Hacker News Reader

Comment by writegit | Hacker News Reader

Super cool project and write up.

What's the state of EEG these days?

Last I interacted with one of these systems it seemed more like confimation bias from noise.

Had a buddy who bought a "high end" headset, shaved his head to "improve the signal", and it appeared for all intents and purposes that it was mostly only reading concusive activity. He would "show me it's working" by tapping on the exterior of the sensors to get it to display a spike.

Conceptually these systems "make sense" to me, ie. the brain uses electromagnetism to function so one should be able to sense/manipulate those vectors, but an FMRI is MASSIVE and requires a 1-3 Tesla electromagnet to get its fidelity, and even then is only measuring blood flow and correlating that to brain activity.

So what's the hope that a tiny sensor resting on your skin will actually correspond to anything happening inside the brain?

View on HN

We're building an eeg device for sleep, and work with a colleague who is on the Emotiv team, so I've got a bit of experience.

The state of BCI is improving, but one of the challenges is for many of the things we'd want to do, you need electrodes placed in areas that have hair. There are flexible sensors that can find their way through hair, but as somebody with thick curly hair, they are not fit for everyone yet. Then there is movement that also needs to be considered for many of these devices.

This is why something like Muse headband used for meditation is a good starting point. You are placing the electrodes on areas that don't have hair, and the use case is while being still, so you have consistent contact without movement.

At my start-up (https://soundmind.co) we're using forehead mounted electrodes which measure your brain waves during sleep, and we use sound to improve sleep performance. So like Muse (which you can also use during sleep, but does not yet provide stim), we benefit from measuring very small signals when you are mostly still.

I think if we were walking down the street with our headband trying to control cat ears, you'd get lots of EEG artifacts due to movement of the electrodes.

My 2 cents.

EEG and MRI are based on two completely different physical measurements. They have nothing whatsoever to do with each otyer, in terms of why they work. Medical EEG machines are very small, and their size relative to MRI machines does not factor into how effective the EEG is.

I mean, would you doubt that a stethescope can work, because it's so much smaller than an MRI? No, that would be ridculous. Same with your comparison vs EEGs.

Just read the WP articles on MRIs and EEGs. You'll understand it.

Is this written from the perspective of experience with contemporary EEG systems or wiki reading?

> They have nothing whatsoever to do with each otyer, in terms of why they work.

I thought my comment made my knowledge of this explicit? Also I was referring to FMRI.

>> and even then is only measuring blood flow and correlating that to brain activity.

My question is about resolution. FMRIs are our best tools in terms of brain activity when concerned about resolution.

A stethoscope can indirectly tell you if you have fluid in your lungs, but "where" or "why" are far beyond the scope of a stethoscope.

A direct comparison would be:

    stethoscope : is water in lungs? :: EEG : is brain in skull?

One seems useful; the other, self-evident.

So, EEG, is it /still/ confirmation bias in its ability to read/interpret signals in the brain?

Or has there been an appreciable development in EEG's abilities/resolution/functionality?

I'm not an expert. I did some undergraduate courses in neuroscience, and one of those covered how various instruments work, including MRI, EEG, and a few others. I would invite anyone with a better education to correct any errors they can identify in what I've said.

And yes I DID just refresh myself on those WP articles I mentioned :-) and they seemed pretty well written, to me.

I also owned a NeuroSky a while back, and IMHO it was not very useful... But that's because it was a toy, not a medical device. Same underlying measurement principles, but very different in terms of actual operation.

One of the main differences is that medical devices are always attached to bare skin with conductive gel applied under the sensor. Also, medical devices have more sensors. This vastly improves the signal quality, as compared to the toy devices.

In research work, medical EEGs have been successfully controlling computers for several decades, long before NeuroSky or the Necomini ears came to market.

Tl;Dr, you don't need an FMRI to control a computer... EEGs work fine, but none of the existing EEG toys have been particularly well designed in that regard.

"Confirmation bias" seems overly harsh. EEG does actually measure brain activity and it's certainly useful for some clinical/research tasks--if you don't mess it up (which is very easy to do). The trick is that you have to play to its strengths, which involve measuring large-scale activity patterns, like sleep (or epilepsy). If you want to measure something very localized, you need another technique (ideally with a hole in the skull).

In fact, I'd say that there is no overall "best" tool.

fMRI has the best (non-invasive) spatial resolution, but since it relies on blood flow, the temporal resolution is sluggish.

EEG has great temporal resolution, but even with fancy source-reconstruction techniques, the spatial resolution is very poor. It's certainly useful for some things, especially those related to global "state" factors. It's also very portable--if you can control EMI and movement artifacts.

MEG is something of dark horse: very good temporal resolution, and the spatial resolution often good--mostly. Since it relies on detecting magnetic fields, it cannot detect neural (electrical) activity that is radial to it. The other big drawback is that it required a large and expensive system with cryogenically-cooled superconducting detectors. However, the newer OPM detectors are cheaper and work at room temp, so more real-world things are possible.

fNIRS, PAT, and ultrasound seem like they might be good in some applications too.

EEG studies are rife with issues, but there are a number of very well replicated results. Obviously they do nothing to localize an effect, but there are absolutely clear correspondences between the readings of an EEG and at least some cognitive activity.

Will you get that nice FMRI resolution? Absolutely not. But the effects you do get are fascinating, and super weird.

Now that's the EEG I've seen in labs. Not sure if there is a similarly good commercial offering.

the underlying concept of EEG hasn't changed since the initial ones in 1929. Basically just electrodes trying to detect whatever LFP (aggregate electrical potentials) comes across the cranium. The clinical ones are considered a rough gauge of brain activity, good enough to reliably detect large seizures or a generalized slowing in patients who are delirious but that is about it.

There's MEG which tries to do the same thing but using magnetic detectors instead of scalp electrodes, but those systems are large and unwieldy

In surgery, they do intracranial subdural electrodes which give you better cortical LFP because the electrodes are almost directly on top of the brain and not having to measure through bone.

Then there are the single unit electrodes or things like the Black Rock systems Utah array where thin probes penetrate the brain into deeper structures but obviously have much more special circumstances where they are able to be used.

Reply on news.ycombinator.com