/ /

  • linkedin
  • Increase Font
  • Sharebar

    The Retinator II: Judgment Day?

    Most human aspect of computers is they are, as humans, prone to err

    Hollywood loves trilogies, and though not all sequels are created equal, as a general rule, the second installment tends to be the best. The examples of this phenomenon are numerous (Godfather Part II, Back to the Future II, The Empire Strikes Back, etc.). However, the most-cultured Ophthalmology Times readers will no doubt point to Terminator II as the best example of all. 

    Those same Ophthalmology Times readers also will remember that 7 years ago, this space held an editorial by Peter J. McDonnell, MD (OphthalmologyTimes.com/Revenge). For those lacking robot-like recall, the editorial commented on “Automated early detection of diabetic retinopathy” (Abramoff MD, et al. Ophthalmology. 2010;117:1147-1154), which demonstrated that computer algorithms had equaled the ability of retina specialists to detect diabetic retinopathy, potentially ushering in a world of machines replacing physicians—hence, “The Retinator.”

    Of course, that doomsday editorial was written by my very own chairman just as I was starting residency at the Wilmer Eye Institute. At the time, I found only two ways to interpret the news:

    1. A machine would say “Hasta la vista, baby” to my career as a retina specialist even before my training was over, or
    2. I would someday be able to sit on a beach, getting paid to “supervise” a computer as it did my job. Despite the apocalyptic signs, I gambled on the latter and finished residency and fellowship anyway.  

    As predicted, the sophistication of automated diagnostic systems has since improved to super-human levels. Much of the recent research has revolved around “deep learning,” a form of artificial intelligence whereby the computer is trained on representations of data and “discovers” features within these data correlated to disease.

    In diabetic retinopathy, Google and others have developed deep learning algorithms trained on color fundus images to detect referable diabetic retinopathy and macular edema with high sensitivity and specificity (Gulshan V, et al. JAMA. 2016;316:2402-2410). But, as some have noted, these algorithms operate as a “black box”—i.e., we don’t know what features within the retinal images the machine is using to classify disease (Wong, TY, et al, JAMA. 2016;316:2366-2367).

    Meanwhile, the human brains behind the original “Retinator” have taken a different approach toward developing a high-performing algorithm for the detection of diabetic retinopathy and macular edema. Instead of a “black box,” their algorithm identifies explicit lesions (e.g., microaneurysms, intraretinal hemorrhages, exudates) and uses deep learning to train multiple detectors for these lesions, thus mimicking the visual cortex of the retina specialist to hone in on pathology (Abramoff MD, et al. IOVS. 2016;57:5200-5206). Their company, IDx, recently joined forces with IBM’s Watson (yes, the computer that has conquered the world’s greatest chess masters) in efforts to further develop and distribute its computing resources for improving global health care. 

    Job security, for now

    New Call-to-action


    You must be signed in to leave a comment. Registering is fast and free!

    All comments must follow the ModernMedicine Network community rules and terms of use, and will be moderated. ModernMedicine reserves the right to use the comments we receive, in whole or in part,in any medium. See also the Terms of Use, Privacy Policy and Community FAQ.

    • No comments available


    View Results