back to article AI of the needle: Here's how neural networks could detect nighttime low blood-sugar levels using your heart beat

Academics have applied for a patent describing how a neural network can detect low blood-sugar levels by analyzing heartbeat patterns rather than a blood sample. Keeping track of glucose levels is annoying and painful. Multiple times per day, diabetics have to prick their finger, place the small drop of blood on a test strip, …

  1. JetSetJim Silver badge

    Too little data - only 4 people for 2 weeks? Nice piece of work, but is the AI merely learning the subjects patterns rather than something that can be more widely applied. Needs to be run on 4000 people for a year, methinks, to get anywhere near useful.

    And I wish folks would stop patenting "X with an AI", we're back in the "X on an internet" days, it seems

    1. Trollslayer Silver badge

      I assume this is a 'proof of concept'.

    2. ibmalone Silver badge

      I'm always a bit surprised by what you can get into Nature! They have to patent before they publish though, so that's why it's patented at this quite preliminary stage. As for "X with an AI", haven't looked at the patent, but I'd hope there's some detail on the setup used, unlike "X on an internet" the trick is finding the right machine learning framework and training process for the problem (increasingly off-the-shelf admittedly).

      1. JetSetJim Silver badge

        Patent text not available yet as it's not passed muster - it's in the back 'n' forth stage of application, but the text from their poster says the invention "use[s] artificial intelligence to detect hypoglycaemic events from raw ECG signals acquired using off-the-shelf non-invasive wearable or ambient sensors", so on the face of it, it is "ECG on an AI" which in my mind may well prove valuable, but shouldn't be patentable as it's just a mathematical/computational technique

        1. ibmalone Silver badge

          Oh dear, that poster!

          The paper does give details on what they used, also in the Reg article, I hope the patent will contain some of that, otherwise it would definitely be too broad. (Bung it into some kind of ML framework is fairly apparent to anyone familiar with the state of the art, which framework to bung it into and what edges need rounding off is not until you start doing the work.)

          Interestingly they also do some interpretability work, looking at what the algorithm is picking up in the heartbeat.

          1. ibmalone Silver badge

            (On reflection I should clarify "Oh dear, that poster!" refers to the wall decoration rather than the esteemed commentard.)

            1. JetSetJim Silver badge
              Pint

              Have to admit that a reference to me didn't occur to me!

    3. ForthIsNotDead
      Coat

      Absolutely. And while we're at it, where the hell is the Blockchain, IoT, and cloud angles? They really missed a trick there. Pfft.

      1. Waseem Alkurdi Silver badge

        Not really

        IoT: I could see an ECG in a future iteration of the Apple Watch.

        Cloud: Chocolate Factory just released Google Health, which can connect to your ECG to provide a cloud-powered predictive insulin schedule (psst, buy more insulin!)

        Blockchain: And you thought you're going to pay for all that insulin with, er, real money?

    4. Clunking Fist Bronze badge

      "X with an AI"

      iDetect. eDetect. BloodBot. myBlood. myBloodBot. myDetect. Reg TM thank you.

  2. Anonymous Coward
    Anonymous Coward

    "The team applied to patent their technology"

    Good luck with that. If it's possible to patent it, then the system is even more worse than I thought. There is nothing special about it, data into a black box, data out. Anyone (with basic coding skills) can do it.

    Now an easy to use, cheap device that does this, that sould be patentable.

    Next up, penis size, will it satisfy her, should patent that.

    1. JetSetJim Silver badge

      > Now an easy to use, cheap device that does this, that sould be patentable.

      Agreed. The iWatch (not cheap, natch) does an ECG, so this should be doable on an iPhone in software, ergo no patent should be allowed (in the current right-pondian world, at least). The idea seems to be "apply a mathematical algorithm to a collected ECG trace to determine blood-glucose levels".

      You can get an ECG device for ~£100 on Amazon, possibly down to around £50 so just need the data feed from that

    2. Jedit Silver badge
      Joke

      "Next up, penis size, will it satisfy her, should patent that."

      Patenting a Sexual Satisfaction Predictor would require multiple trials with the same woman and a variety of men with different sized penises. Was this your idea or your partner's?

      1. Anonymous Coward
        Anonymous Coward

        Re: "Next up, penis size, will it satisfy her, should patent that."

        Damn it, didn't think of it that way, errm, honey about that trial.....

        1. Anonymous Coward
          Anonymous Coward

          Assumptions...

          I didn't know 'anonymous coward' had any sexual orientation connotation - so who knows whether it will be a fun experiment for them or not.....

