Researchers design wearable tech that can sense glucose levels more accurately
uwaterloo.ca
uwaterloo.ca
Type 1 diabetic here - for what it's worth, CGMs aren't particularly invasive. At least in comparison to the many many years of finger pricking! But a smart watch solution would be cool. (I actually do get my CGM readings on my smart watch, which is really nice!)
I know Apple has also worked on this stuff in the past, but from what I remember the accuracy wasn't good enough to be safe for diabetics. I'd be really curious to see accuracy stats on this in comparison to Dexcom and Freestyle CGMs.
I would definitely be excited to use something like this, but for me, the biggest quality of life improvements for me will be continued improvements with closed loop CGM + insulin pump systems.
Until then, Freestyle with Omnipod Dash in a close loop with iAPS was a game changer for me: Almost no peaks anymore, HBA1c on the level of a non diabetic person…
Nevertheless, good luck in productising it and I’ll be certainly trying it once it’s available…
In my experience, the quality control isn't very good (some patches will read much more accurately than others) and accuracy isn't that good when you get out of normal ranges.
I don't think the "invasive" nature of the Freestyle is a problem at all, but it would be nice to see some innovation on either the cost or the accuracy or both.
In no way would I describe CGM as solved, and this would go a long way towards filling many of the gaps, especially in younger / older / less compliant patient populations.
Main points are (fsl2 based):
- latency, currently 10min.
- accuracy, fine in normal range, but when you have a low blood sugar suddenly the latency spikes a lot.
- values when being under the shower too high.
- start up time of 60min could be lower.
-open up the hardware for any app to read
Besides, 10% off doesn’t often matter:
At 0-70 mg/dl the pump should suspend insulin either way. At 110-600 mg/dl the pump should ensure enough IoB by bolus, increase basal, and monitor either way. In that 70-110mg/dl the 10% MARD kind of matters for clinical decision-making, but not much. 90mg/dl is about as healthy as 81 and 99.
Patients are sometimes fussy by this inaccuracy but forget the tremendous benefit of trend indicators, let alone closed loop systems. Both of these have a much much larger positive impact to health than blood glucose being 10% above or below target impacts health negatively.
CGM with <=10% MARD, whether in wrist form factor or sensor form, is good enough for treatment. Of course, same as most readers, I have my doubts about this article.
Indeed though, advancements in (affordable) closed loop tech matter more than where the CGM is worn.
Oil slick is a skin safe tar remover and searching for that using a pummice stone was reccomended on reddit.
But also in times where we have the Libre 3 which is so tiny that you legit don’t even notice it, a CGM on your wrist is not worth the loss of accuracy for T1 I guess (assuming your insurance pays for it).
You might find this interesting: "A bi-hormonal fully closed loop system"
Disclaimer: not a doctor or biologist
In the article, the researcher claims "No other technology can provide this level of precision without direct contact with the bloodstream", so it sounds like they're claiming it's better than existing CGMs in a way that might be clinically relevant. Not sure if that's plausible or whether they are directly measuring blood glucose rather than interstitial.
Even at the size of a brick, or without conveniently hiding the power supply off-camera, forgoing needles would still be a huge boon to diabetics. Why not get the concept working and demo some hard stats, then miniaturize?
To clarify, the actual science they did is interesting (to me at least, as someone not in the field). The paper is linked here: https://www.nature.com/articles/s44172-024-00194-4
What is bullshit is the completely unwarranted conclusions in the title or in the quotes in the article. This is classic "science by press release".
If you notice in the paper, they didn't do any testing, at all, with actual humans (or animals) and their blood sugar levels. The paper is mainly about the design of this "metasurface" which they claim allows higher resolution and sensitivity of a millimeter-wave radar system. The leap from what they've done to "no more needles for diabetics" is about 100x of "draw the rest of the owl".
Again, to emphasize, I'm not denigrating the science they've done. I'm denigrating the hyping of it.
They did mention in the article that clinical trials are on-going.
