Neurofeedback training and its purported mental benefits is a money making business and therefore information about it is flooded in inflated claims of its benefit and new age spiritual BS. I’ve just started looking at the published research, at initial reviews reveal uncontrolled studies and lots preaching to the neurofeedback training choir.
Does anyone know what empirical claims can be made about the mental effects of neurofeedback training? For example, “neurofeedback training results in short term increases in attention span”. Or maybe not even tied to “benefits” but more objective correlations in measurements like “long term neurofeedback training correlates with persistent coherence in subjects’ alpha signals from EEG”.
Neurofeedback is basically just the use of a feedback signal (a metric) as means towards a conditioning procedure: desirable brain signals (as indicated by your metric) are rewarded by positive reinforcement.
Yes, there are very interesting and worthwhile results, but it all critically depends on a) how the metric you’re using is derived from the raw data (what you give feedback about) and b) how you use the feedback signal in your intervention to achieve systemic changes.
There is some solid evidence for EEG neurofeedback for ADD (treatment for ADHD has lower effect sizes but is still worthwhile), check out this meta-analysis: http://ashevillebrainbalance.com/wp-content/uploads/2012/02/Efficacy-Neurofeedback-TreatmentADHD1.pdf
Anyhow, the standard metrics (algorithms) for neurofeedback are all rather easy & boring: eg. band power within fixed bands (eg beta/theta) are log transformed and a ratio is calculated. Ratio is then thresholded to use as on/off signal for feedback signal (eg. party obstructing view of video sequence, playing music etc).
There are more interesting feedback metrics currently being explored…
Then it gets really interesting if you use fMRI instead of EEG, but this’ll stay out of the realm of QS and a lot of it is still rather unexplored due to the high cost.
Success depends on whether you find a signal in your data that readily identifies whether you’re in a desirable state or not. This is easy for attention related processes, but harder for things related to your emotional state and near impossible for the absolute ‘content’ of your thoughts.
Please do keep us posted on your progress and feel free to reach out to me personally for if a public forum is too open for sensitive topics you might want to discuss!
So I’m trying something new. The goal of biofeedback is training the ability to consciously intervene in one’s physiology. I am looking at machine learning research on EEG in the context of brain computer interface. Through this lens, I hope to see what conscious mental tasks people can actually model from EEG signals with decent prediction performance. From there, I hope to look at the fMRI literature to see if repeated use of a brain computer interface application left some persistent effect that looks you can call the results of “mental training”.
I don’t think there is any question that most people can (with a bit of practice) consciously change their heart rate and brainwaves (to some extent). So the question should be how well different training methods work, right? Some methods might use some kind of biofeedback, others just plain meditation etc.
I disagree. Heart rate and HRV I leave aside. As for neurofeedback, while there may be short term changes, the question is whether there are long term cognitive trait changes. While in the neurophysio/cog-psy literature one can find claims of changes in individuals over time (usually measured in weeks), I haven’t found anything that attempts to classify the EEG/fMRI profiles from long-term (years?) meditators in non-meditative rest state from non-meditators. On the other hand, there is machine learning research for classifying EEG profiles of ADHD subjects from non-ADHD subjects. If you are doing neurofeedback to train your mind to some end, like some level of meditative ability, then these kinds of distinctions between individuals are what really matter.
Also, one wants to be able to distinguish the effects of neurofeedback from the effects of just doing an attention exercise. This recent study shows an example where kids with ADHD doing true neurofeedback and kids doing the same exercise but getting fake feedback both improved in attention measures over time, but differences between the groups were not significant. The implication is your $200 neurofeedback device is works no better then sitting down and doing a Sudoku every day.
This study even more discouraging – implies there are no benefits of neurofeedback on ADHD kids when they take Ritilin.
[quote=“Robert_Ness, post:6, topic:792”]
Also, one wants to be able to distinguish the effects of neurofeedback from the effects of just doing an attention exercise.[/quote]
I’m not sure “the effects of neurofeedback” is something that can be isolated. One can compare the effectiveness of specific exercises, and it’s quite possible that those that happen to use some form of neurofeedback turn out to be no more effective than those that don’t. But that doesn’t tell us much about neurofeedback in general (except perhaps that it’s no silver bullet).
Neurofeedback is the follow-up to the biofeedback fiasco of the 80s. Nice story and narrative, but reliable results that have any impact are lacking. Perhaps we simply have not found yet the right features…
I agree. Here is a good read about all the problems entailed in modeling EEG, such as the assumption the signal is linear (which may be a fair assumption in practice for a short term neurofeedback session, I don’t know). I suspect most building commercial or consumer EEG neurofeedback devices earn their chops using Matlab’s EEGlab. Anyone know much about this tool set and how it may or may not be biasing feature selection and the types of models applied?
Yes I used EEGlab for most of my data analysis. It’s a third party open source plugin and not really used for neurofeedback. You’d want to use EEGlab if you analyse data offline especially for event related averaging and independent component analysis (used for cleaning data with many channels from noise, separating the data into maximally independent signals). Feel free to send me an email if you have any questions regarding analysis of your own data in EEGlab.
For BCI applications (and neurofeedback probably), I can recommend BrainBay (easiest software there is), BCI2000(http://www.bci2000.org/) or OpenVibe(http://openvibe.inria.fr/). They all support a wide variety of devices and can be very flexible but vary in how easy they are to get set up. Try BrainBay first if you want to get a feel for EEG data manipulation online, try BCI2000 or OpenVibe if you want to dig deeper. BCI2000 is really aimed at BCI applications, though.
