a novel self-guided approach to alpha activity training alpha.pdf · a novel self-guided approach...

International Journal of Psychophysiology 83 (2012) 282–294

Contents lists available at SciVerse ScienceDirect

International Journal of Psychophysiology

j ourna l homepage: www.e lsev ie r .com/ locate / i jpsycho

A novel self-guided approach to alpha activity training

Geert J.M. van Boxtel a,b,⁎, Ad J.M. Denissen c, Mark Jäger c, David Vernon d, Marian K.J. Dekker a,c,Vojkan Mihajlović c, Margriet M. Sitskoorn a

a Department of Psychology, Tilburg University, Tilburg, The Netherlandsb Department of Electrical Engineering, Eindhoven University of Technology, Eindhoven, The Netherlandsc Brain, Body, and Behavior Group, Philips Research, Eindhoven, The Netherlandsd Canterbury Christ Church University, Canterbury, Kent, UK

⁎ Corresponding author at: Department of Psycholog90153, 5000 LE Tilburg, The Netherlands. Tel.: +31 13 4

E-mail address: [email protected] (G.J.M. van Box

0167-8760/$ – see front matter © 2011 Elsevier B.V. Alldoi:10.1016/j.ijpsycho.2011.11.004

a b s t r a c t
a r t i c l e i n f o
Article history:Received 13 April 2011Received in revised form 2 November 2011Accepted 9 November 2011Available online 26 November 2011

Keywords:Alpha activityInstrumental conditioningNeurofeedbackRandom beta trainingRelaxation

Fifty healthy participants took part in a double-blind placebo-controlled study in which they were eithergiven auditory alpha activity (8–12 Hz) training (N=18), random beta training (N=12), or no training atall (N=20). A novel wireless electrode system was used for training without instructions, involving water-based electrodes mounted in an audio headset. Training was applied approximately at central electrodes.Post-training measurement using a conventional full-cap EEG system revealed a 10% increase in alpha activityat posterior sites compared to pre-training levels, when using the conventional index of alpha activity and anon-linear regression fit intended to model individual alpha frequency. This statistically significant increasewas present only in the group that received the alpha training, and remained evident at a 3 monthfollow-up session, especially under eyes open conditions where an additional 10% increase was found. Inan exit interview, approximately twice as many participants in the alpha training group (53%) mentionedthat the training was relaxing, compared to those in either the beta (20%) or no training (21%) controlgroups. Behavioural measures of stress and relaxation were indicative of effects of alpha activity trainingbut failed to reach statistical significance. These results are discussed in terms of a lack of statisticalpower. Overall, results suggest that self-guided alpha activity training using this novel system is feasibleand represents a step forward in the ease of instrumental conditioning of brain rhythms.

© 2011 Elsevier B.V. All rights reserved.

1. Introduction

Keeping upwith the pace ofmodern society seems to be increasinglydifficult for many people. It has been estimated that work-relatedstress negatively affects at least 40 million workers in 15 countriesof the European Union, costing 20 billion Euros annually (EuropeanCommission, Employment and Social Affairs, 1999, cited in Mentalhealth policies and programmes in the workplace, WHO, 2005).Strategies for copingwith such stress can be diverse, for instance, sittingin a comfortable chair with eyes closed and listening tomusic, practisingyoga, mindfulness training or other meditative techniques, or un-dergoing a physical massage. Here we initiated a program whichexplored whether instrumental conditioning of alpha activity (alphaneurofeedback) could help people to relax in an enjoyable way at lowcost.

y, Tilburg University, P.O. Box662492; fax: +31 13 4662370.tel).

rights reserved.

1.1. Alpha activity1

Although the functional significance of brain rhythms in the alpharange is not exactly known, an increase in alpha activity has oftenbeen related to a decrease in cortical activity. Right from the discoveryof the alpha rhythm it was clear that it was greater with eyes closed,and blocked when the eyes were opened. Presumably, the absence ofinput to the visual cortex led the brain to exhibit an “idling” rhythm.More recent theories (e.g., Klimesch et al., 2007) focus on the conceptof cortical inhibition. This perhaps gives alpha activity a more activeconnotation, but still assumes a relationship between cortical inactivityand increased alpha activity, which is further supported by research si-multaneously recording EEG and fMRI (Goldman et al., 2002).

In the light of these theories it comes as no surprise that alpha activityin humans has often been related to relaxation at the behavioural level.Berger (1929) already observed that alpha activity could be blocked by

1 In line with Shaw (2003), we use the term ‘alpha activity’ to denote brain rhythmsin the alpha frequency range, including the traditional posterior alpha rhythm, the cen-tral mu rhythm, and the temporal tau rhythm.

http://dx.doi.org/10.1016/j.ijpsycho.2011.11.004

mailto:[email protected]


http://www.sciencedirect.com/science/journal/01678760

283G.J.M. van Boxtel et al. / International Journal of Psychophysiology 83 (2012) 282–294

attention or mental effort — the opposite of relaxation. Lindsley (1960)related brain rhythms to arousal states, from coma to overactivity, andassumed that alpha activity corresponded to a state of “relaxed wakeful-ness”, in which attention wanders, free associative thought is favoured,but which is still characterised by behavioural efficiency. Increasingalpha activity by instrumental conditioning has been reported toincrease subjective feelings of relaxation and well-being (Hardt andKamiya, 1978; Nowlis and Kamiya, 1970; Watson et al., 1978; Rice etal., 1993). One of the first systematic studies in this field is that ofNowlis and Kamiya (1970). They asked participants to voluntarilyproduce alternate intervals with either as much or as little alpha aspossible. Alpha levels differed between those trials, as did the verballabelling that the participants attached to the trials. For high alphalevels, verbal reports included relaxation, awareness of breathing,letting go, not focusing, floating, feelings of pleasure, security, andwarmth. For low alpha levels, reports included being alert and vigilant,tension, agitation, and visual attentiveness. A similar pattern wasreported by Hardt and Kamiya (1978) who found that enhancingalpha activity was associated with a reduction in anxiety levels,but only for those with initially high levels. Since that time manystudies and reviews have been published on the relationship be-tween alpha activity and topics like relaxation (e.g., Gruzelier,2002), meditation (e.g., Cahn and Polich, 2006), and hypnosis (e.g.,Perlini and Spanos, 1991; Gruzelier, 2002). Unfortunately, much ofthe research focusing on manipulating alpha to enhance mood hassuffered from methodological difficulties such as the absence of ad-equate control conditions (e.g., see the review of Vernon et al., 2009,for a discussion).

1.2. Double-blind, placebo-controlled study design

In recent years there has been much improvement in the easewith which brain potentials can be recorded, and in the speed ofcomputers that make it possible to calculate on-line feedback basedon these potentials. Thus, more elaborate studies have become possible,and adequate control conditions have become much more feasible,as well as the use of a double-blind, placebo controlled design, inwhich both the participants and the experimenters are uninformedabout the experimental conditions. An important question in thedesign of this type of study is what constitutes an adequate controlfor alpha activity training. Many studies have used ‘treatment-as-usual’ in which the control group does not come to the laboratoryfor training, or a ‘no feedback’ condition, in which the participantsjust sit and relax but receive no training. Such control conditionsare often very dissimilar to the training condition, and thereforenot ideal for a vigorous test. Alternatively, ‘mock’ training hasbeen used (e.g., Egner et al., 2002), in which the participants receivefeedback of a ‘typical’ training session or the feedback of anotherparticipant, which is not temporally linked to the brain potentialsrecorded in the actual session. Such a control condition is clearlyan improvement over no feedback conditions, because the controlusually resembles the training much more. An interesting designwas recently proposed by Hoedlmoser et al. (2008). They wantedto train the sensorimotor rhythm (SMR; 12–15 Hz), and used train-ing of random frequencies (bins of 3 Hz) between 7 and 20 Hz, ex-cluding the target frequency range of 12–15 Hz. Hence, the realtraining group always received feedback based on SMR, but thecontrol group received feedback based on one of the other bins,and also a different bin in each training session. Thus, their controlgroup received real feedback training, as did the experimentalgroup, just not in the frequency band of interest. We consider thisdesign to be an improvement over the mock feedback set-up becausethe control group receives real training in a situation that is otherwiseidentical to the experimental group. Hence, we incorporated this intoour design in which alpha training could be compared to random betatraining.