> “We have a minimum viable product that’s already being used in clinical trials, and while there’s more work to be done, we’re much closer to a full marketable device,” Shaker said.
Absolutely no information about what this "clinical trial" entails, or what phase it was. Most importantly, to get an initial assessment of the accuracy of the device, no clinical trials are necessary - you simply need to do a test that compares the blood sugar reading from the device against the current gold standard, most likely first in some animal model.
If their device was really as far along as the title and quotes are implying, they would be showered with so much money it would make the Theranos peak valuation look small. The only evidence they've provided (which, again, I'm not saying is insignificant) is that the "metasurface" they have developed enhances the resolution and sensitivity of a radar system against a beaker of water.
Theranos also had an MVP in this sense :)
One possibility is that they want to sell this technology to a big company without publicly disclosing all their trade secrets. However, this research could have been sponsored by a public grant, which would have compelled them to share some information. Therefore, they published a paper that appears more like a patent application than a research paper with solid data. It’s still noteworthy that it was published in Nature.
FWIW, it was not published in 'Nature' but in 'Communications Engineering', a journal by Nature Portfolio (formerly known as Nature Publishing Group, part of Springer Nature). It is a new Open Access journal, established only in 2022. Given the track record of their 'Scientific Reports' journal [1], I would be rather cautious regarding the quality of the works published at 'Communications Engineering'.
IMHO, Nature Portfolio is doing their 'Nature' journal a disservice by hosting all of their journals at nature.com. I guess this is intentional, letting their less prestigious journals profit from Nature's prominence.
[1] https://en.wikipedia.org/wiki/Scientific_Reports#Controversi...
https://en.wikipedia.org/wiki/Noninvasive_glucose_monitor#Ne...
(different technique)
This is not my field of expertise, and maybe I am misunderstanding the papers. But it seems that there is little evidence that non-invasive glucose monitoring via measuring dielectric properties works reliably in practice. No in-the-wild studies, no investigation of potentially confounding factors.
Take for example citation 22 from the paper. A study where the authors propose a new antenna design. They seem to measure how the pancreas changes size during insulin production by monitoring its dielectric properties. IIUC, they look for a dip in the frequency spectrum caused by absorption of a certain frequency band.
But their measurements show an even larger effect when measuring on the thumb instead of the pancreas. This effect is not explained at all. (My guess: after having patients fast for 8-10 hours, giving them glucose will have an effect on the whole metabolism, resulting in higher blood flow, and that's what they measured).
Also, while they operate the antenna in the GHz range, they use a cheap USB soundcard (sampling rate 44.1 kHz) for capturing the signal. I did not understand this at all. They also repeatedly use the term "dielectric radiation". Seems to be a rather uncommon term?
The "machine learning algorithms" mentioned in the title seem to be a simple linear regression? They claim an accuracy of ~90% and show some sample results. The complete study data is only available upon request, however.
[22] S.J. Jebasingh Kirubakaran, M. Anto Bennet, N.R. Shanker, Non-Invasive antenna sensor based continuous glucose monitoring using pancreas dielectric radiation signal energy levels and machine learning algorithms, Biomedical Signal Processing and Control, Volume 85, 2023, 105072, https://doi.org/10.1016/j.bspc.2023.105072
Edit: read the paper, now more confused
> Commercial CGM devices have certain drawbacks in diabetic measurement during daily activities such as food intake, sleeping, exercise and driving. The drawbacks are continuous radiations from devices
So they think a drawback of CGM is the (Bluetooth) radiation, and their alternative is to zap the pancreas with, um, magic dielectric radiation? Or magic radiation that results in “dielectric” backscatter?