Hi Martin,
So you do your own signal processing on these software packages or does it select any features for you, or both? Are they all PC focused?
What kind of hardware do you recommend?
Cheers,
Rboert
[quote=“Robert_Ness, post:11, topic:792”]
So you do your own signal processing on these software packages or does it select any features for you, or both? [/quote]
Yes, everything is “hand-crafted” to the specific research question, but there are a couple of examples that you can use and modify on the respective websites. I recommend starting this ways.
Matlab runs on all platforms, but the other programs are limited to PCs I think.
I was recently asked the same question by a QS friend, Yoni, and this was my answer:
I can’t say much about NeuroSky it seems like they removed all but the most recent products from their store, maybe their more recent devices are better.
I’ve also pre-ordered the Muse, but it’ll take a while until it’s there.
EPOC is a decent device, but you might want to modify the device in order for it to give good signals with all electrodes. There are some nice libraries to work with Android.
What I’m a big fan of is are the** truesense devices**. (op-innovations.com) At least with a previous version of the device you had to be very careful to not completely drain the batteries, (inherent drawback of otherwise great Li-Ion batteries) but I think this is fixed in the more recent versions. You get 11-16hrs of runtime, rawdata (EDF+ format!) from 1ch EXG, 3 axis accelerometer, plus a remote to set markers. For 45 USD per device. That’s unmatched and will remain so for a good while. As far as I can tell, data quality is decent enough. You really get only 1 channel on the forehead, but you can wear multiple at a time and combine datasets. I haven’t tried measuring through hair, but I’m sceptical it’ll work well. If you need a cheap device that works, and you’re comfortable with handling raw data, buy 5 of those rather than a single EPOC/Muse/Melon what have you.
I’ve just received an email stating that the first units are shipping. To be honest, after watching the market closely for the last eight (!) years, I’m thrilled if working products are shipped at some point.
In my opinion, for neurofeedback (apart from EEG) there is no need for more than 2 channels (really for beta/alpha/… you need just one, and 2 you need for hemishperes syncronicity training). So most of the hardware described here is too excessive.
Choosing of the hardware is first of all choosing sensors and headband because they shall be comfortable and give good signal. Also BT4.0 interaction with smartphone is a great value.
Another point is dry electrodes vs. wet electrodes. Wet electrodes are not convenient to use, but dry electrodes may not give good performance. The future IMHO is for dry elctrodes.
Another good point is open documentation for software developing.
As a bottom line there seems no really good home “amateur” neurofeedack devices right now, perhaps Muse which is new and no reviews yet. Also waiting for Melon.
I would probably class my view as cautious rather than skeptic; here’s my two-cents if your interested.
My background is in designing biofeedback technologies (e.g. games that adapt play according to a given psychological state such as mental workload). I would agree that studies on clinical biofeedback, to which you are referring, where biofeedback techniques are used to bring about beneficial long term outcomes (e.g. reduced stress or increased attention) can be somewhat problematic to parse and find meaningful results (e.g. suitable controls, follow-up studies).
A good starting point for the clinical efficacy of biofeedback treatments is the “Evidence-based practice in biofeedback and neurofeedback” review of 2008 by the Association for Applied Psychophysiology and Biofeedback (AAPB) which provides a break-down of treatments and their current efficacy. Currently their is only one medical condition for which biofeedback was the superior choice by comparison to control and alternative treatments and it was not a problem treated by neurofeedback.In 2005 their was a meta-review of neurofeedback treatments for ADHD (“Electroencephalographic Biofeedback in the Treatment of Attention-Deficit/Hyperactivity Disorder”). This review concluded that while neurofeedback has a positive impact the design of such treatments needs improving to be of benefit.
From the literature I’ve read, clinical neurofeedback has its uses. The physiological mechanic certainly works (i.e. user’s ability to manipulate a signal). However more studies are needed investigating its long term impact. I would certainly agree that some commercial neurofeedback devices make somewhat questionable claims, especially given the technologies association with the new-age movement which can be a little fluffy on performance.
Would you please explain these modifications you mentioned for the EPOC?
Regarding Neurofeedback:
I have both EPOC and Neurosky MindWave Mobile. Played around with several of the apps in Neurosky and Apple store that works with the MindWave Mobile headset. Any ideas where I can find algorithms on how these apps work? For example: http://www.home-of-attention.com/en
This app says it can train my mind for Relaxation, Attentiveness, Attention, Concentration. You can connect Neurosky headset, play an audio and it will play/pause the audio depending on the measurements. What I don’t understand is: how does the app makes that decision? Is there any papers that says adjusting the audio feedback in certain way will benefit relaxation or attention. Is it a simple adjustment like:
measure attention/relaxation from the headset
if too low pause the audio
measure again
if still too low, reduce the expectation
measure again
if high enough play the audio and keep the new expectation
An aspy friend who I no longer hold high opinion of (since I think he uses being “aspy” as an excuse from treating people poorly) pressured me into doing neurofeedback (he also pressured people around him often to do weird things).
Overall, it seemed like an experience meant for those that buy into general quackery. I think there is so much we don’t understand about the body, and the practice of psychology still remains to be an artform or statistics at best. Anything beyond this descends into manipulative practice (and probably malpractice in a court of law).
When I was hooked up, most of my “channels” were static (not fluctuating greatly). I was very high in theta waves which I was told was good. This seemed not unlike what I watched in a scientology documentary. Beware of cultists with PhD’s.
