Slashdot Mirror


Google Successfully Uses Machine Learning To Detect Diabetic Retinopathy (betanews.com)

BrianFagioli writes from a report via BetaNews: Diabetic eye disease is caused by retinopathy. Affected diabetics can have small tears inside the eye, causing bleeding. Over time, they can lose vision, and ultimately, they can go blind. Luckily, Google has been trying to use machine learning to detect diabetic retinopathy. Guess what? The search giant has seen much success. Not only are the computers able to detect the disease at the same level as ophthalmologists, but Google is actually slightly better! "A few years ago, a Google research team began studying whether machine learning could be used to screen for diabetic retinopathy (DR). Today, in the Journal of the American Medical Association, we've published our results: a deep learning algorithm capable of interpreting signs of DR in retinal photographs, potentially helping doctors screen more patients, especially in underserved communities with limited resources," says Lily Peng, MD Ph.D., Product Manger at Google. She goes on to say "our algorithm performs on par with the ophthalmologists, achieving both high sensitivity and specificity. [...] For example, on the validation set described in Figure 2, the algorithm has a F-score of 0.95, which is slightly better than the median. F-score of the 8 ophthalmologists we consulted (measured at 0.91)."

30 comments

  1. I programmed a cheap Android phone by Anonymous Coward · · Score: 0

    Did something similar with one of these phones for a project in my master's curriculum. My team characterized different eye conditions and applied DSP to cancel out the effects. It worked OK. Basically what we were doing was to mask a spot on the retina or priapism of the cornea's reticulum.

    1. Re:I programmed a cheap Android phone by Anonymous Coward · · Score: 0

      Would you anticipate seeing DR self-screening machines
      next to the drugstore blood pressure machine?
      Seems like it could be done with current technology.

    2. Re:I programmed a cheap Android phone by bargainsale · · Score: 1

      Priapism? Well, I suppose it makes you go blind ...
      The cornea does not have a "reticulum." Either.

      --
      Aberrations have appeared in my destiny prognostication engine!
    3. Re:I programmed a cheap Android phone by sexconker · · Score: 1

      He's just slutting out his aliexpress referral link.

  2. Diagnosis as a Service by Anonymous Coward · · Score: 0

    They're also really good at diagnosing sex addiction. Google gives every user a Porn Score from 0 to 99 based on your search history. I was thrilled to receive a bug bounty for causing an integer overflow.

  3. Hairy Palm Syndrome by Anonymous Coward · · Score: 0

    HPS will do the same, or so I have read.

  4. That's nice.... by mark-t · · Score: 0

    ... so they can detect something for which there is no known cure sooner.

    Thrilling.

    So now you'll know even sooner that you're going to lose your eyesight eventually.... I suppose this might give you more opportunity to prepare for it, but I expect all that it will actually do is give you more time to be dreading the future.

    1. Re:That's nice.... by Anonymous Coward · · Score: 0

      Just because it can't be cured does not mean it can't be controlled to some degree with lifestyle changes.

    2. Re:That's nice.... by Anonymous Coward · · Score: 0

      blindfighting skill farming.

    3. Re:That's nice.... by Anonymous Coward · · Score: 0

      A diabetic usually knows if he has diabetic retinopathy. Needing to go to an ophthalmologist is usually the second or third clue. Just like neuropathy, retinopathy is a complication that's fairly easy to self diagnose.

    4. Re:That's nice.... by Anonymous Coward · · Score: 2, Insightful

      Exactly. Speaking as a thirty year type one diabetic who has gone through the "waking up with black spots" phase (whoops, there's your retinopathy), having it confirmed by the doc (more like, they inject you with this crazy neon yellow fluid and shoot a bright light into your eyes looking for the exact areas of damage), and the two month long series of laser treatments to stop the retinal bleeding (dear heaven, thank you for the miracles of modern medicine - not to mention my sexy ophthalmologist) and ongoing work with the endocrinologist to get my blood sugar back in control... to the OP, denial of your condition (assuming you're diabetic) will only bring the complications on sooner. Management, baby - that's all we've got for the time being.

    5. Re:That's nice.... by bargainsale · · Score: 5, Informative

      Nonsense. Diabetic retinopathy is very treatable. (It's actually probably the most valuable thing I myself do during my working week.)
      Certainly it doesn't always work in every case. But that's true of any treatment you care to mention.
      Panretinal photocoagulation (with lasers) has saved the sight of hundreds of thousands of people. Over the past few years, on top of this, there have been major advances using antiVEGF treatments like Lucentis/ranibizumab.
      It is *eminently* worthwhile for diabetics to be screened for eye disease. The problem is not that we can't treat it - it's that treatment is best done *before* the patient notices any problem. Hence, screening.

