EEG Neurofeedback Sonification literature 3.7.
3.7.9. van Boxtel,
Probably, the most rigorous and methodologically-sound EEG neurofeedback sonification study was reported by Geert J.M. van Boxtel, (van Boxtel et al., 2012). In this randomised, double-blind, „Between Subject‟ placebo-controlled study, 50 healthy participants received 15 sessions of one of three interventions. One group received auditory alpha activity training (N=18), a second group received random beta training (N=12), and the third did not receive any training at all (N=20).
The subjects were able to listen to music of their own choice, and in the two training groups the EEG band power recorded from the sensorimotor strip in the centre of the head would affect the sound quality of the music. Thus, in the alpha group as the alpha power decreased, the quality of the music would decrease proportionately in real-time. In the active control group a different Beta band would be selected randomly every session in order to inhibit any learning effects. In the passive control group subjects would have their EEG recorded while listening to their music, but there would be no change in the sound quality.
In a comprehensive test battery of pre- and post-measures of 26 channel EEG and Event-related potentials, as well as mood rating scales, quality of life inventories, sleep rating and guided interviews, van Boxtel et al., were able to show that only the real alpha training group could increase their EEG alpha activity by 10%, and that the increase “remained evident 3 months after the last training session”. In an exit interview, the alpha training group did feel more relaxed, but despite showing trends in the right direction no statistically
authors suggested that this was due to the lack of statistical power because the group sizes were too small.
Although this was an excellently designed and controlled study, there are three main issues that could explain the failure to show a behavioural change with the EEG training. First, the authors suggest that modern life is stressful and point out that 40 million workers in Europe suffer from the negative effects of stress. (Europe has a population of around 740 million people, so 40 million would be around 5.4% suffering from stress). The 50 subjects were recruited from a website, and there was no suggestion they came from a stressed population. This would mean than on average there could be 3 subjects suffering from stress-related conditions, and the majority of subjects would not be particularly stressed or likely to receive significant benefit behaviourally from general alpha enhancement training.
The second issue is that the subjects were allowed to listen to their own choice of music, without control for the genre and style of the music. Given that the participants were not informed of the purpose of the study, it is quite possible that they were listening to arousing music. This is supported by the fact that, despite numerous studies showing the relaxing effect of listening to relaxing music, in this study even the music-only group failed to show significant increases in relaxation.
Of course it could be claimed that the behavioural measures used in this study were capturing trait properties of the subjects that should not be expected to change over a short period of time. Or that the measures were just not sensitive enough to any change that may have happened.
But perhaps the most significant difference between this study and the others mentioned above is that the subjects were not given any instructions or even told their EEG could control the sound quality; they were only asked to “sit back and relax”. The authors suggested that, as the participants were listening to music that they knew well, and the reduction in sound quality was very obvious, the task was a “very intuitive feedback mechanism”. This could be correct, as with this very simple sonification method the training group did make lasting changes in their alpha levels that remained for 3 months. But the lack of explicit instruction may account for the lack of behavioural change.
One explanation for this may be that, despite the human auditory system‟s incredible ability to continually trace or alert to sound, it also has the ability to block unwanted or irrelevant sounds. Maybe precisely because the subjects were familiar with the music they were listening to, they were able to ignore the distortion.
This highlights an interesting and potentially critical distinction in the neurofeedback domain, that between explicit and implicit instruction. In most clinical neurofeedback training, a great deal of explanation is given to the subjects about the procedure, and they are encouraged to focus explicitly on trying to manipulate the EEG activity and how they feel as they change the EEG parameter. But in studies that use implicit neurofeedback, it is not so clear what the learning mechanisms would be.
3.7.10.
Hardt, 2012
37 years after his first paper on alpha neurofeedback, a pioneer in the field James V. Hardt, (2012) reported on a study of 40 adults undergoing an intense alpha neurofeedback training procedure, which consisted of daily sessions of between 76 and 120 minutes for 7 consecutive days.
The alpha amplitude of four channels (O1, O2, C3 and C4) controlled the amplitude of four tones (400-800 Hz) (i.e. AM) played on four different speakers in four different spatial locations.
The training protocol consisted of both alpha enhancement and alpha suppression. “For epochs of 2 minutes at a time, the trainees sat in the dark with their eyes closed listening to their feedback tones wax and wane based on the strength of the filtered EEG signals. Then a “ding” sounded and the tones stopped. For the next 8 seconds, the monitor displayed color-coded numerical feedback of their alpha brain-wave integrated amplitude...”
In order to capture any “positive psychological results by reducing anxiety and other psychopathology” a pre and post battery of four well known measures was recorded, the Minnesota Multi-Phasic Personality Inventory, the trait forms of the Multiple Affect Adjective Check List, Clyde Mood Scale, and Profile of Mood States.