1.3. Alpha activity training system

An important motivation for our study concerns the valorisation ofscientific knowledge.Wewanted to investigate whether instrumentalconditioning of alpha activity could be applied in a consumer productaimed at normal, healthy, everyday users. For instance, we pictured aperson commuting on the train, wearing a device similar to an MP3player, with earplugs or a headset, listening to his or her favouritemusic while at the same time receiving relaxation therapy. This ambi-tion resulted in various important design choices. First, we used audi-tory feedback for training, because this would resemble an MP3-likedevice. Specifically, because we wanted people to listen to theirown favourite music, this also meant that the feedback needed to bea manipulation of music characteristics. The three aspects of soundwaves are amplitude (perceived as loudness), frequency (perceivedas pitch), and complexity (perceived as timbre). Pilot work showedthat a simple high-pass filter on the sound greatly affected musicquality, making the music sound very distant and thin. This turnedout to be the basis of a very intuitive form of feedback, especially be-cause people know their own favourite music very well. In fact, thefeedback was so intuitive that no instructions were needed to begiven to the participants for training to occur. Using auditory feed-back is also the reason for incorporating a third condition in our de-sign — a group of participants who did nothing else other thanlistening to their favourite music. In this way, we hoped to assessany differences of alpha activity training not only in comparison to awell-controlled beta training condition, but also relative to listeningto music only. Listening to music is perceived by many people asrelaxing, and for a new device to be successful it should add valueto that already relaxing experience.

Secondly, if such a device is to be used outside of the laboratory inreal-life situations, the measurement of alpha activity should begreatly simplified. It goes without saying that using electrodes thatrequire conductive gel or paste are out of the question. Innovationis required to reach our ambition of valorisation. Our ultimate aim isto use dry electrodes connected (with or without wires) to a record-ing system built into an MP3-like device. However, as an intermediatestep, here we used water-based electrodes mounted into the connec-tion band of a consumer-style auditory headset, connected to a porta-ble amplifier that transmitted the signals via Bluetooth to a computerthat controlled the feedback (in the future the computer could bereplaced by an MP3-like device or smartphone). Using this set-up,the participants could apply the headset by themselves, and the com-puter software would give them feedback about signal quality. For thepurpose of this study, an experimenter applied some control signalssuch as ECG and EOG that could be disposed of in real life situations.Nevertheless, such a set-up enabled us to investigate whether such awater-based electrode system, which is likely to result in signals oflower quality then with a conventional system, would be sufficientlysensitive to enable adequate alpha activity training which in turnwould increase relaxation. Our system already represents a big im-provement over conventional neurofeedback equipment, which ismostly gel-based and can only be used by trained personnel.

The brain rhythm that can be measured over the central scalp iscalled the mu or sensorimotor rhythm (SMR), and is often of a slightlyhigher frequency than the alpha rhythm. Functionally, it is thought tobe similar to the alpha rhythm in that it blocks in response to motoraction, just as the alpha rhythm blocks in response to visual input(Pfurtscheller and Aranibar, 1979). Thus, we wanted to know wheth-er training at central sites is related to subjectively experienced relax-ation, just as training at more posterior sites is thought to be, and alsowhether central training generalises to other brain areas. There issome, but not much, evidence that training of brain rhythms maygeneralise over the scalp (e.g., Egner et al., 2004).

To summarise, we investigated the effect of alpha activity trainingat central locations on alpha activity itself, measures of heart rate

284 G.J.M. van Boxtel et al. / International Journal of Psychophysiology 83 (2012) 282–294

variability, and subjectively experienced relaxation.Weused an innova-tive set-up with water-based electrodes in young healthy participants,and relatively short training periods in 15 sessions on consecutivedays. Using short training periods in young participants (mostly collegestudents) who might be considered already very relaxed perhapsmeans that the chances of finding strong effects are not very high. Onthe other hand,we reasoned that such a situation constitutes a powerfultest, and that finding such effects would be quite meaningful andimportant.

2. Methods

2.1. Participants

Participants were recruited by means of a website that explainedthe procedures involved in the research in great detail. A total num-ber of 171 persons (48 male, 123 female) indicated on the websitethat they wanted to participate in the research. Their age rangedfrom17 to 27 years (mean20.5 years). One personwas rejected becauseof hearing problems; 109 persons either did not follow up on ourrequest, turned out to be unavailable at the time of the research,or decided to cancel their participation. The remaining 62 providedwritten informed consent and were randomly assigned to one ofthree groups (see Design and procedure): alpha training (A), randombeta training (B), or control (C, music only). Twelve persons had to beexcluded from the final analyses, either because they failed to completethe training sessions (N=5), or because of technical problems duringthe sessions (N=7). The final group consisted of 50 participants(36 female, 14 male), mostly psychology students, all right handedand in good health with normal or corrected-to-normal vision andhearing. They were distributed over the three groups as follows: A(alpha training): N=18 (12 female), mean age 20.68 years (18–25,S.D. 1.79 years); B (random beta training): N=12 (9 female), meanage 20.59 years (19–23, S.D. 1.50 years); C (control, music only):N=20 (15 female), mean age 21.02 years (18–25, S.D. 2.06 years).The three groups did not differ statistically in mean age (F(2,47)=0.26, p=.775). All participants were paid about € 300 plus travelexpenses for their participation.

Although the participants were informed about the overall purposeof the study, details such as the different conditions were not revealed.The participants as well as the experimenters were blind to the experi-mental condition that a particular participantwas randomly assigned to(double-blind design).

2.2. Design and procedure

An overview of the experimental design can be found in Fig. 1a.After signing in on the website, participants filled in a medical ques-tionnaire, and performed on-line versions of a Sternberg memorytask and a 2D mental rotation task. The medical questionnaire wasscreened for a history of psychiatric or neurological disease by qualifiedpersonnel.

2.2.1. Pre-training, post-training and follow-up measurementsThe neurofeedback training period (see neurofeedback training) was

immediately preceded and followed by a quantitative EEG (qEEG)session that lasted about 4 h. The same recording was again con-ducted at a follow-up 3 months after the second measurement. Weaimed to achieve a one week delay between the first qEEG sessionand the start of training and between the end of training and the secondqEEG session.

The qEEG recording sessions were conducted in a dimly-lit andquiet experimental chamber that was not electrically shielded. Themeasurements consisted of a qEEG including neuropsychologicaltests according to the protocol developed by the Brain ResourceCompany (Brain Resource, Inc.). EEGwas acquired using a Neuroscan

system using a Quickcap and 40 channel NuAmps recording from 26electrodes located according to a 10–10 system from the followingsites: Fp1, Fp2, F7, F3, Fz, F4, F8, FC3, FCz, FC4, T3, C3, Cz, C4, T4,CP3, CPz, CP4, T5, P3, Pz, P4, T6, O1, Oz and O2, referred to algebraicallylinked ears. Horizontal eye movements were recorded from electrodesplaced 1.5 cm lateral to the outer canthus of each eye. Vertical eyemovements were recorded with electrodes placed 3 mm above themiddle of the left eyebrow and 1.5 cm below the middle of the leftbottom eye-lid. In addition, facialmuscle activitywas recorded bilateral-ly from the orbicularis oculus and the masseter muscles. Electrodermalactivity was recorded from the left index finger. Heart rate was recordedfrom an electrode placed on the left wrist, and R–R intervals weredetected by a computer algorithm that was manually checked andcorrected if necessary. EEG data was recorded relative to the virtualground and referenced offline to linked mastoids; EMG and heartrate data was recorded using a bipolar set-up. Electrode impedancewas kept below 5 kΩ. Data was low-pass anti-aliasing filtered at100 Hz (12 dB/octave roll-off) and then sampled at 500 Hz using a22 bit A/D converter.