I do find myself wondering whether a watch- or patch-sized object could get a usable NMR signal from glucose. Maybe a neodymium magnet and a very carefully shaped probe antenna to compensate for the horribly nonuniform magnetic field? Maybe an AC field with no permanent magnet at all? I found a reference suggesting that measuring glucose in blood outside the body by 1T NMR is doable but marginal, so this may be a lost cause.
https://doi.org/10.1016/j.bspc.2023.105072
The OP paper is a bit lacking in any actual details of how glucose is being detected…
Usefull? It is if you use it. I do triathlons and knowing exactly where my blood-sugar level is at would allow me to focus better on the type of nutrition and the impact of it while working out. It would also tell me if i was a bit down before a race, so i can take some food.
Basically: this is a game-changer for amateur athletes, which would create a tremendous market for it. People i know already use the patches to measure as well, or lactate measurements, ketone measurements, etc. and that's just at the casual amateur level.
Another application that springs to mind is knowing when to eat instead of just having lunch and sugar-crashing 2 hours later in the office.
If I am connected to my body, I can also feel it by own biological sensors. But I do see the use case, to get another data input, for those cases where I am distracted and don't pay too much attention on myself.
There's a theory that says you basically won't ever gain weight if you prevent your blood sugar from going above a certain level. So it's an objective way of knowing how much to eat and when.
Also, to warn when blood sugar is too low. Some people (myself included) often get so into work (or whatever) that we forget to eat, with adverse consequences. An alert is very helpful.
Preventing non-hereditary diabetes could be much cheaper from a societal perspective.
I measure my heart rate one time for fun. It was insanely high at rest. I felt perfect, full of energy but it was completely obvious i needed to introduce rest days. I conviscated the heart rate monitor and everything was back to normal 3 days later.
We just reached (in a clinical trial) a comparable accuracy as early-stage invasive devices that got FDA approved with a shoe-box-sized device and we still have some work to do. The pre-print of our publication is here: https://www.researchsquare.com/article/rs-5289491/v1
I'm excited to see new developments but in this case, I'm not sure this will reach the market anytime soon.
Man, you have no idea. I'd gladly buy a shoebox-sized thing. No finger pricking and no test strips to buy is the king. It could be 4U rackmount thing for all I care, as long as it was noninvasive and accurate.
I see this a lot. People seem to ignore the “viable” part of MVP. If there’s more work to be done to make it a full marketable device, it isn’t viable in its current stage.
I wish them luck.
RF-based approaches have the problem that they are not specific to glucose. A molecule of glucose absorbs infrared light at specific wavelengths due to its size and types of bonds. It does not have specific absorption of radio frequencies. In this paper, researchers measured glucose in pure water at concentrations 100X physiological levels. I'd like to see this work with whole blood or a tissue phantom, or measure glucose independently from any other solute.
I fear we can assume that, although the approach might be novel, it can't replace needles for accurate measurement. But maybe I am overlooking the performance comparison.
https://www.nature.com/articles/s44172-024-00194-4
EDIT: No, i don't think it's mentioned...
I didn't read the paper yet, but I predict from the comments in this HN thread, that the proposed system is essentially a dielectric spectroscopy setup optimized for glucose detection (or any number of proxy byproducts/complexes/etc..)
Check this wikipedia page for example: https://en.wikipedia.org/wiki/Dielectric_spectroscopy
Look at the picture on the right, from typically lower to higher frequencies there is the motion response of ions, the reorientation response of molecules with a dipole moment, the excitation of vibrational modes in a molecule and the electronic excitation of electrons switching orbitals...
EDIT: since I have not read the article, I do not vouch for its authenticity (above my paygrade)
"Breakthroughs" in this field are a dime a dozen: https://finance.yahoo.com/news/liom-cracks-holy-grail-non-22...
My buddy is one of the few guys that has a sound (no pun intended) technology that might work. But future will tell. I won't give a link. Yes, the company secured funding.
I do agree on one part regardless of any of that though... I'm at the point of waiting for the one who actually sells me said device instead of the one that says they'll soon be able to.