  3. Alister Silver badge

    I fail to see why an "AI" is required for this, plain old data crunching without tacking on neural networks or anything else should be all that is required to identify patterns of ECG activity. As the poster above, I think "with an AI" seems to be done for the sake of it, sometimes.

    1. Anonymous Coward
      Anonymous Coward

      Buzzword, investors etc.

      Also its easier to let the 'ai' figure out the pattern

      1. osakajin

        BlockchAIn surely?

      2. Steve K Silver badge

        Quite

        1. Tiny sample size excluding potential patient beneficiaries

        2. Some correlation found.

        3. Patent.

        4. …?

        5. Profit!!!

    2. Dan 55 Silver badge

      What, don't you like not knowing how the answer is arrived at and spuriously incorrect results?

    3. Anonymous Coward
      Anonymous Coward

      I can see why

      One algorithm size doesn't fit all (different individuals, etc), ergo AI for the flexibility factor?

  4. Timmy B Silver badge

    A bit out of date

    Currently you can use a Freestyle LibreLink (like my OH) that allows you to test and monitor levels without needing to do finger pricks anywhere near as often. Far more convenient and more comfortable. Expensive, though.

    1. Korev Silver badge

      Re: A bit out of date

      I've had to call an ambulance twice for a friend who over-relied on a similar system.

      1. Timmy B Silver badge

        Re: A bit out of date

        I think the issue is in "over-relied". As with most monitoring you do need to have a measure or hypo and hyper awareness. It's not as accurate as there is a lag in the measurements but for at least 75% of tests it's perfect. You can't, for example, use it as a driving test due to the lag.

    2. Sweeper

      Re: A bit out of date

      Libre is available on Prescription, and hence free, to type 1 diabetics in the UK on the NHS as long as one of several conditions are met.

      1. Timmy B Silver badge

        Re: A bit out of date

        The nightmare of it is that the conditions can be changed by your GP practice. Ours have limited the age range and various other parameters so it makes it virtually impossible. They also recently changed the prescription meters, strips and needles to the cheapest they could legally purchase.

        1. tyne

          Re: A bit out of date

          The DVLA now accept the glucose measurements from the Libre as suitable for confirming your safe to drive, unless you are a PSV or HGV driver.

          1. Timmy B Silver badge

            Re: A bit out of date

            I did know this but had forgotten.... Thanks for giving the aged cells a prod.

      2. Snapshot
        Flame

        Re: A bit out of date

        I'm an IDDM Type 2. My treatment is identical to a Type 1 but because I started as a Type 2 my Libre costs me £100 per month. Snarl....

        1. Timmy B Silver badge

          Re: A bit out of date

          What's bonkers is that our Dr surgery does not recognise adult onset type one as distinct from type two. But only where the Libre is concerned. For everything else it does. My OH has adult onset type one. Double Snarl!!

  5. Sweeper

    As a type 1 Diabetic since the early seventies I have to say that the concept that low blood sugar leads to low heart rates is almost entirely inaccurate. In fact, when I was diagnosed having a higher than normal pulse rate was one of the symptoms we were told to watch out for with low blood sugar levels.

    1. Alister Silver badge

      Yes, I'm not sure where El Reg gets that from, as if you read the abstract the identification of a hypoglycemic event seems to be based on ECG changes, not heart rate, in particular a prolonged QT interval, an increase in the R/T amplitude and a raised S/T segment.

      Unfortunately all of those can also be a consequence of Ischemic Heart Disease, and a high proportion of Type 1 and 2 Diabetics will also suffer from IHD.

      I don't see how you could reliably differentiate between the two without careful calibration for each individual patient.

      1. JetSetJim Silver badge

        AI on ECGs seems to be a fruitful area of research, right now. There's one that predicts age and sex, one for atrial fibrillation, one forwill you die soon (presumably from a heart condition!) and now one for blood-sugar.

        The problem in several of these is there isn't a clear way of saying why the AI is predicting the outcome in question, which could be somewhat unnerving as a patient on the receiving end in front of a doctor shrugging...

    2. Anonymous Coward
      Anonymous Coward

      ECG contains a lot more than heart rate.

  6. tyne

    This sounds like an AI solution looking for it's problem... rather than a problem that needs to be solved.

    As a long term diabetic I now only use the traditional finger prick technique a couple of times a week at most. The rest of the time I wear a patch that last 2 weeks, that takes a measurement and stores it in it's memory every few minutes, scan the patch with an NFC smartphone or it's dedicated reader 4-5 times throughout a day and I get graphs of my blood glucose levels over the whole 24 hour period.

    I find keeping track of my blood glucose neither annoying or painful, it's just a fact of life. I suspect the faff of effectivley rigging myself up to an ECG that can take accurate enough readings would be considerably more effort.