The EEG was recorded during periods of 2 min eyes open and eyesclosed, and the data recorded in these intervals is used in the presentreport. Event-related potentials in an auditory oddball task, Go/NoGotask, and continuous performance task were also recorded, but arenot reported here. Also not included here are the data collectedfrom the neuropsychological testing module (IntegNeuro).

After the second and third qEEG sessions, interviewswere conductedwith each participant individually and these were recorded.

2.2.2. Training sessionsIn the 3 weeks after the first qEEG measurement, the participants

came to the lab to undergo the neurofeedback training on 15 consec-utive working days. This turned out to be practically impossible forsome participants, but everyone completed the 15 training sessionswithin 4 weeks. The training sessions took place in a normal officeroom, in which each participant was seated in a comfortable recliningchair in front of a small table with a laptop on it. There were five suchchairs and tables with laptops in the room, separated by wooden par-titions, so that 5 participants could be trained at the same time by asingle experimenter. The whole training session was automated asmuch as possible. The experimenter supervised the whole trainingsession, and only took action in case something was wrong (usuallybad electrode contacts, which were automatically signalled).

The participants were given a set of headphones that they used forlistening to their favourite music. Participants could either bring theirown music for that particular day on an MP3 player, or they could selectthat day's music from a playlist containing thousands of songs from amyriad of artists. Therewas no limitation to the kind ofmusic participantscould listen to. Categories included genres like hard rock, easy listeningand classical music.

EEG was measured using Ag/AgCl ring electrodes (outer diameter:12 mm; inner diameter: 5 mm), mounted at the inner side of theband connecting the ear covers of the headphone. The electrodeswere positioned roughly over C3 (measured against A1, mounted inthe left ear cover) and C4 (measured against A2, mounted in theright ear cover). Tap water was used to optimise the contact betweenthe electrodes and the skin. No impedance checks were made, but thequality of the data was assessed on-line via the ratio of the power inthe 49–51 Hz (noise) range, relative to the power in the 4–35 Hz (signal)range. The signals were amplified (DC-400 Hz) and sampled at a rateof 1024 Hz by a 24 bit A/D converter on a Nexus-10 portable device(MindMedia B.V., The Netherlands). The signals were transmittedvia Bluetooth to a PC that controlled the experiment, and at thesame time stored the data on a flash SD card for off-line analysis.

Horizontal and vertical EOG was measured from electrodes at theouter canthi and above and below the left eye, respectively. The electro-cardiogram (ECG) was recorded from electrodes attached to the left

Fig. 1. Schematic representation of the time line of the whole experiment (a), and of a single neurofeedback training session (b).


and right wrists for ease of electrode application. Skin conductanceand respiration were also measured, but are not reported here. Allthese signals were amplified and sampled by means of the Nexus-10device.

The neurofeedback protocol was applied as follows. Eight timesper second (each 125 ms) a segment of the preceding 4 s was filteredby a third-order Butterworth band-pass filter from 4 to 35 Hz, and afirst order notch filter at 50 Hz. To be usable for a feedback update,this segment should fulfil three criteria: (i) no clipping or overflowof the amplifier; (ii) peak-to-peak of at most 200 μV; (iii) the ratio be-tween the line noise (49–51 Hz) and EEG power (4–35 Hz) should besmaller than 1.0. The feedback update depended on the group that aparticular participant was assigned to:

A. Alpha training. For each channel (C3–A1, C4–A2), the relativealpha power was calculated as the sum of the power in the alphaband (8–12 Hz), divided by the sum of total power (4–35 Hz). Alow cut-off frequency of 4 Hz was used in the calculation of totalpower to be relatively free of slow drift artefacts. If both channelsshowed a good epoch of 4 s, then the average of the two channelswas used. If only one channel yielded a good epoch, then only thatchannel was used. If neither channel resulted in a good epoch,then no feedback update was given. The relative alpha measurewas filtered by a first-order IIR filter with a time constant of 4 s.In this way, the speed of the changes in the feedback wassmoothed somewhat, and did not go back and forth too quickly.The feedback measure was used to drive the cut-off frequency of afirst-order high-pass filter built into the audio path of the musicplayed through the headphones. This cut-off frequency equalled2 Hz if the current alpha level was greater than the maximumalpha level for the previous part, and 2000 Hz if the currentalpha level was lower than the minimum alpha level for the pre-vious part. For intermediate levels, a linear interpolation wasdone in such a way that the cut-off frequency was high for lowcurrent alpha levels and low for high alpha levels. Using the fol-lowing formula: fc=2000−(2000−2)∗(alpha−min_alpha) /(alpha_max−alpha_min), where fc=audio cut-off frequency,alpha=current alpha level, min_alpha=minimum alpha level,and max_alpha=maximum alpha level. This procedure resulted

in a very intuitive feedback mechanism in which a person's ownwell-known favourite music sounded thin and distant if alphalevels were low because the low tones were removed. Whenalpha levels were high, however, the music sounded rich andfull, as the participants were used to hearing it.

B. Random beta training. The procedure for this condition was exactlythe same as for the alpha group, except for the EEG frequency usedto drive the audio filter, which was always in the beta range. Thebeta level was a 4 Hz bin (same bin width as the alpha range),randomly chosen on each day (session) between 14 and 30 Hz.Beta bands on different days did not overlap. The minimum andmaximum beta levels were calculated from the first eyes openmeasurement of that day in the band that was chosen for thatday. Thus, the random beta group also received training, but notin a fixed frequency band, and not in the alpha range.

C. Control: no feedback (music only). In this condition, the quality of themusic did not vary as a function of an EEG rhythm. The participants inthis group always listened to their favourite music in full quality.Thus, there was no training in this condition, but the participantsdid complete all tasks just like the participants in the other twoconditions.

Importantly, all participants received no instructions about thetraining. They were not told that they were expected to increase thequality of the music; they were only asked to ‘sit back and relax’.

A training session on a particular day always consisted of the samesequence of tasks. After the signals were determined to be valid, a base-line measurement of 5 min rest with eyes opened was recorded, fol-lowed by 5 min with eyes closed. After that, 3 neurofeedback trainingintervals of 8 min duration were interspersed by cognitive tasks lastingabout 5 min each. The (fixed) sequencewas:flanker task, neurofeedbacktraining 1, stop-signal task, neurofeedback training 2, Stroop task, neuro-feedback training 3, N-back task. The purpose of this sequence, which isdepicted in Fig. 1b, was fourfold. First, we wanted to avoid participantsfalling asleep during one long neurofeedback session in which theysimply listened to their favourite music in a comfortable recliningchair. Secondly, we wanted to assess the effects of alpha trainingon simple cognitive tasks administered longitudinally. Thirdly, by al-ternating conditions of relaxation and cognitive work, we hoped to


capture details relating to the dynamics of the alpha rhythm. Finally,it allowed us to match alpha levels to the audio filter.

2.2.3. Questionnaires and subjective dataWe also assessed possible effects of the neurofeedback training on

mood, perceived stress, sleep quality, and general quality of life. Theparticipants were asked to fill in questionnaires on the first, middleand last training sessions, as well as in the follow-up session. Becausethe first and last training sessions were close in time to the pre- andpost-training qEEG sessions, we used the questionnaire data fromthose sessions to match the alpha power and heart rate variability(HRV) data on the pre- and post-training qEEG sessions. The ques-tionnaires that the participants filled in were:

Mood: we used the abbreviated Dutch version of the Profile ofMood States (POMS; Van der Ark et al., 2006).Stress: the tension subscale of the POMS was used as a measure ofsubjectively perceived stress. In addition, we used the Dutch ver-sion of the Perceived Stress Scale (PSS, Cohen et al., 1983).Sleep: a Dutch translation of the Karolinska Sleepiness Scale (KSS;Akerstedt, 1990) was administered. The participants also filled inthe Dutch version of the Sleep-50 questionnaire (Spoormakeret al., 2005).Quality of life: the Dutch WHOQoL-bref is an abbreviated versionof the WHO Quality of Life questionnaire that correlates wellwith the full version (Trompenaars et al., 2005).Subjective relaxation: at the beginning of each training session, theparticipants also indicated their momentary subjective relaxationstate using a visual analogue scale (VAS) presented on the computerscreen, onwhich they had to position a vertical mark on a horizontalline using the computer mouse. They also did this after each sepa-rate phase of a training session (eyes open, eyes, closed, trainingperiods, cognitive tasks). Thus, there were 10 subjective relaxationmeasurements taken in each of the 15 training sessions.