And yes, skin color or skin temperature (fever!) does not matter for his measurement. I once submitted an SBIR grant for this project, but have no involvement anymore with this project. The NIH found this approach "highly innovative" but thought it can not be realized bc the technology is "prohibitive expensive". I assume they did not really read the proposal. I explicitly wrote that, while such a device costs 50k on the market, a slimmed down version produced in quantities would cost a few hundred dollars. They just had a short look at the approach, googled the hardware behind it, and rejected it. An idiot and google is a very dangerous combination!
So glad we've rebranded primitive ML and basic control mechanisms as AI.
Soon a Fast Fourier Transform will be rebranded as AI as well.
AI predates C. Actually AI predates lisp:
> IPL was used to implement several early artificial intelligence programs, also by the same authors: the Logic Theorist (1956), the General Problem Solver (1957), and their computer chess program NSS (1958).
https://en.m.wikipedia.org/wiki/Information_Processing_Langu...
For someone glucose-curious like me, the Stelo is just right. The app provides a slightly laggy real-time graph of measurements (collected every 5 minutes and reported every 15 minutes). It uses some heuristics to identify rising/falling events, and it'll notify you for the steeper cases (but I'm not sure about hypoglycemic event notifications, as I had a couple of those while sleeping and found out only when I woke up). On Android, it uses the Health Connect hub to sync some health data with other apps. It provides a rudimentary event-log function to add meal, exercise, and FYI notations. You'll also get a daily time-in-range wrap-up.
The more interesting analytics are in clarity.dexcom.com, which is a website that visualizes data that the app is constantly pushing, with a couple-hour lag. There you will find more graphs, tables that group measurements by day and hour, and various expected calculations (average, standard deviation, CV, calculated GMI). And you'll also find the all-important export to CSV button, which gives you all the sensor data. Using that I was able to import everything to Google Sheets, where I did a linear regression with finger-prick measurements to ascertain the sensor's (mild) deviation.
The Stelo feature set is clearly designed to provide all the data eventually, but not quickly enough to be useful for diabetics who need real-time info for glucose management. That's how they'll continue segmenting the Stelo and G7 audiences. I have no problem with that; if OTC GCM cost continues to drop, and they become as prevalent as annual lipid/metabolic labs, I could see the incidence of lifestyle-attributable T2D dropping, which would obviously have massive benefits to society.
https://www.notebookcheck.net/Orange-Pi-Watch-D-Pro-New-smar...
I'll take a closer look at the paper in Nature.
The paper is way overhyped. They've just built a meta material near field antenna. Nothing new I can see.
There are other papers available via Googling that glucose can be measured at around 4.2GHz. Seems hard to do it reliably, though.
They also know how to turn very complicated things into miniaturized production products.
If they could provide enough proof that it works the way they say it does, I bet they could find a really good suitor there. And you know the various other health/smart watch companies would love to get one up on Apple, outside of the obvious direct benefit it would provide their customers.
The challenge is, there's no a priori proof that a method can't work, because the information you need to make that assertion has to come from the same kind of research as trying to make it work. So far the start-ups have all failed in the same way, which is that a signal that looks promising in a test tube can't be reliably distinguished from the myriad sources of variation in the living system.
Note that I'm not in any way discouraging the work, just offering some historical context for the problem space.
Hats off to these researchers, who went a different direction.
Better health really can start with glucose monitoring for everyone. Because we could all learn about sugar's effect on our bodies.
There is also a paper from them doing it with the Google's Project Sali dev kit, the radar that was mostly demonstrated with gesture recognition, but looks like it's useful for other things too like this. They also show it can be also used to detect glucose levels in drinks too like Coke vs diet Coke.
Besides the current one posted, they have multiple other publications about the topic:
Using Project Sali back in 2018: https://scholar.google.ca/citations?view_op=view_citation&hl...
This one is also nice detailed one: https://www.mdpi.com/2072-4292/12/3/385
and looks like the current one is about improving this technology with enhanced signal-to-noise ratio.
I wonder if this is actually a viable path to detecting blood glucose. Wouldn't it be sensitive to other substances that affect the dielectric properties of a solution as complex as blood?
