Wearable device for noninvasive optical brain imaging | Hacker News Reader

Wearable device for noninvasive optical brain imaging

spie.org

59 pointsgeox13 comments

I guess that's more practical than what another BCI company wants to do, e.g., implant 10k's of nanoscale sensors directly in the brain and eventually establish a read-write interface.

That's really amazing. 2kg still seems pretty heavy though, more than a typical motorcycle helmet

I don't understand the breakthrough here, they miniaturized fNIRS?

'Tis also my read on the technology. They originally intended to miniaturize MEGs, which would have been remarkable, but diverted to the more immediately feasible. With $100m in personal funding, I felt MEG was absolutely the right route.

I'm let down a bit by the recent marketing as well - when thinking optical sensing for neurology, you really think optogenetics like applications.

Isn't that a bit apples v oranges? I've had a MEG and I'm floored by the technology but would a better analog be a PET? We're looking for two different things, metabolism v saturation but it seems like they're both in the physical or structural realm than the electrical.

One of the main advantages of TD-NIRS is that the signal it's imaging is "electrical".

Modalities like PET, BOLD fMRI, and CW-NIRS do depend upon saturation changes. For BOLD and CW-NIRS, it's the change in blood oxygen saturation.

TD-NIRS images the fast optical signal that is correlated with electrical activity in the cortex. MEG images the magnetic fields correlated with electrical activity in the cortex. IMO, they're pretty similar.

> TD-NIRS images the fast optical signal that is correlated with electrical activity in the cortex. MEG images the magnetic fields correlated with electrical activity in the cortex. IMO, they're pretty similar.

So I guess the way I look at it, is that the field is what is augmented in a MEG, that makes the substrate emit an amplified signal naturally there. While a TD-NIRS injects a laser signal into the skull to measure back reflections... both inject energy into the system, but one is a reflection vs. an emission.

Is this the wrong way to look at it?

And anodyne33, yes, it is 100% apples to oranges. I was abreast of their original aims and MEG was on the table at the time - I had hoped that would remain the aim as it to me appears more beneficial, based on how I view the sensing above.

Not a breakthrough. This technique has been known about for at least two decades.

Most fNIRS uses the amplitude-based, continuous-wave modality to compare chromophore concentrations resulting from thermovascular coupling.

This uses time-domain based. What this means more formally is that it uses the impulse response created from a fast optical imaging source to then detect scattering changes in the cortex that ideally correspond to neuronal activation (or lack thereof).

I was actually working on a very similar device a few months ago. I had to give up as the chip shortage made the specialty ICs required to pull this off damn near impossible to buy.

There are a couple of things that make TD-NIRS a bit trickier. First off, it relies upon counting photons. This makes it susceptible to all sorts of noise, coupled with the fact that you need a photodetector with a very fast rise time and at least 10-20% detection of incident photons upon the detector.

Benefits - Extremely fast (millisecond-range) neuronal activity detection - Less susceptible to motion artifacts - Very localized detection, scattering is well-modeled

Drawbacks - Requires extremely fast sampling rate - Above sampling rate makes multiplexing difficult - Still susceptible to all kinds of noise

Mary Lou Jepsen's Open Water was looking at something similar -

I'm going to say something stupid simple: Any technique imaging the brain outside the skull is hard. Much of these IR technologies are noble in terms of their general science and engineering learnings, but in terms of practicality, sub-optimal.

Curious to know if you've experimented with other modalities? My base is fNIR and EEG device manufacturing, while just being exposed to (f)MRIs, MEGs and the like.

I currently do research in MRI.

I'm not entirely sure what you mean about IR technologies. Almost all medical imaging done today is done outside the skull. The only exception is ECoG, which is only medically used for patients with severe epilepsy. This is because open-brain surgery is an extraordinarily risky and expensive proposition.

Every single imaging modality has strengths and weaknesses. It is the goal of the physician, and of the radiologist, to choose the appropriate imaging modality for the patient.

NIRS is not always the best choice, especially not for medical imaging. But it's a good choice if you are looking for a portable modality that can image neuronal activation in the cortex.

EEG is already difficult because you can't just add probes to increase spatial resolution. There is a fundamental limit the information that can be reliably gathered solely based upon the sodium-ion voltage potentials of neurons.

I love these technologies, thanks for engaging with me about them.

IR roughly meant NIRS - was just playing fast and loose with the modality for 'things that suffer from extreme scatter that I believe will prove impossible (for the intended use) for the foreseeable future'

MRIs as I understand it align then flip water molecules that are a super easy signal to read, it comes down to stitching techniques and the hardware affordances to make that easier.

So, now I'm more curious, you believe you solved the scatter issue but were supply chain constrained?

Your assessment is also mine: spatial and temporal resolution requirements drive which sensing techniques one would use.

My aims were consumer based, so much less concerned about precision, more analysis turn around time. EEG was torture, yet rewarding in that realm - though to many techie's dismay, it only offered a single bit of resolution > I still think that's enough.

Edit: If it's still not clear what I mean (I am not a grad student in any respect), I think that anything one has to 'inject' then read the reflection is a suboptimal approach to reading the naturally emitted (albeit amplified) signal.

Why not try a micro approach, 1024 pin sized probes that make an end point connection to the scalp. Most EEGs I have seen have probes the size of a coin. Is there a reason for this? From what I know of electronics the size of the electrode only matters if you are passing high amounts of current through them.

I don't think kernel has demonstrated fast optical imaging of direct neuronal activity here. The article clearly mentions they detect oxygen activity corresponding, like conventional fNIRS or fMRI.

I've been looking at this field and my conclusion was that non-invasive optical imaging of direct neuronal activity, while possible in theory, it requires several magnitudes of improvement in today's technology. Even Openwater is detecting blood flow (and not individual neurons). Wrote my thoughts here: https://notes.invertedpassion.com/Consciousness/Fast+optical...

Curious to hear what you were building and whether you actually got close to doing fast optical imaging. Happy to chat offline, my email is here: https://invertedpassion.com/about/

Reply on news.ycombinator.com