2.3. Data analysis

2.3.1. qEEG measurementsThe signal analysis in the alpha frequency domain in the pre-

training, post-training, and follow-up sessions was computed on2 min eyes open and 2 min eyes closed segments. Alpha powerspectral density was estimated using the Welch (1967) methodon digitally filtered EEG. The EEG and EOG data channels werefirst off-line filtered with a 50 Hz first order (6 dB down per octave)notch filter for removing line noise, and then by a third-order(18 dB down per octave) Butterworth high-pass filter at 1 Hz andlow-pass at 65 Hz. EEG spectra were computed and the EEG recordwas then segmented in 75% overlapping intervals of 4 s. Thus, eachepoch was shifted 1 s forward in time, and there were 120 segmentsin total. Horizontal and vertical EOG channels were used to correct theEEG signal for eyemovement artefacts using a linear regressionmethodwith separate coefficients for horizontal and vertical eye movements(cf. Verleger et al., 1982). This correction was only applied if theEOG in a particular segment exceeded 60 μV. The segments werethen transformed to the frequency domain using a Hanning windowfor tapering. The FFT power values were then transformed to a log10scale and all frequency components (0.25 Hz resolution)were averagedto form the frequency bands Alpha (8–12 Hz), Beta (15–30 Hz), andTotal Power (4–35 Hz). Relative alpha and beta power were calculatedas the ratio between alpha or beta and total power.

We also used a novel method to estimate the individual alpha peak(IAF). Peak alpha varies as a function of cognitive performance andmayrepresent a permanent functional parameter (e.g., Klimesch, 1997),and is therefore a useful and acceptable measure. This method is

more fully described in Appendix A, and used in the Results sectionas an addendum, not a replacement, of alpha power measured asthe mean in the 8–12 Hz band. However, we feel that it capturesthe essential features of the peak alpha concept better than traditionalmethods.

2.3.2. Heart rate variabilityThe raw R–R intervals were exported to the Kubios package for

HRV analysis (Tarvainen et al., 2009). This package produced a cardiota-chogram, which was again checked by eye for any remaining artefacts(there were none). The package calculates time domain measuresHRV, of which we used the following: the standard deviation of theRR intervals (SDNN), the root mean square of successive differences(RMSSD), the number of successive intervals differing more than50ms (NN50), the corresponding relative amount of these intervals(pNN50), and the HRV triangular index. The frequency domain mea-sures we used were: low frequency power (LF; 0.04–0.15 Hz), highfrequency power (HF; 0.15–0.4 Hz), also in normalised units (nLFand nHF, relative to total power), the LF/HF power ratio, and peakfrequencies for each band. All measures were calculated based onFFT transformed heart period series, and on autoregressive modelsof the series. Finally, the following non-linear methods of HRVwere used: the SD1 and SD2 axes of the Poincaré plots, approximateentropy (ApEn), sample entropy (SampEn), detrended fluctuationanalysis (DFA), and the correlation dimension D2. All measures,which followed international recommendations (Malik et al., 1996;Berntson et al., 1997) were log10 transformed to make their samplingdistributions Gaussian.

2.3.3. Statistical analysisBefore statistical analysis, we checked all measures for normality of

the sampling distribution, and log-transformed the data if necessary.We then computed linear mixed models with fixed between-subjectsfactor Group (levels A=alpha, B=beta, C=control), and randomwithin-subjects factors Sessions (levels 1=pre-training, 2=post-training, 3=follow-up), and Participant. For tests involving differentelectrodes, we defined further appropriate fixed factors as necessary.Covariance structure was determined based on the information criteriagiven by the statistical output, but did not differ very much betweencovariance types. We always used an autoregressive (1) pattern withrandom intercept and random slope, because that seemed the bestmatch to our design with three sessions. Mixed models have theadvantage being able to cope withmissing values. Because of a varyingnumber of missing values for each dependent variable, the number ofdegrees of freedom for identical tests may vary somewhat betweendependent variables. Observed power was calculated at an alphalevel of 0.05. All calculations were done using SAS PROC MIXED 9.1.

3. Results

3.1. Alpha power recorded with qEEG and with the novel system

Fig. 2 shows the full power spectra of all participants as measuredwith the novel headset on the first training day. These data resemblespectra recorded with conventional EEG systems quite well, and thealpha peak is clearly discernible around 10 Hz. Fig. 3 shows grandaverage spectra recorded from C3 and C4 with the conventionalfull-cap EEG system on the first qEEG measurement, together withspectra recorded with the novel system on the first training day.This figure shows that the systems were remarkably alike in thealpha frequency range; the difference at higher frequencies arisesbecause of the different ways that 50 Hz line noise was filtered.Alpha power (8–12 Hz) recorded with the qEEG system and withthe novel system could not be distinguished statistically (qEEG versusheadset: F(1,43)=1.67, n.s.; eyes open versus eyes closed: F(1,43)=9.87, pb0.01; C3 versus C4: F(1,43)=0.28, n.s.). In addition, the

Fig. 2. Full power spectra of all participants, recorded with the novel EEG system usingwater based electrodes mounted into the headset. Electrode C3–A1, eyes closed. Notethe prominent alpha peak around 10 Hz.

Fig. 4. Percentage of participants who – unprompted – explicitly mentioned that thetraining was relaxing in the exit interview, as a function of group.


correlations between the alpha power values recorded with the twosystems were always between 0.8 and 0.9 (C3, eyes open: 0.88; C3,eyes closed: 0.80; C4 eyes open: 0.88; C4, eyes closed: 0.84). In sum,the novel system allowed recording and thus training alpha activityvery well.

3.2. Guided interviews

As a qualitative measure of the way in which the participants per-ceived the training sessions, we examined how many participants ineach group indicated that the sessions were relaxing, without thembeing prompted. We did not inform the participants that this studywas about relaxation, and asked them neutrally about how they expe-rienced the training. We then counted the number of people thatreported spontaneously that they thought the training was relaxing.The resulting percentages are depicted in Fig. 4. In the alpha traininggroup (A), 53% of the participants spontaneously remarked that theyexperienced the training sessions as relaxing, as opposed to 20% inthe random beta group (B) and 21% in the music only control group

Fig. 3. Grand average power spectra of C3–A1 (left) and C4–A2 (right), recorded under eyheadset.

(C). Hence, about twice as many in the alpha training group sponta-neously mentioned relaxation compared to the control groups.

3.3. Behavioural data (questionnaires)

Statistical tests on the questionnaires are given in Table 1. As canbe seen in the table, there were only a few effects of the trainingconditions on the behavioural scales. Only the Environment dimen-sion of the Quality of Life scale, and Factors Influencing Sleep,yielded statistically significant effects across sessions, as a functionof group. The means from these effects (Fig. 5) indicated that theywere mainly due to the music only control condition not to one ofthe training conditions. However, looking at the means displayedin Fig. 5, it is striking that, with the exception of those two tests,all patterns are largely as one would expect based on the hypothe-sis that alpha training – not random beta or music only – inducesrelaxation. Perceived stress decreases over sessions in the alphagroup, whereas it remains roughly the same in the other groups.POMS Vigour increases in the alpha group and decreases in randombeta and music only control groups. The POMS Negative Moodscales all seem to decrease to a greater extent in the alpha groupthan in the other groups. Similar relations can be observed for the

es open and eyes closed conditions from the conventional qEEG system and the novel

image of Fig.�2

image of Fig.�3

image of Fig.�4

Table 1Statistical effects for behavioural data (questionnaires).