I had some issues with the Libre 2 sensor getting knocked off or loosening after a few days in the shower. But the libre 3 is smaller than a US quarter coin and lasts 15 days and doesn't snag on anything. Will this watch exceed that? Because if not, then I'm not looking to switch.
Lastly for diabetics or pre-diabetics that are only using finger sticks, I CANNOT stress enough how important a CGM is to your health. You learn so much about how your body actually works with certian foods rather than low frequency (but somewhat more accurate) finger sticks and that information dropped my A1C like a rock in just a few months. Tell your older relatives also. These are life saving devices.
"The system’s key components are a radar chip, which sends and receives signals through the body, an engineered “meta-surface”, which helps focus these signals for better accuracy, and microcontrollers, which process the radar signals using artificial intelligence algorithms. The algorithms improve the accuracy and reliability of the readings by learning from the data over time."
>The actual research: basically a better antenna that can distinguish between distilled water and a ~25000 mg/dL glucose solution absent any other factors. For context: At around 500 mg/dL you're dead.
I've looked over the studies/tests they've done and they look decent, though the accuracy is not that great.
You could also look at the stock economics to see that this company has not behaved like one with a bright future.
The existing alternatives do have access to the bloodstream.
> The existing alternatives do have access to the bloodstream.
So? Who says otherwise? People don't want invasive tests that have direct contact with their bloodstream.
Who says that's not addressed by this technology?
> directly improve quality of life a lot more than the annoyance of applying the sensor under the skin
How annoying is it? They don't cut a hole and insert it. How can you say how much it would improve the quality of life?
I don't see where all these assumptions about this technology come from.
I'm more than convinced we had this tech ages ago but it was not profitable to deploy it. Now either relevant patents are finally expiring or the market has changed enough to allow such players to enter.
But the hurdle today isn't in the physical sensing technology, it's how to detect an accurate signal from extremely noisy data.
Patents aren't the main blocker -- nobody has built this at all in a way that is accurate enough. And Apple has been trying hard, and they have nothing to do with Big Pharma.
I participated in studies with prototype devices to measure blood glucose and nobody I talked to mentioned patent blocks.
For others reading; this is about a small radar technology, having nothing to do with research in the detection of blood glucose whatsoever.
How do you know? The date on the article says October 29, 2024
Also - flashback to Rockley Photonics ($RKLY) - and their years-long promise of non-invasive glucose monitoring.
Does it mean AI models are used for data fitting? Or clustering?
For data generated in such low scales, wouldnt statistical methods or procedural methods be sufficient or efficient or both?
The answer to your last question is yes, especially when it’s from raw signals.
Tbf, there are applications from devices that do use deep learning methods but from experience they are not practical except on very edge cases.
In my experience when marketing wants to use AI, they will. Regardless of whether it is ML, basic statistics or even just a few if-else blocks.
It used to be the way you describe up to about 2-3 years ago, now the term is meaningless.
Generally, yes. Attenuation varies by frequency, and I guess blood sugar and pressure, though.
But as for this guy and his invention, don’t forget that in clinical setting, Theranos has shown more evidence of their product working. And GlucoWatch, a similar idea two decades ago, was FDA-approved and made it to market though still wasn’t clinically useful due to poor accuracy. Then, I’m not even talking about the charlatan cottage industry around glucose sensing watches, nor am I talking about how CGMs are a (generally) solved problem in diabetes.
Let’s wait for some clinical trials of the applied blood glucose sensing before we pop the champagne? It quite likely won’t happen, welcome as that invention would be.
Good to know I'm not generally then.
It’s also a little presumptuous to believe that someone would want to wear a bulky medical device for an easily manageable and non-life threatening condition all the time. If not wearing a medical device doesn’t kill you, then why do you need to wear it? Form factor and comfort is important.
Edit: It's here: https://www.nature.com/articles/s44172-024-00194-4
But he's clearly spent his entire career on this!