I want to be excited, but the blinking and chewing signals demonstrated in the paper are not true EEG signals, they are muscular artifacts. Based on my own experiments with OpenBCI[0], specifically the cython board and the ultracortex with spike electrodes, this paper does not demonstrate the most difficult part, measuring actual brain generated EEG signals.
I could not separate any signal out from the noise, neither with the provided software (which ostensibly filters and bins alpha, theta, beta, and gamma waves), nor while trying to roll my own filtering. That's not to say that this isn't good work, but I think the real difficulty is in the electrode hardware. I suspect that passive electrodes are unlikely to be very useful for the hobbyist, at least not at consumer level prices, which are still quite high. Maybe there was just something unlucky about my hair/scalp, but even the ostensibly simple closed eye alpha signal eluded my measurement attempts.
It's been a year or two, I haven't kept up if there have been developments, open to suggestions! I don't trust myself to build active electrodes, a reasonably priced kit would be awesome.
Re: muscular artefacts, can anyone elucidate the nature of "pre-speech signals" captured by electrodes along the jaw? There was a MIT student a few years ago who built a headset shaped device that could hear his thoughts because it picked up his "inner voice" via electrodes on his face. My understanding is that the signals are the same as those produced in speech, but weaker (below the threshold of actually activating the muscle).
>"The headset-like device, AlterEgo, is a sensory and auditory feedback system that uses neuromuscular signals from the brain’s speech system to extract speech. When we talk to ourselves internally, our brain transmits electrical signals to the vocal cords and internal muscles involved in speech production. With AlterEgo, an artificial intelligence agent is able to make sense of these signals and prepare a response. The user can hear the AI agent’s responses through vibrations in the skull and inner ear, thus making the process entirely internal. The AI agent can also send the information to a computer, to help an individual with a speech disability communicate in real-time."
Ah you're right, they measure the closed eye alpha signal. I'm just skimming with ctl+f and didn't see it, did they mention specifically what electrodes they used for their measurements?
Idk, but the signal doesn't look like an alpha wave you'd get from a good EEG. The video on their repository doesn't seem to be using the normal electrode set, nor use the common procedure of applying the electrodes.
Hey HN,
Sebastian here, one of the co-authors of the paper. Let me know if you have any questions! We'll be launching on Crowd Supply with the device soon, so you can buy it if you want. We have other devices in the pipeline as well. Check out https://www.hackerbci.com/
Everything says €0.00. How much do you actually have to spend to get the necessary electrodes, instrumentation amps, etc.? The raspberry pi seems like the simplest part. Thanks.
Thanks. I came across this a while back, that also looks interesting: a "poor nerd's fMRI" that uses infrared imaging instead of magnetic fields. It would be cool to have something like that too.
This device can be easily used for machine learning enthusiasts to create projects for controlling robots and mechanical limbs using the power of thought
I wonder how this compares to OpenBCI. I was very close to dropping the money on an electrode cap and the OpenBCI hardware but maybe it was good I didn't if this ends up being more affordable. I wonder how the resolution compares - I remember reading that OpenBCI needed some tweaking to achieve a 250hz sample rate.
It's not clear to me it could be very different cost-wise, it's going to rely on one of a limited number of bio-amplifier chips.
To be fair OpenBCI have changed business models a bit; if you're a hobbyist you might go AliExpress but a researcher with a genuine budget will probably pay the OpenBCI "premium" for the same hardware.
The other aspect not discussed is that you really want it to be battery powered and wireless if you can (safety reasons), and a raspberry pi device is going to require a lot more battery or have much less runtime.
A year ago James Burton tried out some EEG devices for his project [1], and it seemed like noise filtering (funny enough he had to filter the AC frequency commonly used in UK grid) and things such as "skin moisture" affected the electrode outputs quite a bit.
Is this still the case when using EEG devices, including this implementation?
A couple years ago I worked on a piece of EEG software. It was super cool, but also deeply dampened my optimism for BCI tech. If EEG (a very serious and expensive technology with tons of smart people working on it) can’t do much better than getting pretty basic electrical signals from the brain, I don’t see how any consumer priced device could allow you to actually “control things with your thoughts”.
Basically, an EEG picking up brainwaves is like standing on one side of a canyon and listening for whispers from your friend on the other side, and occasionally a caravan of trucks drive through. The brains electrical signal is super weak, and you have to pick up what you can from the other side of the skull by comparing voltages in different areas. The thing is, if someone raises their eyebrows, that electrical muscle impulse is enough to blow out whatever signal you were seeing. Skin moisture, the entropic and decaying nature of the universe, all that stuff is fighting against you. It’s a miracle that EEG works as well as it does for the medical purposes where it frequently provides life-saving information.
Looking at the photo of the circuit board I can't see any sign this is opto isolated from the RPI's power source. What safety precautions are being taken?
This is cool, but to be usable for serious research (i.e. peer reviewed publication) it has to be compared to a gold standard device to make sure the data is valid.
Can this be used for sleep tracking?
How precise do the measurements need to be to give useful data about one’s sleep? (e.g. how much deep sleep one had.)
It looks like the sample rate is more than high enough to differentiate between the typical alpha, theta, and delta waves, which should be enough to discern sleep stages. The hard part is a quality electrode-scalp contact which maintains relatively constant impedance over the course of sleep, especially after movement. EEG is all about signal/noise, because ultimately you are amplifying tiny currents which are easily drowned out by noise from contact movement and I think even RF interference (you can trivially pick up 60hz mains noise even if you're sitting ≈10 feet from the nearest powered appliance/wire).
OpenBCI sells electrode caps which they claim are sufficient for sleep studies, unfortunately IIRC they are somewhat inconvenient because the electrodes require wetting and/or gel which must be applied to each electrode every time the cap is worn. I have yet to shell out the $500+ to give it a try, but that's still way cheaper than medical grade devices.
If you want sleep tracking, you can use Muse S [0]. You can stream EEG readings in real-time using [1].
However it's really, really hard to get _useful_ data about one's sleep. The results computed by Muse are garbage. The only successful product I know is Dreem [2] but they've exited the consumer market.
At the moment, I think the easiest way is to get an ECG (yep, for the heart, not the head) and process the readings using z3score-hrv [3].
I could not separate any signal out from the noise, neither with the provided software (which ostensibly filters and bins alpha, theta, beta, and gamma waves), nor while trying to roll my own filtering. That's not to say that this isn't good work, but I think the real difficulty is in the electrode hardware. I suspect that passive electrodes are unlikely to be very useful for the hobbyist, at least not at consumer level prices, which are still quite high. Maybe there was just something unlucky about my hair/scalp, but even the ostensibly simple closed eye alpha signal eluded my measurement attempts.
It's been a year or two, I haven't kept up if there have been developments, open to suggestions! I don't trust myself to build active electrodes, a reasonably priced kit would be awesome.
0. https://openbci.com/