Effect Group Session Group∗Session

df F Power df F Power df F Power

Perceived stress 2,47 0.10 0.06 2,89 0.37 0.11 4,89 0.44 0.15

MoodTension 2,47 0.07 0.06 2,86 2.35 0.46 4,86 1.30 0.39Anger 2,47 0.33 0.10 2,86 0.05 0.06 4,86 0.39 0.14Fatigue 2,47 0.50 0.13 2,86 0.46 0.12 4,86 1.43 0.43Depression 2,47 0.10 0.06 2,86 0.56 0.14 4,86 0.96 0.29Vigour 2,47 1.14 0.24 2,86 0.24 0.09 4,86 0.68 0.21

Quality of lifePhysical 2,47 0.12 0.07 2,86 0.19 0.08 4,86 2.19 0.62Psychological 2,47 0.27 0.09 2,86 0.70 0.17 4,86 0.31 0.12Environment 2,47 1.56 0.32 2,86 0.44 0.12 4,86 2.79⁎ 0.74Social relationships 2,47 0.05 0.06 2,86 0.59 0.15 4,86 0.24 0.10

SleepRating 2,47 2.36 0.45 2,78 1.81 0.37 4,78 1.02 0.31Sleep duration 2,47 1.42 0.29 2,78 0.59 0.15 4,78 1.63 0.48Complaints daily function 2,47 0.81 0.18 2,78 1.56 0.32 4,78 1.21 0.36Factors influencing sleep 2,47 0.12 0.07 2,78 0.17 0.08 4,78 4,50⁎⁎ 0.93Nightmares 2,47 0.98 0.21 2,78 2.33 0.46 4,78 0.36 0.13Circadian rhythm sleep disorder 2,47 1.38 0.28 2,78 7.50⁎⁎ 0.94 4,78 1.46 0.43Insomnia 2,47 2.98 0.55 2,78 0.09 0.06 4,78 0.76 0.23

⁎ pb .05.⁎⁎ pb .01.


Quality of Life and the Sleep scales. Thus, the formal tests might notreveal statistically valid differences, but it is intriguing that mostvariables show the expected pattern. Such a situation suggests a possi-ble power problem— that is, theremight have been too fewparticipantsin the study to draw statistically valid conclusions. The overall lowpower values listed in Table 1 support such an interpretation.

Subjective relaxation identified using the visual analogue scale(VAS) was not measured during the pre-training, post-training, andfollow-up measurements. However, we compared the average re-laxation scores of the first and the last training session, as a meansto investigate whether subjective relaxation changes during thetraining were specific to the training groups. Unfortunately, thiswas not the case. Mean relaxation scores were slightly smaller inthe last compared to the first training session, but the correspondingeffect failed to reach significance (Session: F(1,47)=0.36, n.s., power=0.09; Group: F(2,47)=2.20, n.s., power=0.43; Session∗Group:F(2,47)=0.43, n.s., power=0.12).

3.4. qEEG measurements

We performed omnibus tests on relative alpha power (8–12 Hz/4–35 Hz) for a subset of posterior electrodes, C3, C4, P3, P4, O1, O1,and we defined the appropriate within-subjects factors Position (C,P, O), and Laterality (left, right). As expected, alpha power was largerfor eyes closed than for eyes open (F(1,47)=491.15, pb0.001,power=1.00), and larger at occipital and parietal electrodes com-pared to central electrodes (F(2,94)=8.06, pb0.001, power=0.95).The difference between eyes open and eyes closed was largest in thealpha training group and smallest in the music only control group,and the random beta group had an intermediate value (F(2,47)=5.79, pb0.01, power=0.85).

In all three groups, alpha power increased over sessions(F(2,90)=66.96, pb0.001, power=1.00). Importantly, the increasein relative alpha power was largest for the alpha training group(F(4,90)=20.60, pb0.001, power=1.00), especially in the eyesopen condition (F(4,89)=6.39, pb0.001, power=0.99). Fig. 6 providesa visual representation of this effect. The figure not only shows the

greatest increase in the alpha training group, but also that alpha levels,especially in eyes open, continued to increase after the post-trainingmeasurement to the follow-up meeting 3 months later. For eyes open,the increase between the pre- and post-trainingmeasurements roughlyequals a further increase between the post-training and the follow-upmeasurement. Apparently, the participants learned how to controltheir alpha level, andmaybe through rehearsing evenwithout receivingfeedback were able to increase it further. At any rate, the effect of alphatraining did not fade after training.

We also tested relative beta power (15–30 Hz/4–35 Hz) to inves-tigate whether they would show a similar pattern to that seen foralpha activity. For relative beta power, a completely different patternof results was found. Relative beta power was larger in eyes opencompared to eyes closed (F(1,47)=126.13, pb0.001, power=1.00) —as expected the opposite of alpha power. The difference between eyesopen and eyes closed was exactly equal for the alpha and randombeta training groups, and smaller in the music only control group(F(2,47)=9.48, pb0.001, power=0.97). Beta power decreasedover sessions (F(2,90)=13.99, pb0.001, power=1.00), especiallyin the random beta group (F(4,90)=15.50, pb0.001, power=1.00).

We then used a non-linear regression model (Appendix A) to esti-mate the individual alpha frequency and its power. The IAF power (Cparameter in the model) was greater for eyes closed than for eyesopen (F(1,47)=1074.14, pb0.001, power=1.00), and largest overoccipital electrodes and smallest over the central pair (F(2,94)=47.24, pb0.001, power=1.00), especially for eyes closed (F(2,94)=36.91, pb0.001, power=1.00), and this scalp distribution was mostprominent in the alpha training group (F(4,94)=3.70, pb0.01,power=0.87). As in the relative alpha power, IAF power estimatedby non-linear regression increased from the pre-training to thepost-training session, only for training with eyes open (F(2,89)=3.89, pb0.05, power=0.70), and continued to increase in the alphatraining group but not in the other groups (F(4,89)=6.83, pb0.001,power=0.99). No other statistically meaningful differences werefound for the C parameter in the model. The IAF itself (D parameterin the model) was smaller in the random beta group compared tothe alpha and music only control groups (F(2,47)=19.00, pb0.001,

Fig. 5. Averaged scores of the behavioural questionnaires, separately for groups and recording sessions.

289G.J.M

.vanBoxtelet

al./InternationalJournalofPsychophysiology

83(2012)

282–294

image of Fig.�5

Fig. 6. Increase in relative alpha power over sessions, as a function of experimental group, separately for recordings made with eyes open and eyes closed.


power=1.00), especially for eyes open (F(2,47)=35.20, pb0.001,power=1.00) and smaller at central than at parietal and occipitalelectrodes (F(2,94)=24.78, pb0.001, power=1.00), especially inthe eyes open task in the random beta group (F(2,94)=2.73,pb0.05, power=0.73). Turning now to the possible effects of trainingin the IAF, we found that in the recordings with eyes closed the indi-vidual alpha frequency slightly decreased from the pre- to the post-training session, relative to the eyes open recordings. However, atthe follow-up measurement, both eyes open and eyes closed record-ings showed IAF values that had returned to the pre-training level(F(2,89)=6.04, pb0.01, power 0.87). This effect was somewhat stron-ger in the random beta group compared to the other two groups(F(4,89)=2.58, pb0.05, power=0.70). Importantly, there were nosession effects of alpha activity training; the IAF seems to be a rela-tively stable subject characteristic uninfluenced by the alpha trainingprocedure.

We also produced topographic maps of power at the IAF resultingfrom the non-linear regression model described in Appendix A. Fig. 7illustrates the difference between post-training and pre-training peakalpha activity (eyes open) as indexed by the C-parameter, along with

Fig. 7. Upper: topographic map (eyes open) of the difference between post- and pre-training peak alpha power measured by the C parameter in the nonlinear regression fit(see Appendix A; scale in log10(μV2)); middle: probability map of the paired t-test ofpost- versus pre-training peak alpha activity; bottom: significance map of same t-test.

the corresponding probability and significance maps. It can be seenthat the alpha increase that resulted from the alpha training mainlyinvolved the left fronto-central areas (near to the training site), andsurprisingly also the right occipital area (distant from the trainingsite). No appreciable changes were observed for the random betaand music only control groups. The activity recorded when the eyeswere closed did not show a statistically meaningful increase inalpha activity from the pre-training to the post-training session.

