Please create an account to participate in the Slashdot moderation system

 



Forgot your password?
typodupeerror
×
Biotech AI Medicine

A New AI-Powered Eye Exam Reduces Errors By 74% (sciencemag.org) 34

sciencehabit quotes Science magazine: The classic eye exam may be about to get an upgrade. Researchers have developed an online vision test — fueled by artificial intelligence (AI) — that produces much more accurate diagnoses than the sheet of capital letters we've been staring at since the 19th century. If perfected, the test could also help patients with eye diseases track their vision at home...

[W]hen the researchers ran their "Stanford acuity test" (StAT) through 1000 computer simulations mimicking real patients, the diagnostic reduced error by 74% compared with the classic eye test, the team reports this month in the Proceedings of the AAAI Conference on Artificial Intelligence. The simulations work by starting with a known acuity score and factors in the types of mistakes a human might make. It then virtually "takes" the different eye tests in order to compare how accurate they are. The team used this instead of actual patients because it starts with the "true" acuity — something unknown in a human.

You can take StAT yourself at myeyes.ai, although the team cautions that the test isn't meant to replace doctor visits just yet.

This discussion has been archived. No new comments can be posted.

A New AI-Powered Eye Exam Reduces Errors By 74%

Comments Filter:
  • that makes headlines today and we'll never hear again! Yay!

  • by 93 Escort Wagon ( 326346 ) on Saturday June 06, 2020 @04:02PM (#60153840)

    ... is this test can diagnose eye problems experienced by a programmed simulation of a human better than a real doctor can. But simulated humans don't buy many pairs of eyeglasses or contact lenses, I suspect.

    Seriously, research at such an early stage doesn't seem newsworthy. Well, maybe if you're running their university department's PR team, it's worth posting on at that level.

    • Seriously, research at such an early stage doesn't seem newsworthy.

      Man you guys are just never happy are you:
      - Product not on the market: "We're too early, research never amounts to anything. Let us know when I can buy it."
      - Product is on the market: "OMG Slashvertisement alert. Why am I seeing ads! Msmash is ruining Slashdot the corporate sellout!

  • by Skip Whiffle ( 1452747 ) on Saturday June 06, 2020 @04:03PM (#60153844)
    I took the test, but I had no idea how to tell the system when the letter was indecipherable and just looked like a box, at which point I just had to guess. I think the test could have been more accurate, or at least more enjoyable to take, if it had a button for "I can't tell."
    • Comment removed (Score:4, Insightful)

      by account_deleted ( 4530225 ) on Saturday June 06, 2020 @04:19PM (#60153894)
      Comment removed based on user account deletion
      • I tested it on a 27" 4K monitor at a 0.5m distance and it doesn't work. Not enough resolution for testing that close as some tests were just a few pixels but the poor design of this site says it works at any distance when it obviously doesn't.

        This whole thing looks like a middle school science project.

    • by AmiMoJo ( 196126 )

      Can someone explain how the test works? It's not like any eye exam I've ever had.

      The letter M appears on screen and seems to rotate. I'm supposed to press the arrow keys... To indicate what?

      And at the end it only seems to give you a score out of 20, so basically it's just measuring your eyesight and not actually helping to correct it, right? Seems a bit useless if it doesn't help pick the right lens. What is the score out of 20 used for?

      • Can someone explain how the test works? It's not like any eye exam I've ever had. The letter M appears on screen and seems to rotate. I'm supposed to press the arrow keys... To indicate what?

        It shows an "M", "W", "E" and "backwards E" and you're suppose to indicate which way the open part of the letter is pointing, so Down for "M", Up for "W", Right for "E" and Left for "backwards E". As noted elsewhere, there's no way to indicate "can't tell" and you just have to enter something with an accidental correct being a error case on the part of the test.

      • It measures acuity - which is the medical term for resolution. This is the starting point for measuring eye health. In that it is degraded both due to refractive error and structural problems with the eye. Assessment of refractive error requires optical instruments - which can be as simple as a selection of multiple lenses - or sophisticated automated instruments which measure the shape of the eye and its refractive power using optical and imaging techniques. Typically a proper eye examination would include
        • by AmiMoJo ( 196126 )

          Thanks for that detailed explanation. I've never had such an eye test, it's always been a series of tests to determine what lenses I need to correct my vision rather than simply measuring how good it is.

