comparison of Discrete-Trial-Based SMR and SCP Training and … · 2020. 5. 18. · Jacques Duysens...

Documents

Transcript of Comparison of Discrete-Trial-Based SMR and SCP Training and … · 2020. 5. 18. · Jacques Duysens...