3.5. Heart rate variability

The mean R–R interval was not influenced by the training (e.g., ef-fect of Sessions: F(2,90)=2.29, n.s.). Overall heart rate variability(SDNN) did not respond to training either, but in the random betagroup was greater for eyes closed than for eyes open, whereas the re-verse was true for the music only control group (Task∗Group interac-tion; F(2,47)=4.46, pb0.05, power=0.74), and the same patternwas found for RMSSD (F(2,47)=3.35, pb0.05, power=0.61). Spec-tral analysis was used to decompose the variability further, and weobserved the same effect for high frequency power (HF; F(2,47)=4.61, pb0.05, power=0.75), but not for low frequency power(F(2,47)=2.51, p>0.05). No other effects of training on any of theHRV parameters were found, so it appears that our training did notaffect heart rate variability systematically. The full analyses and in-volved means are available as supplementary information.

4. Discussion

The purpose of this study was to investigate whether alpha activitytraining using an innovative self-guided system with water-basedelectrodes mounted in an audio headset was feasible. There are severalindications that supported this feasibility. First, the system was able torecord EEG in the alpha range, of which the spectral properties corre-spond to what is usually reported in the literature. Spectral plots ofEEG records show power decreases as a function of frequency, with aclear alpha peak around 10 Hz. In addition, alpha power recorded witha conventional qEEG system and with the novel system was statisticallyindistinguishable. Importantly, alpha training using our new systemresulted in an increase in alpha activity as recorded using a conventionalEEG system. Alpha activity increased specifically in the group thatreceived alpha training, but not in the other groups in which either

image of Fig.�6

image of Fig.�7


random beta or no training was given. Beta activity did not increasein the alpha training group, indicating that the training elicited a fre-quency specific effect, and did not simply increase overall EEG power.The present findings add to the growing body of literature on thetrainability of alpha activity in relation to relaxation (e.g., Dempsterand Vernon, 2009; Egner et al., 2002; Gruzelier, 2002). These specific ef-fects of alpha activity training on alpha activity as opposed to betaactivity were evident despite a stringent control condition that includedreal training of random beta frequencies (Hoedlmoser et al., 2008).For the most part the result of random beta training was statisticallyindistinguishable from a second control condition in which the partici-pants merely listened to their favourite music. The inclusion of thesetwo control conditions is a strong point of the present study, and demon-strates the trainability of alpha activity beyond doubt.

Alpha activity recorded with the eyes open exhibited the largestincrease. It is tempting to relate these findings to the fact that thealpha training was given with the eyes open. Alpha training is oftengiven with eyes open because the training itself may rely on informationprovided by the visual feedback (see Angelakis et al., 2007). However,the differential effects of training with eyes open and eyes closed onpost-training recordings of eyes open and eyes closed activity, is notwell studied (see Vernon et al., 2009). The present study was notdesigned to specifically study these effects, but our results suggestthat the effects with eyes open training may not generalise to trainingwith eyes closed.

Another important result of the present study is that the increase inalpha activity as a result of alpha training persisted, and increased slight-ly, at the follow-up session 3 months after the last training session. In theinterval between the post-training measurement and the follow-upmeasurement, the participants did not receive any additional training,or other information about the experiment. Furthermore, these effectswere specific to the alpha training group, so the changes brought aboutby the alpha trainingwere long lasting. It is possible that the participantsin the alpha training group, unlike those in the random beta or musiconly control groups, learned to produce a solid alpha rhythm that theycould possibly evoke at will whenever needed. Such a possibility notonly supports the conclusion that use of such an innovative self-guidedsystem can elicit learning, with regards to altering alpha activity, butthat due to the self-guided nature of the learning it is also more easilymaintained. However, further research is needed in order to substantiatesuch a claim.

This pattern of results was roughly similar when alpha activity wasdefined broadly as being in the 8–12 Hz range, relative to the totalrange of 4–35 Hz, and in the model of the individual alpha peakusing a non linear regression analysis. The regression model has theadvantage over the broad definition in that it is much more specific,and it has the ability to model the individual alpha frequency as wellas the power at that frequency. Although the training conductedhere was based on the broad 8–12 Hz power band and was sufficientto show clear changes as a function of the training, future researchmay wish to utilise an approach based on the IAF in order to furtherdevelop and refine this method. Such an approach will allow for amore individualised training scheme, and also avoid suboptimal train-ing effects for participants who have an IAF outside of the 8–12 Hzrange. Related to the above, an important finding of this study wasthat alpha training increased power within the IAF range, but did notalter the IAF itself, which remained stable across groups andmeasure-ment sessions. This is remarkable because our study lasted for severalweeks and 72% of the participants were female. Luteal phase is knownto affect the IAF (Bazanova and Aftanas, 2008), but appeared not tohave influenced our results differentially for our training groups. Theinsensitivity of the IAF to alpha activity training is important as it re-lates to possible side effects. The present method of training using aneasily applicable electrode set-up and a very intuitive form of auditoryfeedback was able to increase alpha activity without displaying signsof potentially adverse side-effects.

The results of our study also support the notion that instrumentalconditioning of alpha and other brain rhythms using double-blind,placebo-controlled experiments is feasible. Using a random beta train-ing protocol as a control, as utilised by Hoedlmoser et al. (2008), andalso recently applied in the gamma neurofeedback studies of Keizer etal. (2010), seemed to work quite well. Importantly, the random betagroup did not exhibit any adverse effects of the random beta training.Hence, random beta might be considered an appropriate control in sit-uationswhere EEG activity outside the beta band needs to be trained.Our second control group, in which the participants listened to theirfavourite music without receiving any training, constitutes a lessthan ideal control for the alpha training group in that those groupsnot only differ in alpha training, but also in receiving rewards ofwhatever nature. It is instructive to see that alpha activity trainingproducedmore alpha activity along with verbal reports of relaxation,than the music only control group. This is surprising because listeningtomusic by itself is considered to be a relaxing experience. Our researchwould suggest that combining alpha activity training with listening tomusic adds an extra degree of relaxation.

It could be argued that using random beta training as a controlcondition is less than optimal. The original work by Hoedlmoser etal. (2008) used rhythms above as well as below the frequency rangetrained in the experimental condition (12–15 Hz in their case), butwe used only higher frequencies in our control condition. Using equalcontrol frequency bins above and below the alpha range would haveresulted in training, in some sessions, of frequency ranges where arte-facts may contaminate the signals, such as eye blinks in the delta andsometimes theta range. Using only frequency bins higher than alphaavoided such possible problems, perhaps with the drawback thatRolandic beta activity was trained, which is not independent ofRolandic alpha (mu). Indeed, the beta group did show a slight in-crease in alpha activity as a result of training. However, this increasewas smaller than in the alpha group, presumably because a differentfrequency bin was used on each day (RANDOM beta), whereas thesame alpha bandwas used in the alpha group on every day. Althougha case could be made for using random control frequencies outsidethe Rolandic beta band (higher than 24 Hz), it can also be arguedthat the current design provided an extra rigorous test in that thechance for finding significant differences were minimised, howeverstill found. Finding the optimal control condition in neurofeedbackstudies remains an important issue for future work.

