Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
×
Biotech Cellphones Medicine

Researchers Can Accurately Measure Blood Pressure Using Just A Cellphone Video Of Your Face (utoronto.ca) 37

A new study published in the American Heart Association journal Circulation: Cardiovascular Imaging found that blood pressure can be measured accurately by taking a quick video selfie. An anonymous reader quotes this announcement from the University of Toronto: Kang Lee, a professor of applied psychology and human development at the Ontario Institute for Studies in Education and Canada Research Chair in developmental neuroscience, was the lead author of the study... Using a technology co-discovered by Lee and his postdoctoral researcher Paul Zheng called transdermal optical imaging, researchers measured the blood pressure of 1,328 Canadian and Chinese adults by capturing two-minute videos of their faces on an iPhone. Results were compared to standard devices used to measure blood pressure. The researchers found they were able to measure three types of blood pressure with 95 to 96 per cent accuracy.
Lee co-founded a company to turn their technology into a smartphone app (named Anura) that reports stress-level measurements and resting heart rate from a 30-second video of your face -- with plans to release a new version also returning blood pressure results sometime this fall in China.

The university also notes that the researchers now hope to extend their technology so it can measure blood-gluclose levels and cholesterol.
This discussion has been archived. No new comments can be posted.

Researchers Can Accurately Measure Blood Pressure Using Just A Cellphone Video Of Your Face

Comments Filter:
  • Bleh, video is needed -- so we won't know Mona Lisa's blood pressure nor cholesterol. Waiting for better doctors than these... :p

  • But at my doctor, I always have white coat hypertension, because I'm afraid he'll forbid me booze and meat.

    • Take a Valium

      • Xanax, when taken as directed, is a safe and effective palindrome.

        • Princess Xanax is not to be trifled with; her song of Peace and Sleep and Contentment is powerful -- and highly addictive. Also, a highly controlled substance, not that easy to get (legally).
          FWIW, there are some 100% legal OTC things you can take to temporarily reduce your blood pressure. Beet juice concentrate is one. Agmatine Sulfate is another. Add a little l-theanine or valerian root extract to those, an you're calm as can be. Take them about 30-40 minutes before checking your BP.
          Of course what you re
    • You don't actually believe that 'doctors' are some sort of higher life-form, omniscient and omnipotent, who can force you to not eat meat or have a drink or two, do you? Also just because someone managed to slide through medical school doesn't mean they graduated at the top of their class with honors, either.
      Just get regular exercise and don't be stupid about what you eat every day and you'll be fine.
    • white coat hypertension

      My highest reading was after they left me for ten minutes sitting next to a box of rubber gloves and a tube of lubricant.

      • My highest reading was after they left me for ten minutes sitting next to a box of rubber gloves and a tube of lubricant.

        You do realize that could be read in more ways than one, don't you? You could have been fearful or aroused.

        I get a kind of white coat hypertension too. When I visit my physician it's often because of pain from my arthritis, which is not helped by my having to walk down a long hallway with a hard concrete floor to get to the office. Then when in the office I'm asked their standard set of questions on what might be causing foot pain and inflammation like heart failure, diabetes, and kidney problems. At fi

        • By doing this they also confirm both that you are who the craft says (screwups happen) and that you are aware and involved in your medical care. If you're giving answers that don't match your chart, you may have vascular dementia or a drug side effect. If you don't know what you're taking, and why, how can you give informed consent?
    • BTW, why does my spellchecker not work on titles here?

    • by tlhIngan ( 30335 )

      But at my doctor, I always have white coat hypertension, because I'm afraid he'll forbid me booze and meat.

      No doctor can order you to do anything. They can write a prescription, but you're not forced to get it filled. It's generally a good idea, though. But you're also allowed to question them, like if they prescribe you an antibiotic when you have a cold or flu (antibiotics don't work against viruses. You need an antiviral for that).

      And doctors know you're human. They'll never ask you to ban red meat or al

  • Don't believe it (Score:4, Interesting)

    by schwit1 ( 797399 ) on Saturday August 10, 2019 @10:59AM (#59073976)

    It's just a scam to get your face in a database

    • That's not a stretch at all.

      The vein of gold in these apps is the data, right? Insurance companies would pay for this. Cops could detect that you're hypertensive. Big Pharma could direct ads to your lovely face.

    • by AmiMoJo ( 196126 )

      I hope so, because the potential for abuse if it is real is pretty alarming.

      • They got average measurement bias, for systolic blood pressure, of 0.39 mm Hg (0.35 to 0.44) with standard deviation of 7.30 (7.28 to 7.32).
        For diastolic blood pressure average measurement bias was -0.20 mm Hg (-0.23 to -0.17) with SD of 6.00 (5.98 to 6.02).

        They were using data from subjects whose systolic blood pressure was between 100 and 139 mm Hg and diastolic pressure was between 60 and 89 mm Hg.

        If your measurement error ranges from 0.28% to 0.39% (0.39 mm Hg as percent of 139 to 100 mm Hg) and your st

    • ..yeah, this, my first thought was, "Gee, wonder how much data that app is stealing from you that they're not telling you about, and how much it tracks you?"
    • The app that tells people if one's underwear is on too tight should be pretty accurate.