This short but unclear paper highlights the differences between research done to validate a commercial clinical intervention and a pure experimental research design. There was no control group included, in this „Non-blinded‟ „Within Subject‟ design and very little information was given about the protocol
measures. Somewhat oddly, Information was given on the length and material of the electrodes but not on why the electrode locations were chosen or the technical details of the EEG recordings.
This was a time consuming study where each trainee spent 10-12 hours at the training centre each day for 7 consecutive days and did neurofeedback training for around 9 to 14 hours.
But still, despite these short comings the majority of psychometric measures did show a highly statistically significant improvement.
3.7.11.
Wang, 2013
In a two part study, Sheng Wang, Yan Zhao, Sijuan Chen, Guiping Lin, Peng Sun and Tinghuai Wang (2013), first selected 24 high and 24 low trait anxiety participants from a pool of 358 undergraduate students using the Chinese version of the State-Trait Anxiety Inventory. They then recorded event related potentials (ERP) of the 48 participants while they performed an Emotional Stroop task.
The Emotional Stroop task is a variation of the well-known colour Stroop task, were the word for a colour and the colour of the “ink” the word is written in can be either, congruent (i.e. the word “red” in red ink) or incongruent (i.e. the word “red” in blue ink) In different trials participants are instructed to push a button to identify the colour of either the ink or the word. Many studies have shown that participant‟s reaction time increases in the incongruent trials and this is known as the Stroop effect. In the Emotional Stroop task, participants must identify the colour of the text of three different emotional categories of words (negative,
a negative bias and focus more on negative stimuli and therefore have a slower reaction to the negative words in the Stroop task.
Event related potentials of the brain are computed by recording the electrical response of the brain to hundreds of trials and then averaging the trials to cancel out the background noise of brain processing that is not related to the task. This leaves the brain response specifically associated with identifying the stimuli and responding to the task. ERPs have excellent temporal resolution and look at brain activity in time windows of around 500 milliseconds. The ERP Brain response to stimuli is characterised by positive and negative fluctuations at different times in relation to a baseline period just before the stimuli. So for example a well-known ERP is called the P300 and is a positive deflection at around 300 milliseconds after a stimulus is presented to a person.
Wang et al. showed that in the high trait anxiety participants only, there was a significant main effect of increased reaction time to negative words. The ERPs also showed longer latencies and increased amplitudes for the P300.
In the second „Single-blinded‟, „Between Subject‟ study design, the 24 high trait anxiety participants were randomly assigned into one of two groups; ether EEG feedback group (n. =12), or sham feedback group (n. =11).
The training consisted of one continuous 27 min session, twice a week for a total of 15 sessions. The feedback was measured from C3 or C4 and the alpha activity (8–13Hz) varied the volume of an „ocean waves‟ sound also when the alpha was over a threshold set to a range of 0.7 to 1.5 times the baseline average, a „warble‟ sound played.
Again, no training or pre vs. post EEG data was reported but more meaningfully there was a significant reduction in the P300 latencies and also reduction in reaction time for negative words on the emotional Stroop test for the neurofeedback but not the sham group.
3.7.12.
Ramirez, 2015
In a „Non-blinded‟, „Within Subject‟ pilot study Rafael Ramirez, Manel Palencia- Lefler, Sergio Giraldo and Zacharias Vamvakousis (2015) trained a multi-channel EEG protocol intended to increase both arousal and valence in an elderly depressed population. Arousal was calculated as the frontal beta to alpha ratio (i.e. (beta of F3 + F4 + AF3 + AF4) / (alpha of F3 + F4 + AF3 + AF4)) and Valence as frontal Alpha asymmetry (F4 alpha - F3 alpha). The EEG was collected with a cheap consumer devise and custom software.
In 10 sessions (2 per week) of 15min each, participants chose a set of 5 or 6 music pieces and their arousal and valence would affect the loudness and tempo of the notes of the music, to make it sound "happier" as they moved towards a more positive mood.
Ramirez explained: “The system consisted of a real-time feedback loop in which the brain activity of participants was processed to estimate their emotional state, which in turn was used to control an expressive rendition of the music piece. The user's EEG activity is mapped into a coordinate in the arousal- valence space that is fed to a pre-trained expressive music model in order to trigger appropriate expressive transformations to a given music piece (audio or MIDI).”
The Beck Depression Inventory was used as the main pre vs. post measure of change and was claimed to show a statistically significant reduction in depression. But the data was only of 5 people and not a lot of confidence can be given to these claims. Arousal and valence scores are given of the session data, but with the information given it is difficult to make any conclusions.
The sonification system looks very interesting but insufficient evidence of its utility was presented. Clearly they were working with a difficult population with many health issues; however the minimum expected data from a pilot study of this nature would be some user feedback to establish if this elderly depressed population enjoyed the music manipulation or found that having their favourite tune that they have known and loved for years tampered with was disconcerting.