It was also found that training of alpha activity at central sitesincreased alpha activity more posteriorly. This is a quite surprisingfinding, although to some extent consistent with others who havefound that neurofeedback training can lead to changes beyond targetfrequency and location (e.g., Egner et al., 2004). The topographic mapsindicated that some effect of the training was also present aroundcentral sites, at least to the left, which is the hemisphere contralater-al to the dominant right hand in our participants. These maps, alongwith the statistical results using mixed linear models, point to effectsof central alpha activity training on more posterior sites associatedwith the classical alpha rhythm. In all statistical tests, alpha activitylevels were largest at the occipital electrodes, with parietal electrodesalso displaying considerable activity. This is what is usually reportedfor the classical alpha rhythm (Shaw, 2003). It is unclear how the ac-tivity in central and posterior areas is related. A first possibility isthat we trained the mu or sensorimotor rhythm and that this rhythmis somehow coupled to themore posterior alpha rhythm, possibly viathe thalamus. This possibility seems unlikely because the mu rhythmis usually slightly faster than the alpha rhythm (Pfurtscheller andAranibar, 1979). A second option is that the posterior alpha rhythmspreads via volume conduction to more central sites, where it is pickedup by our sensors and increased by the training. Indeed, Nunez (1995)suggested that posterior alpha activity may spread toward the frontalregion, and as a consequence can be picked up by central electrodes.However, if this were the case, it would seem that the increased alpha

2 This possibility was suggested to us by an anonymous reviewer.


activity resulting from the training (not necessarily the absoluteamount of alpha power) would be strongest at the central electrodes.In contrast, we found that the effect of training was maximum at oc-cipital sites. A third possibility which is more consistent with the pat-tern of results is suggested by the work of Thatcher et al. (1976).Based on theoretical and neurophysiological grounds, they proposeda model of alpha activity involving two generators; one posterior(parieto-occipital) and one more central (centro-parietal), with re-ciprocal connections and even mutual competition for neuronal re-sources. There is some support for this model of alpha generationbased on coherence between brain regions, but the present data canneither support not falsify the model. The distant effects of trainingspecific frequencies and locations are clearly an interesting issuefor further study.

The present study further indicates that in young, healthy partici-pants, a long lasting change in alpha activity can be obtained in a limitednumber of short training sessions. We used 15 sessions, and in eachsession, the duration of the training was 24 min. Yet this limitedamount of training, without explicit instructions, was sufficient toproduce an increase in alpha activitywhich remained evident 3 monthsafter the last training session, without any additional training. Furtherworkmight reveal that even less training is capable of producing similarstable changes. Of course, shortening the amount of training in eachtraining session may also restrict the possibility of eliciting a robusteffect. We currently hypothesise that three main reasons contributeto the success of the training method used. First, the ease of settingup the EEG system hasmany advantages to start with. Using a standardEEG system, which involves attaching electrodes with gel, sticking EOGelectrodes around the eyes, etcetera, can be irritating for the partici-pants, resulting in a higher state of stress before the training beginsand is also more time consuming. With the current system, it takesless time to set participants up and as a consequence they may bemore relaxed. Secondly, the intuitiveness and familiarity of the auditoryfeedbackmight also have played a role. For instance, providing feedbackabout brain activity is often very indirect. Nevertheless, modern appli-cations using game-like interfaces represent a significant improvement.However, our system worked by manipulating the participants' ownfavourite music to provide feedback regarding target cortical activity.Given that the participants know their own music very well and assuch would be able to detect changes in tone quickly and easily itrepresents amode of feedback that is both engaging and informative.A third reason, which admittedly may be a little more speculative,concerns the alternation of cognitive effort and relaxation that waspart of our training scheme. For instance, a sign of good cardiovascularcondition is the ability to recover quickly fromphysical effort and can bemeasured by the recovery of heart rate after the physical exercise(Chorbajian, 1971). It may be that similar processing capabilitiesare important for peak mental condition, and the fact that our train-ing incorporated alternating cognitive tasks with relaxation may bemore effective than simply training absolute alpha levels per se.

Relating the increase in alpha activity resulting from the trainingto subjective and cardiovascular indices proved to be cumbersome.Nevertheless, use of a semi-structured interview resulted in abouttwice as many participants in the alpha training group mentioningthat the training was relaxing, without them being prompted for it.However, this represents only a very crude measure of relaxation.We expected to find relations between (the increase in) alpha activityand behavioural measures of stress, relaxation and sleep. However,althoughmany of the analyses showed the expected pattern we failedto obtain statistical support for the hypothesised relations. As men-tioned above, it may be possible that our groups were too small sothat tests lacked sufficient statistical power. In addition, it is possiblethat alpha activity and relaxation defined behaviourally representdescriptions of a similar process but at entirely different levels.

The same may be true for the lack of a clear relationship betweenalpha activity and heart rate variability seen in our study. It was

expected that indices of relaxation, alpha activity, and HRV wouldbe positively correlated. Surprisingly however, it is difficult to findstatistically meaningful correlations in the literature. In the presentstudy, the inclusion of alternating sequences of training and cognitivetasks in the protocol may have prevented such effects to occur in thatthat the interaction between relaxation and cognition may haverendered the cardiovascular system insensitive to training effects.2

To sum up, we compared training of alpha activity to training ofrandom beta and no training control, and found that alpha activitytraining at central sites enhanced alpha activity at posterior sites.Although the relationship between alpha activity and the behaviouralindices of relaxation were weak we have demonstrated the feasibilityof alpha activity training using an easy, wireless, water-based electrodesystem combined with an intuitive form of auditory feedback based onthe participant's own favourite music. We consider our system to be animportant step forward in the development of a system for instrumen-tal conditioning of brain rhythms that can be used both in the clinic aswell as at home. In our view, the availability of such a system wouldsolve a number of technical weaknesses that surround such systemscurrently, so that more emphasis can be placed on factors that reallymatter in these studies, such as the protocols to be used, the numberand nature of the training sessions, the influence of different instrumen-tal conditioning schedules, and the transfer of the trained brain area toother areas, to name but a few.

Acknowledgements

The invaluable assistance of Fienke Geurdes and Vivian Holten insetting up this study and in performing long and tedious work inrecording the pre-training, post-training, and follow-upmeasurementsis gratefully acknowledged. The training sessions also benefited fromtheir efforts, as well as that of Juliëtte van der Wal, Francisco Steenbak-kers, and Emma Rainford.

Appendix A

Measuring peak alpha power in individual participants and at dif-ferent electrode channels has proved to be cumbersome, and manydifferent methods exist. Very often, total alpha power is reportedas the average power of a single frequency band located somewherebetween 7 and 13 Hz (usually 8–12 Hz). The work of Pfurtscheller,and later Klimesch and colleagues, suggested that it is important to dis-tinguish between upper and lower alpha bands (e.g., see Klimesch,1999, for a review). Researchers often use fixed upper and loweralpha bands (for instance, 8–10 Hz and 10–12 Hz), or 2-Hz bandsimmediately below and above the Individual Alpha Freqency (IAF).Klimesch (1999) reported that the IAF is usually located in theupper alpha band, and that the lower alpha band is usually broaderthan the upper alpha band. As a consequence, the use of more specificmeasures to denote the level of ‘alpha activity’ has been advocated.

We usednon-linear regression tomodel the IAF and its power. Usingleast squares, the error in the following function was minimised:

Log10 Power fð Þð Þ ¼ Aþ Bf þ Ce− f−Dð Þ2

2E2 þ �

where:

f frequency, 6–35 Hz, 0.25 Hz resolution [Hz]A a constant, reflecting individual variations in absolute

power [log10(μV2)]B the linear change in power as a function of frequency

[log10(μV2)]C the power at the individual alpha frequency [log10(μV2)]D the individual alpha frequency [Hz]

Fig. A1. Example of the model fit in a single subject (no. 10, alpha training group). The black trace represents the spectrum of averaged activity at the C3 electrode; the red trace isthe fitted model. Left: eyes open; right: eyes closed.


E the width of the alpha peak [Hz]measurement error.

Conceptually, this model consists of the sum of two components. Alinear regression line (A+Bf) represents the background power, whichdecreases as a function of frequency. An inverted U-shape (the Ce term)represents the alpha peak that arises on top of the linear decrease.

An example of the fit of this model on the data of an individualparticipant in our study is presented in Fig. A1. The figure showsthat the model overlaps the real data. The root mean squared of themeasurement error (RMSE) was small but differed statistically fromzero (F(1,241)=3984.06, pb0.001). The RMSE tended to co-varywith some of the model parameters, and correlated significantly andpositivelywith theA (r=0.31), C (r=0.13), andD (r=0.06) parameters,and negatively with the B (r=−0.06) and E (r=−0.26) parameters.However, the highest correlation of the RMSE and the parameterswas 0.31 (A), meaning that a maximum of 9.61% of the variance in theA parameter was explained by measurement error. For the important Cparameter, the percentage of explained variance due to measurementerror was only 1.85%.

