June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Validation of a deep learning system for the detection of diabetic retinopathy in Indigenous Australians
Author Affiliations & Notes
  • Mark Chia
    Institute of Ophthalmology, University College London, London, London, United Kingdom
    Moorfields Eye Hospital NHS Foundation Trust, London, London, United Kingdom
  • Fred Hersh
    Google Inc, Mountain View, California, United States
  • Rory Sayres
    Google Inc, Mountain View, California, United States
  • Pinal Bavishi
    Google Inc, Mountain View, California, United States
  • Pearse Keane
    Institute of Ophthalmology, University College London, London, London, United Kingdom
    Moorfields Eye Hospital NHS Foundation Trust, London, London, United Kingdom
  • Angus Turner
    University of Western Australia Centre for Ophthalmology and Visual Science, Perth, Western Australia, Australia
    Lions Eye Institute, Nedlands, Western Australia, Australia
  • Footnotes
    Commercial Relationships   Mark Chia None; Fred Hersh Google LLC, Code E (Employment), Alphabet Inc, Code I (Personal Financial Interest); Rory Sayres Google LLC, Code E (Employment), Alphabet Inc, Code I (Personal Financial Interest); Pinal Bavishi Google LLC, Code E (Employment), Alphabet Inc, Code I (Personal Financial Interest); Pearse Keane DeepMind, Roche, Novartis, Apellis, BitFount, Code C (Consultant/Contractor), Big Picture Medical, Code I (Personal Financial Interest), Heidelberg Engineering, Topcon, Allergan, Bayer, Code R (Recipient); Angus Turner None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 2092 – F0081. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Mark Chia, Fred Hersh, Rory Sayres, Pinal Bavishi, Pearse Keane, Angus Turner; Validation of a deep learning system for the detection of diabetic retinopathy in Indigenous Australians. Invest. Ophthalmol. Vis. Sci. 2022;63(7):2092 – F0081.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : The increasing demand for diabetic retinopathy (DR) screening represents a significant burden for eye care services. Deep learning systems (DLS) for DR detection have shown promise in bridging the gap between demand and availability of health resources. However, an important limitation is a tendency for poor generalisability. External validation within populations intended for use is therefore critical. Indigenous Australians are an understudied ethnic group who suffer disproportionately from diabetic-related blindness. This study evaluates the performance of a DLS for DR detection in an Indigenous Australian population.

Methods : We performed a retrospective external validation study comparing the performance of a DLS against a retinal specialist for the detection of more-than-mild (mtmDR) and vision-threatening diabetic retinopathy (vtDR). The validation set consisted of 1682 consecutive, single-field, macula-centered retinal photographs from diabetic patients at an urban Indigenous primary health service in Perth, Australia. Adjudication by a panel of three retinal specialists and trained graders served as the reference standard.

Results : The validation set comprised 1682 eyes of 866 patients (mean age 54.9 years, 52.4% female). For mtmDR detection, sensitivity of the DLS was superior to the retinal specialist (98.0% [95% CI:96.2-99.1%] vs 88.0% [95% CI:84.5-91.2]) with a small reduction in specificity (95.0% [95%CI:93.5-96.5%] vs 97.9% [97.0-98.8%]; both significant by two-sided permutation test, p < 0.001). For vtDR, the DLS’s sensitivity was again superior to the human grader (96.2% [95% CI:93.9-96.3%] vs 84.5% [95% CI:79.3-89.7%], p<0.001) with a slight drop in specificity (95.5% [95%CI:94.3-96.7%] vs 98.3% [95%CI:97.6-99.0%], p<0.001). These results are comparable to reported performance in other populations, including Indian and Thai diabetic screening populations.

Conclusions : In this retrospective sample, the DLS showed improved sensitivity and similar specificity compared to a retinal specialist for the detection of mtmDR and vtDR. This demonstrates the potential of the system to support DR screening amongst Indigenous Australians, an underserved population with a high burden of diabetic eye disease. Further prospective validation in real-world clinical settings is required.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.

 

Figure 1. Flow diagram of image classification by reading center and deep learning system (DLS).

Figure 1. Flow diagram of image classification by reading center and deep learning system (DLS).

 

×
×

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.

×