      --
      Aberrations have appeared in my destiny prognostication engine!
    6. Re:That's nice.... by bargainsale · · Score: 3, Informative

      That's exactly wrong and terrible advice if any diabetic believes it. Diabetic retinopathy is best treated (and often urgently *needs* treatment to prevent permanent damage) *before* the patient notices symptoms.( I am an ophthalmologist who specialises in exactly this field.)

      --
      Aberrations have appeared in my destiny prognostication engine!
    7. Re:That's nice.... by dmr001 · · Score: 2

      Blowing my mod points for the opportunity to clarify why we screen for diabetic retinopathy: By the time a diabetic patient has visible diabetic retinopathy, laser photocoagulation treatment cannot always repair the damage. The goal is to find the bleeding before risking significant visual loss, when treatment tends to be more successful. This is why most organizations (like the American Diabetes Association) recommend yearly dilated eye exams for diabetic patients. Unfortunately, screening can be expensive for underinsured or uninsured patients, or those without access to ophthalmologists or optometrists. As a primary care physician I asked if I couldn't get trained myself to save the cost for my uninsured patients, and got basically a bunch of eye-rolling. Somewhat like neuropathy, you don't want to be able to self-diagnose it: you want to prevent it before it becomes noticeable, which is awfully close in many patients to the point of also being irreversible. Diabetic neuropathies tend to be easier and cheaper to diagnose currently and don't need a specialist (and could probably be done by patients themselves with monofilament examinations). The same isn't true for the eyeballs —yet.

    8. Re:That's nice.... by Anonymous Coward · · Score: 0

      Diabetic Retinopathy has been known to be reversible in certain cases on a low fat plant based (vegan) diet since the 1950s:

      http://nutritionfacts.org/video/can-diabetic-retinopathy-be-reversed/
      and
      https://youtu.be/gIcSPwsNUyE?t=10m6s

    9. Re:That's nice.... by mark-t · · Score: 1

      I was under the impression that it was not something that could be prevented by treatment If I am wrong about this then I misspoke

    10. Re:That's nice.... by Anonymous Coward · · Score: 0

      I don't see where Anon is giving advice about anything. And while it would be great to diagnose retinopathy before it becomes noticeable, in general it just isn't likely to happen because people usually don't go to the doctor until they notice a problem. Hell, people don't even go to the doctor to get diagnosed with diabetes until they are suffering from a serious complication. I know, because I went for several months with an unquenchable thirst, blurring vision, numbness in fingers and toes, fog, tiredness, frequent urination, etc. before I finally went to the doctor and was diagnosed with late onset type 1. The final straw was going from around 185 lbs. to 130 lbs.in the space of about 2 weeks. Quite a few people have to be hospitalized before they get diagnosed, and you can be certain if they're at that stage, they've have already suffered from diabetic retinopathy to some degree.

    11. Re:That's nice.... by TheRaven64 · · Score: 2

      According to my optician, it's also one of the early signs of diabetes in a lot of people. It's something that they routinely check for (it's very easy to do if you have two photos of the retina from different times and not something that it should be difficult to train an expert system to do). The first time I had my retina photographed, my optician was telling me that the previous week they'd caught early signs of diabetes in a child that they were fitting for glasses: they were able to begin insulin treatments nice and early and significantly reduce the amount of damage that occurred before more obvious symptoms.

      --
      I am TheRaven on Soylent News
    12. Re:That's nice.... by Anonymous Coward · · Score: 0

      It can be treated - but not prevented (unless you sort out the diabetes before you get diabetic retinopathy).

    13. Re:That's nice.... by Anonymous Coward · · Score: 0

      > I suppose this might give you more opportunity to prepare for it

      Like many other diabetic complicatuons, improving blood glucose control can not only help prevent further damage but can allow the body to heal much of the existing damage. It's not perfect: after almost 50 years of Type 1 diabetes I have notable kidney damage and some spots on my feet that don't sweat. But my skin and blood pressure are good clues to how well I'm controlling my diabetes. And at my age, I face the potential of eye surgery for *both* cataracts and to treat any retinopathy. But the laser surgery for retinopathy has proven very effective for many long term diabetics, especially if it's done early.

      There is also a known cure for Type 1, in testing at Mass. General Hospital. Look up Dr. Faustman's work at http://www.faustmanlab.org/. They're in their second year of a five year trial on human subjects, the *second* round of testing, of a cure for most Type 1 diabetes. It involves using the BCG vaccine to trigger a change in the immune system, one that allows adult stem cells to change to insulin producing cells without being destroyed. If it works well and is held up by the FDA, I'm taking a trip to India to arrange BCG vaccines administered regularly and to arrange extremely tight blood sugar control during the treatment. The tight blood sugar control during treatment is why other labs were unable to reproduce the results on lab animals at first.

    14. Re:That's nice.... by BarbaraHudson · · Score: 1

      Some fool wrote:

      A diabetic usually knows if he has diabetic retinopathy. Needing to go to an ophthalmologist is usually the second or third clue. Just like neuropathy, retinopathy is a complication that's fairly easy to self diagnose.

