July 2019
Volume 60, Issue 9
Open Access
ARVO Annual Meeting Abstract  |   July 2019
FDA-authorized autonomous AI for diabetic retinopathy screening in clinical routine
Author Affiliations & Notes
  • Bianca S Gerendas
    Vienna Reading Center, Department of Ophthalmology, Medical University of Vienna, Austria
  • Martina Neschi
    Vienna Reading Center, Department of Ophthalmology, Medical University of Vienna, Austria
  • Martin Michl
    Vienna Reading Center, Department of Ophthalmology, Medical University of Vienna, Austria
  • Lupyr Kostiantyn
    Vienna Reading Center, Department of Ophthalmology, Medical University of Vienna, Austria
  • Gabor Gy Deak
    Vienna Reading Center, Department of Ophthalmology, Medical University of Vienna, Austria
  • Yvonne Winhofer
    Department of Endocrinology, Medical University of Vienna, Austria
  • Alexandra Kautzky-Willer
    Department of Endocrinology, Medical University of Vienna, Austria
  • Michael David Abramoff
    IDx Inc, Coralville, Iowa, United States
    Department of Ophthalmology and Visual Sciences, University of Iowa, Iowa City, Iowa, United States
  • Ursula Schmidt-Erfurth
    Vienna Reading Center, Department of Ophthalmology, Medical University of Vienna, Austria
  • Footnotes
    Commercial Relationships   Bianca S Gerendas, IDx, LCC (F); Martina Neschi, None; Martin Michl, None; Lupyr Kostiantyn, None; Gabor Deak, None; Yvonne Winhofer, None; Alexandra Kautzky-Willer, None; Michael Abramoff, Alimera Life Sciences (C), IDx Inc (I), IDx Inc (E), IDx Inc (C), University of Iowa (P); Ursula Schmidt-Erfurth, None
  • Footnotes
    Support  IDx Research Support
Investigative Ophthalmology & Visual Science July 2019, Vol.60, 4776. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Bianca S Gerendas, Martina Neschi, Martin Michl, Lupyr Kostiantyn, Gabor Gy Deak, Yvonne Winhofer, Alexandra Kautzky-Willer, Michael David Abramoff, Ursula Schmidt-Erfurth; FDA-authorized autonomous AI for diabetic retinopathy screening in clinical routine. Invest. Ophthalmol. Vis. Sci. 2019;60(9):4776.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : In 2017, there were 425 million patients diagnosed with diabetes, a prevalence estimated to increase to 629 million by 2045. The American Diabetes Association recommends regular eye examinations in people with diabetes for diabetic retinopathy (DR). An autonomous AI screening device (IDx-DR) for the detection of more than mild DR, was cleared by the FDA in 2018 for use in people with diabetes and no visual complaints. The purpose of this study is to evaluate the use of IDx-DR in clinical routine.

Methods : Included were all available patients with type I or II diabetes, without any subjective eye symptoms, no previous diagnosis of DR and no confounding eye disease, who visited the Department of Endocrinology at the Medical University of Vienna for a routine clinical examination (~4000 patients/year). A medical student who had never performed an eye examination, received a four hour training to independently operate the AI system. The acquired images (four 45° color fundus images, for each eye one macula- and one disc-centered) were transferred to the Vienna Reading Center (VRC) and evaluated for DR according to the International Clinical Diabetic Retinopathy severity scale by a certified and masked image grader, as well as a retina specialist. The IDx-DR outputs and the VRC manual grading were then compared and sensitivity/specificity were calculated.

Results : Patient representation was comparable to that of the Austrian diabetes population in terms of gender/age, except for an underrepresentation of diabetes patients aged ≥75, likely due to a higher frequency of accompanying eye conditions. Disease prevalence was 9%. When compared to VRC grading results, IDx-DR identified more than mild DR with a sensitivity of 88.2% and a specificity of 89.0%. Based on the preliminary analysis of the first 200 cases, where only two patients where diagnosed negative by IDx-DR but positive by VRC, the study will be adequately powered with 1670 subjects.

Conclusions : Autonomous AI for DR screening is safe for patients and a useful tool that can be easily applied in clinical routine. While it is not meant to replace the ophthalmologist’s examination, the identification of diabetes patients with no or early signs of DR allows a reallocation of resources to cases that require more intensive management. Automated DR screening is reliable, fast, easy and independent of location and could therefore increase the patients’ compliance.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×