    I luckily dont worry too much about hypo's but, for those that do there are similar solutions that can continually monitor your blood glucose 24 hours a day, and alert you if your in danger, they're currently expensive but they work, are available now, in the UK funded by the NHS for those that really do need them.

  7. Conrad Longmore

    I know when my blood sugars are low at night..

    I know when my blood sugars are low at night because I will typically have long and very vivid dreams, rather like a fever dream. That having been said, it's hardly a scientific method and other non-invasive techniques are available. The problem with *those* is that the DVLA only recognise a fingerprick test as being accurate enough for driving, so you are stuck with those anyway and it's likely your GP will only pay for one testing method.

    Test strips are expensive in the UK - *retail* can be about £25 for 50 strips although the NHS gets them cheaper. This means that many people with diabetes being treated with (for example) metformin or sitagliptin don't get test strips prescribed (or only in low quantities) which seems to lead to poorer overall control of diabetes compared to insulin-dependent diabetics who need to test several times a day. A reliable and inexpensive continuous glucose monitor would be an enormous benefit to many people.

    1. macjules Silver badge

      Re: I know when my blood sugars are low at night..

      My better half says that the test strips are like buying HP printer ink - you get the printer dead cheap and then get ripped off for the inks.

      1. Conrad Longmore

        Re: I know when my blood sugars are low at night..

        Yup, I think you are exactly right. NHS groups get offered the meters for next to nothing I believe..

  8. a_yank_lurker Silver badge

    Diabetic Monitoring

    The amount of monitoring a diabetic needs to do depends on the individual. Some only need it to monitor longish term trends rather than adjusting what one eats at the next meal. Some must be more aggressive in their monitoring. But what tends to used for diagnosis it are insulin, glucose, and A1C levels. The reason for the glucose monitoring is that blood glucose tends to correlate reasonably well with insulin levels, e.g. if your fasting/non-fasting levels are within bounds your insulin level is probably about were it should be. While heart disease is a complication of diabetes I am not so sure about the correlation between heart rate, etc and insulin levels.

    The small sample size and specific sample type makes drawing any conclusions problematic for its value in the real world. Plus I am not impressed with the buzzword bingo.

    1. Timmy B Silver badge

      Re: Diabetic Monitoring

      "The amount of monitoring a diabetic needs to do depends on the individual. Some only need it to monitor longish term trends rather than adjusting what one eats at the next meal. "

      If you're IDD in the UK then you must test within 2 hours of driving and every 2 hours of driving.

  9. Phil Bennett

    82%?

    If you only detect potentially fatal conditions 82% of the time then what the hell is the point?

  10. Anonymous Coward
    Anonymous Coward

    Lots of Issues

    1. Should this be patentable at all?

    At this stage it just seems an idea - maybe something in the ECG signal can be an indicator of nocturnal hypoglycemic events.

    The work done so far is with such a small sample and does not test in the target popualtion or for the target events. Most such ideas fail once real testing is done.

    The real work and value in such developments is the clinical testing and effidenc eof effectiveness so granting a patent just inhibits potential future development by warning off othergroups who might explore the area.

    2. Is this actually convenient?

    Even a basic three wire ECG is inconvnient and will almost certainy fall off during sleep. If the point is to monitor overnight then fair enough if it is worth the effort to solve these sort of issues with some sort of wearable device but a pin prick test will often be far more convenient.

    3. What is the medical application and does this device actually confer a potentially significant benefit in diagnosis, treatment or monitoring?

    4. Are their competing technologies? Certainly I have seen work on trans-dermal glucose monitoing which if it can be made to work sounds a far better more reliable and probably more convenient approach with a lot of applicatiosn not just mmonitoring of events during sleep.

    There is nothing wrong with teh work, but puffing it as more than a long shot unlikely to affect medical practice in the future is not realistic but hey it has a buzzword in its favour.

  11. TrumpSlurp the Troll Silver badge

    Partly accurate?

    The article talks about capturing a blood sample on a test strip then inserting it into a glucometer. Wrong. You insert the strip (which makes an electrical connection) then you collect the drop of blood.

    This seems to be a one trick pony - detect hypos whilst asleep - being hyped as an answer to all aspects of blood glucose monitoring. If it is significantly cheaper than CGM then it might have a place in the market. However if you are already using (for example) a Freestyle Libre with the add on Bluetooth transmitter then it doesn't seem to add any value nor does it seem to offer proper CGM.

    If someone can show that this technology is at least as accurate as finger pricks for measuring high, normal and low BG plus alerting for rapid changes then it might be a player. However there are no indications that detecting an absolute BG level has even been investigated.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2020