July 2019
Volume 60, Issue 9
Open Access
ARVO Annual Meeting Abstract  |   July 2019
Autonomous artificial intelligence (AI) reliably detects diabetic retinopathy
Author Affiliations & Notes
  • Stephanie Klein Lynch
    Ophthalmology & Visual Sciences, University of Iowa Hospitals & Clinics, Iowa City, Iowa, United States
  • James C Folk
    Ophthalmology & Visual Sciences, University of Iowa Hospitals & Clinics, Iowa City, Iowa, United States
  • Michael David Abramoff
    Ophthalmology & Visual Sciences, University of Iowa Hospitals & Clinics, Iowa City, Iowa, United States
    IDx Technologies, Inc., Coralville, Iowa, United States
  • Footnotes
    Commercial Relationships   Stephanie Lynch, None; James Folk, IDx LLC (C); Michael Abramoff, IDx LLC (I), IDx LLC (P)
  • Footnotes
    Support  IDx LLC
Investigative Ophthalmology & Visual Science July 2019, Vol.60, 1537. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Stephanie Klein Lynch, James C Folk, Michael David Abramoff; Autonomous artificial intelligence (AI) reliably detects diabetic retinopathy. Invest. Ophthalmol. Vis. Sci. 2019;60(9):1537.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : To determine whether autonomous AI systems can reproducibly and repeatedly diagnose diabetic retinopathy (DR) from fundus photographs.

Methods : Subjects with diabetes were recruited from primary care clinics. The severity of diabetic DR was determined from 4-widefield stereophotography and macular OCT by the Fundus Photography Reading Center (Madison, WI) according to the ETDRS severity scale. 12 subjects with less than or equal to ETDRS level 20 and without macular edema were included sequentially into the DR- arm. 12 subjects with ETDRS level 35 or higher or macular edema were included sequentially into the DR+ arm. Each subject then underwent 10 complete diagnostic sessions with an autonomous AI system (IDx, Coralville, IA) using 3 different operators and 3 different cameras. The final output (DR+, DR- or insufficient quality), was recorded for each session, resulting in 80 reproducibility and 40 repeatability observations for each arm, or a total of 240 observations. A mixed-model with fixed effects of operator, camera, sequence, dilation and DR+/DR- arm was used to test for statistical significance of reproducibility, and intra-subject agreement was used to test for repeatability.

Results : Subjects had an average age of 54.6 (28-74) years. 62.5% were female, 4% Hispanic, and 83% African American. 5/240 observations from a single subject were of insufficient quality and were excluded from further analysis, leaving 235 observations. 234/235 observations yielded the same diagnostic result indepdenent of operator, camera, sequence, dilation, or DR+/DR- status, for an overall repeatability measure of 99.6% (95% CI 95.4-99.9%). For reproducibility, there was no significant difference for operator (p=0.96), camera (p=0.94), or sequence (p=1.0), while the difference between DR-/DR+ arms was highly significant, as expected (p=0.0002).

Conclusions : In a primary care setting, autonomous AI has high repeatability and reproducibility in the diagnosis of DR. This finding supports the diagnostic patient safety of such systems.

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.

×