          • You will have had such a test if you have ever had your eyes tested for anything, especially if being tested for prescription lenses. Whenever you have your eyes tested for lenses, you get a full eye health check, including acuity, intra-ocular pressure, examination/photography of the retina, visual fields/blind spot measurement.

            The normal acuity test is just to read letters off a Snellen chart https://en.wikipedia.org/wiki/... [wikipedia.org] - the result is given by the lowest row which you can read accurately.
    • by fahrbot-bot ( 874524 ) on Saturday June 06, 2020 @04:30PM (#60153938)

      I took the test, but I had no idea how to tell the system when the letter was indecipherable and just looked like a box, at which point I just had to guess. I think the test could have been more accurate, or at least more enjoyable to take, if it had a button for "I can't tell."

      In those circumstances, only getting it accidentally correct would be an error case, but it would probably be better to be able to proactively indicate that you can't tell which way it's pointing.

    • by ck42 ( 134627 )

      Same problem. Also found that it started off with letters too small to even begin testing. In an actual exam, they increase the size of the eye chart at least until I can BEGIN the test.
      If I simply guessed because I couldn't tell (there's no "I can't read it at all button"), what happens if I guessed correctly? I'd have to think that this would skew the results.
      TLDR: it needs more work.

  • by schwit1 ( 797399 ) on Saturday June 06, 2020 @04:06PM (#60153856)

    Maybe I'm just getting suspicious in my old age.

    • by mykepredko ( 40154 ) on Saturday June 06, 2020 @04:43PM (#60153992) Homepage

      I understand your suspicions and you're right to be cautious. Identify theft is not something to be taken lightly. To make matters worse, there are people out there who would charge you to tell you if your information is available on the Internet.

      Not me, though - because you're a Slashdotter I'll be happy to see if you've been compromised for free, just send me the credit card number along with your social security number.

  • The retina has fine detail, so I'm surprised that there isn't a mechanism to directly measure the lens distortion by viewing the retina from an external instrument while the eye focuses on near and far objects.

    • As Solandri linked below, those machines exist [wikipedia.org]. But 'seeing' strongly involves the brain, not just the retina, and many people can see less well than they should just due to the retina. Also, it's not necessarily best to prescribe for optically perfect distance vision if it's not needed for functional distance vision; that can lead to additional ciliary strain and even a ciliary spasm [wikipedia.org].
  • Ran the test and got results that were better than I was told by my ophthalmologist (Left eye StAT: 20/13, Opthalmologist: 20/15, Right eye StAT: 20/14, Opthalmologist: 20/18).

    I'd be curious to hear from other Slashdotters as to how accurate the test is for them. This is something that we (other than those that read /. on their phone) can test fairly easily.

    I don't know where "AI" comes into it, from RFTA it seems like the magic is using the rotated letter "E" and using something like a successive approxim

    • I scored 20/17 in my left and 20/16 in my right. With my glasses for astigmatism I still got 20/17 in my left, but achieved 20/8 in my right. I also tested at distances of 1.5', 3.5', and 20' with similar results.
  • by Solandri ( 704621 ) on Saturday June 06, 2020 @06:05PM (#60154248)
    Then why not just use a machine which does the entire test automatically [wikipedia.org]? No patient feedback required; the patient just needs to look in the machine. (The same can be done manually [wikipedia.org] as well.)
    • Machines that do eye exams automatically do not give sane results with certain eye conditions.

      Every autorefractor they've put me on plus my optometrists brand new automatic intraocular pressure (glaucoma air puff) tester do not work on my eyes. My optometrist resorted to just touching my eyeball with his finger to judge pressure since that new piece of equipment wouldn't work.

  • Ya, sure. Use a credit card? I'm not that dumb.
  • It does not matter if the test is 100% accurate. A lot of healthcare related stuff is controlled by licensing bodies who will never approve this. Why would they approve something which puts their members out of business.

  • How does this generate an eyeglass prescription that could possibly include astigmatism? How doe sthis check for retnial tears? How does this check for glacoma?

  • It rates me as 20/18 despite having severe astigmatism in both eyes and nearsightedness.
  • Automatic refractors have been around for years. They are highly accurate. Optometrists don't tend to use them, because then why would you need an optometrist? However, ophthalmologists do use them. https://en.wikipedia.org/wiki/... [wikipedia.org]

The one day you'd sell your soul for something, souls are a glut.

Working...