  • Does it mean "30 points over your actual blood pressure" like is currently the accepted method? Doctors love selling placebo-like BP meds and insurance companies love charging you more even though you aren't actually a higher risk.

    • by nagora ( 177841 )

      You could click on the link in the story and read the abstract, which defines "accurately" and mentions the need for further work to get up to clinical accuracy levels for other groups of subjects.

  • Needs some work (Score:5, Interesting)

    by nagora ( 177841 ) on Saturday August 10, 2019 @11:05AM (#59073988)

    The people chosen for the test were within normal blood pressure ranges, so basically the system had a good head start. To be any real use it needs to be able to spot people outside of that range

    Now it does say this in the paper, but the reporting (of course) is not spending much time on the fact that it didn't really do a lot better than if you or I looked at the videos and said "(s)he seems fine to me" and wrote down whatever the current definitions of "normal BP" is.

  • The summary and the story mention measuring three types of blood pressure. I thought pressure was a singular thing from physics - there may be different units, but only one "type" of pressure. If you read the linked paper, they are reading three different pressure measurements (systolic, diastolic, and pulse pressure). Yeah, it's being pedantic, but this is slashdot.
  • Comment removed (Score:5, Interesting)

    by account_deleted ( 4530225 ) on Saturday August 10, 2019 @11:33AM (#59074024)
    Comment removed based on user account deletion
    • by ceoyoyo ( 59147 )

      200 can be a perfectly reasonable sample size. That's why you do statistics, to determine what conclusions you can reasonably draw, given the precision provided by your sample.

      They don't seem to have done the relevant statistics.

    • by swell ( 195815 )

      I test blood pressure daily with a standard cuff device. Actually I test 3-6 times in succession and the responses are wildly different. At the doctor's office and hospital they have fancy devices that are timed to take 3 separate readings at wide intervals. Those results, too, are wildly varied so the machine averages them. Either these devices are terribly inaccurate or my blood pressure changes dramatically in the space of 2 minutes.

      This 'face reader' was tested against devices like mine and found to be

      • by g01d4 ( 888748 )

        responses are wildly different

        I've not seen a good explanation for why that is. Okay, so I just googled and this Introduction [bmj.com] cites a few references.

  • Are those cheap cameras really able to pick up subtle details like skin tension, and other very minute details that indicate high blood pressure that humans don't pick up on?
    • Are those cheap cameras really able to pick up subtle details like skin tension, and other very minute details that indicate high blood pressure that humans don't pick up on?

      These "cheap" cameras can now do 4K HD video at 30fps with 24 bit color. That's a rather high end "cheap" camera on a phone but even a smart phone from a few years ago could do 1080p video at 24fps.

      I expect that with the lower the quality camera it would take better lighting, longer video (more frames for comparison), a better focus, and so on. There's going to be a limit of course, my 20 year old "megapixel" camera might not have the resolution needed. Today a camera capable of 8 megapixel resolution is

  • This advancement in technology brought to you by Dr. Voight and Dr. Kampff . A far faster and less invasive test than the previous method, which was obtaining a bone marrow sample.

  • by serviscope_minor ( 664417 ) on Saturday August 10, 2019 @12:03PM (#59074098) Journal

    Well honestly this smells a little off to me. I know enough computer vision and ML that I ought to be able to read and understand the paper, but the full text isn't available. But I'm pretty skeptical. Given the average paper quality, it sounds like it was a case of throwing deep learning at a problem, getting decent results on a hold out set and calling it a day.

    1300 is not a lot of people for training. Sure people are hard to get but that's not my problem. Secondly I'll bet the dataset is hopelessly biased. The money says there were hardly any black people in the dataset (the institution is in Canada where they make up about 3.5% of the population, so I'd expect around 3.5%), and I'll bet it works less well on people with more opaque skin. But if the validation set has the same bias as the sample then they wouldn't notice the problem.

    There's also that normative problem others have pointed out. I don't know what they mean by +/- in the short description, is that L2 (Gaussian) based at 1, 2 or 3 sigma or is it L1 (percentile) based at 75, 95 or 99%? Hard to know, really, but the range for normal blood pressure isn't very large, so I wonder what results the null hypothesis of predicting the mean is.

    And it's not clear if they were predicting resting heart rate. If so, then the ML algorithm could well also be picking up stuff correlated with blood pressure to improve the accuracy, e.g. weight. Because given the opportunity an ML algorithm will "cheat".

    So yeah I'm super skeptical.

  • Since the dawn of creation; well no, perhaps just in my youth (60 years ago, nearly the same thing), the archetypal high blood pressure sufferer was a red faced businessman. In particular, red-faced. Perhaps a simple colour test would be as accurate.

  • I've talked to many nurses and doctors who won't use the automatic blood pressure cuffs, because they are so inaccurate. Instead, they go back to the old way of using a stethoscope and meter. Some entire hospital or physician systems have eliminated the automatic versions.

    And you're telling me that this facial scanning system can beat that? I doubt it.

"If the code and the comments disagree, then both are probably wrong." -- Norm Schryer

Working...