      No, diabetics don't know they have diabetic retinopathy in the early stages

      . WTF did you get that idea from?

      What are the symptoms of diabetic retinopathy and DME?

      The early stages of diabetic retinopathy usually have no symptoms. The disease often progresses unnoticed until it affects vision. Bleeding from abnormal retinal blood vessels can cause the appearance of “floating” spots. These spots sometimes clear on their own. But without prompt treatment, bleeding often recurs, increasing the risk of permanent vision loss. If DME occurs, it can cause blurred vision.

      Because the disease progresses slowly, you most likely won't notice until stage 4 - when blood vessels in your retina cause floaters that block your vision. You'll probably assume that any temporary blurriness (which doesn't always happen) is caused by fatigue or eye strain, and since you don't really notice it after a while, and adapt to it, you don't realize what happens.

      Diabetic retinopathy may progress through four stages:

      1. Mild nonproliferative retinopathy. Small areas of balloon-like swelling in the retina’s tiny blood vessels, called microaneurysms, occur at this earliest stage of the disease. These microaneurysms may leak fluid into the retina.

      2. Moderate nonproliferative retinopathy. As the disease progresses, blood vessels that nourish the retina may swell and distort. They may also lose their ability to transport blood. Both conditions cause characteristic changes to the appearance of the retina and may contribute to DME.

      3. Severe nonproliferative retinopathy. Many more blood vessels are blocked, depriving blood supply to areas of the retina. These areas secrete growth factors that signal the retina to grow new blood vessels.

      4. Proliferative diabetic retinopathy (PDR). At this advanced stage, growth factors secreted by the retina trigger the proliferation of new blood vessels, which grow along the inside surface of the retina and into the vitreous gel, the fluid that fills the eye. The new blood vessels are fragile, which makes them more likely to leak and bleed. Accompanying scar tissue can contract and cause retinal detachment—the pulling away of the retina from underlying tissue, like wallpaper peeling away from a wall. Retinal detachment can lead to permanent vision loss.

      --
      "Transparent" is a shit show that trades on every stereotype going. A man in drag is NOT a transsexual.
  5. Who called it AI? by JMZero · · Score: 1

    I don't see anyone calling this AI except you. Whatever pattern matching you use to determine when to deploy this garbage stock comment, I think it has misfired here. ..or, wait, do you also have some pointless objection to calling this machine learning?

    --
    Let's not stir that bag of worms...
    1. Re:Who called it AI? by 110010001000 · · Score: 1

      AI or Machine Learning. What is the difference at this point? We are calling basic programs "AI" now. Complete garbage.

    2. Re:Who called it AI? by Anonymous Coward · · Score: 0

      Don't mind him. He's like a drooling grampa who can't stop making racial slurs about the damn gooks or whatever, only his thing is to wet himself and mumble constantly about how this and that aren't really AI.

    3. Re:Who called it AI? by JMZero · · Score: 1

      Well, I guess the difference is that while it would be a stretch to call this AI, it is very reasonable to call it machine learning - and that's what they called it. And yeah, I get it, you have some pointless axe to grind about people generally overusing the term AI. Nobody cares.

      But I will say that this exact comment, which is a waste of space every time it crops up on this kind of story, at least provided me a laugh this time. Your shitty, pouty reply is just icing on the entertainment cake. I would have felt maybe a bit mean if you'd actually copped to shitposting without reading the summary. But I don't imagine you're that kind of person (ie. someone with a shred of dignity or self awareness).

      --
      Let's not stir that bag of worms...
  6. Remember when... by matbury · · Score: 1

    ...Google told us they could predict outbreaks of colds and flu? Let's wait and see how well their latest attempts at diagnosis and prediction go.

  7. deep learning / deep voodoo by Anonymous Coward · · Score: 0

    The article is available here: http://jamanetwork.com/journals/jama for 11/29/16 together with 3 editorials. In addition to the ophthalmology piece details of the convergence of clinical medicine and information technology are discussed particularly as regards pathology and radiology. There is no question that this is important stuff. The bottom line is that the machine demonstrates a high correlation with the humans Everyone knows what criteria the humans are using No one knows what criteria the machine is using because it taught itself.

  8. Better than an ophthalmologist? by Anonymous Coward · · Score: 0

    Let me know when it's better than a retinologist...

    I guess I understand that if you live in the sticks, you make do with what you have. I suppose there are diabetics who see general practitioners or family medicine doctors instead of endocrinologists, but that seems like a bad idea, too.

  9. Better Obamacare through Machine Learning. by Anonymous Coward · · Score: 0

    Google Successfully Uses Machine Learning To Detect Diabetic Retinopathy

    So now this means Google will have an even better health plan for their employees?