The data displayed in Fig. A1 suggests that the measurement errorin these analyses might be largely due to activity in the beta range,where a straight line is fit that shows some deviation from the actualdata in that range. Further work is required to optimise the model incase beta power is to be modeled as well.

In conclusion, the presentmodel provided a good fit of the individualalpha peak.

Appendix B. Supplementary data

Supplementary data to this article can be found online at doi:10.1016/j.ijpsycho.2011.11.004.

References

Akerstedt, T., 1990. Subjective and objective sleepiness in the active individual. Inter-national Journal of Neuroscience 52, 29–37.

Angelakis, E., Stathopoulou, S., Frymiare, J.L., Green, D.L., Lubar, J.F., Kounios, J., 2007.EEG neurofeedback: a brief overview and an example of peak alpha frequencytraining for cognitive enhancement in the elderly. The Clinical Neuropsychologist21 (1), 110–129.

Berger, H., 1929. Über das Elektrenkephalogram des Menschen. Archiv für Psychiatrieund Nervenkrankheiten 87, 527–570.

Berntson, G.G., Bigger, J.T., Eckberg, D.L., Grossman, P., Kaufmann, P.G., Malik, M.,Nagaraja, H., Porges, S.W., Saul, J.P., Stone, P.H., Van der Molen, M.W., 1997.Heart rate variability: origins, methods, and interpretive caveats. Psychophysiol-ogy 34, 623–648.

Bazanova, O., Aftanas, L., 2008. Individual EEG alpha-indices and nonverbal creativity infemale during the spontaneous menstrual cycle in comparison with male subjects.Annals of General Psychiatry 7, S101.

Cahn, B.R., Polich, J., 2006. Meditation states and traits: EEG, ERP, and neuroimagingstudies. Psychological Bulletin 132, 180–211.

Chorbajian, T., 1971. Nomographic approach for the estimation of heart rate recoverytime from exercise in humans. Journal of Applied Physiology 31, 962–964.

Cohen, S., Kamarck, T., Mermelstein, R., 1983. A global measure of perceived stress.Journal of Health and Social Behavior 24, 385–396.

Dempster, T., Vernon, D., 2009. Identifying indices of learning for alpha neurofeedbacktraining. Applied Psychophysiology and Neurofeedback 34, 309–318.

Egner, T., Strawson, E., Gruzelier, J.H., 2002. EEG signature and phenomenology anfalpha/theta neurofeedback following training versus mock feedback. AppliedPsychophysiology and Biofeedback 27, 261–270.

Egner, T., Zech, T.F., Gruzelier, J.H., 2004. The effects of neurofeedback training on thespectral topography of the electroencephalogram. Clinical Neurophysiology 115,2452–2460.

Goldman, R.I., Stern, J.M., Engel Jr., J., Cohen, M.S., 2002. Simultaneous EEG and fMRI ofthe alpha rhythm. NeuroReport 13, 2487–2492.

Gruzelier, J.H., 2002. A review of the impact of hypnosis, relaxation, guided imageryand individual differences on aspects of immunity and health. Stress 5, 147–163.

Hardt, J.V., Kamiya, J., 1978. Anxiety change through electroencephalographic alphafeedback seen only in high anxiety subjects. Science 201 (4350), 79–81.

Hoedlmoser, K., Pecherstorfer, T., Gruber, G., Anderer, P., Doppelmayr, M., Klimesch,W., Schabus, M., 2008. Instrumental conditioning of human sensorimotor rhythm(12–15 Hz) and its impact on sleep as well as declarative learning. Sleep 31,1401–1408.

Keizer, A.W., Verschoor, M., Verment, R.S., Hommel, B., 2010. The effect of gammaenhancing neurofeedback on the control of feature bindings and intelligencemeasures. International Journal of Psychophysiology 75, 25–32.

Klimesch, W., 1997. EEG alpha rhythms andmemory processes. International Journal ofPsychophysiology 26, 319–340.

Klimesch, W., 1999. EEG alpha and theta oscillations reflect cognitive and memoryperformance: a review and analysis. Brain Research Reviews 29, 169–195.

Klimesch, W., Sauseng, P., Hanslmayr, S., 2007. EEG alpha oscillations: the inhibition-timing hypothesis. Brain Research Reviews 53, 63–88.

Lindsley, D.B., 1960. Attention, consciousness, sleep and wakefulness. Handbook ofPhysiology: Neurophysiology. American Physiological Society.

Malik, M., Bigger, J.T., Camm, A.J., Kleiger, R.E., Malliani, A., Moss, A.J., Schwarz, P.J.,1996. Heart rate variability: standards of measurement, physiological interpretation,and clinical use. Circulation 93, 1043–1065.

Mental Health Policies and Programmes in the Workplace, 2005. Geneva, World HealthOrganization (Mental Health Policy and Service Guidance Package).

Nowlis, D.P., Kamiya, J., 1970. The control of electroencephalographic alpha rhythmsthrough auditory feedback and the associated mental activity. Psychophysiology6, 476–484.

Nunez, P.L., 1995. Neocortical Dynamics and Human EEG rhythms. Oxford UniversityPress, London.

Perlini, A.H., Spanos, N.P., 1991. EEG alpha methodologies and hypnotisability: a criticalreview. Psychophysiology 28, 511–530.

image of Fig. A1




Pfurtscheller, G., Aranibar, A., 1979. Evaluation of event-related desynchronization(ERD) preceding and following voluntary self-paced movement. Electroencepha-lography and Clinical Neurophysiology 46, 138–146.

Rice, K.M., Blanchard, E.B., Purcell, M., 1993. Biofeedback treatments of generalisedanxiety disorder: preliminary results. Biofeedback and Self-Regulation 18, 93–105.

Shaw, J.C., 2003. The Brain's Alpha Rhythms and the Mind. Elsevier, Amsterdam.Spoormaker, V.I., Verbeek, I., Van den Bout, J., Klip, E.C., 2005. Initial validation of the

Sleep-50 questionnaire. Behavioral Sleep Medicine 3, 227–246.Tarvainen, M.P., Niskanen, J.-P., Lipponen, J.A., Ranta-aho, P.O., Karjalainen, P.A., 2009.

Kubios HRV — a software for advanced heart rate variability software. In: VanderSloten, J., Verdonck, P., Nyssen, M., Haueisen, J. (Eds.), IFMBE Proceedings, 22, pp.1022–1025.

Thatcher, R.W., Krause, P.J., Hrybyk, M., 1976. Cortico-cortical associations and EEG co-herence: a two-compartment model. Electroencephalography and Clincical Neuro-physiology 64, 123–143.

Trompenaars, F.J., Masthoff, E.D., Van Heck, G.L., Hodiamont, P.P., De Vries, J., 2005.Content validity, construct validity, and reliability of the WHOQOL-Bref in a

population of Dutch adult psychiatric outpatients. Quality of Life Research 14,151–160.

Van der Ark, L.A., Marburger, D., Mellenbergh, G.J., Vorst, H.C.M., Wald, F.D.M., 2006.Verkorte Profile of Mood States. Handleiding en verantwoording. Harcourt TestPublishers, Amsterdam.

Verleger, Gasser, T., Möcks, J., 1982. Correction of EOG artifacts in event-related poten-tials of the EEG: aspects of reliability and validity. Psychophysiology 19, 472–480.

Vernon, D., Dempster, T., Bazanova, O., Rutterford, N., Pasqualini, M., Andersen, S., 2009.Alpha neurofeedback training for performance enhancement: reviewing the meth-odology. Journal of Neurotherapy 13, 1–13.

Watson, C.G., Herder, J., Passini, F.T., 1978. Alpha biofeedback theory in alcoholics: an18-month follow-up. Journal of Clinical Psychology 34 (2), 765–769.

Welch, P.D., 1967. The use of fast Fourier transforms for the estimation of power spec-tra: a method based on time averaging over short modified periodograms. IEEETransactions on Audio and Electroacoustics 15, 70–73.

a novel self-guided approach to alpha activity training alpha.pdf · a novel self-guided approach...

Documents