June 2017
Volume 58, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2017
Deep Learning System for Screening of Diabetic Retinopathy, Glaucoma and Age-related Macular Degeneration Using Retinal Photographs: The DEEP EYE Study
Author Affiliations & Notes
  • Gilbert Lim
    School of Computing, National University of Singapore, Singapore, Singapore
  • Daniel Shu Wei Ting
    Singapore Eye Research Institute, Singapore National Eye Centre, Singapore, Singapore
    Duke-NUS Graduate Medical School, National University of Singapore, Singapore, Singapore
  • Carol Yim-lui Cheung
    Singapore Eye Research Institute, Singapore National Eye Centre, Singapore, Singapore
    Department of Ophthalmology and Visual Sciences, Chinese University of Hong Kong, Hong Kong, Hong Kong
  • Gavin S Tan
    Singapore Eye Research Institute, Singapore National Eye Centre, Singapore, Singapore
    Duke-NUS Graduate Medical School, National University of Singapore, Singapore, Singapore
  • Rina Rudyanto
    School of Computing, National University of Singapore, Singapore, Singapore
  • Alfred Tau Liang Gan
    Singapore Eye Research Institute, Singapore National Eye Centre, Singapore, Singapore
  • Ching-Yu Cheng
    Singapore Eye Research Institute, Singapore National Eye Centre, Singapore, Singapore
    Duke-NUS Graduate Medical School, National University of Singapore, Singapore, Singapore
  • Wynne Hsu
    School of Computing, National University of Singapore, Singapore, Singapore
  • Mong Li Lee
    School of Computing, National University of Singapore, Singapore, Singapore
  • Tien Yin Wong
    Singapore Eye Research Institute, Singapore National Eye Centre, Singapore, Singapore
    Duke-NUS Graduate Medical School, National University of Singapore, Singapore, Singapore
  • Footnotes
    Commercial Relationships   Gilbert Lim, None; Daniel Shu Wei Ting, None; Carol Cheung, None; Gavin Tan, None; Rina Rudyanto, None; Alfred Gan, None; Ching-Yu Cheng, None; Wynne Hsu, None; Mong Li Lee, None; Tien Yin Wong, None
  • Footnotes
    Support  National Health Innovation Centre (NHIC) Innovation to Develop (I2D) Grant, National Medical Research Council (NMRC), Singapore. (NHIC-I2D-1409022)
Investigative Ophthalmology & Visual Science June 2017, Vol.58, 683. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to Subscribers Only
      Sign In or Create an Account ×
    • Get Citation

      Gilbert Lim, Daniel Shu Wei Ting, Carol Yim-lui Cheung, Gavin S Tan, Rina Rudyanto, Alfred Tau Liang Gan, Ching-Yu Cheng, Wynne Hsu, Mong Li Lee, Tien Yin Wong; Deep Learning System for Screening of Diabetic Retinopathy, Glaucoma and Age-related Macular Degeneration Using Retinal Photographs: The DEEP EYE Study. Invest. Ophthalmol. Vis. Sci. 2017;58(8):683.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Deep learning is a breakthrough machine learning technique that has exhibited substantial promise in image classification. The purpose of this study is to evaluate the diagnostic performance of a deep learning system (DLS) in detecting diabetic retinopathy (DR), age-related macular degeneration (AMD) and glaucoma suspect, using a large-scale national screening population of patients with diabetes in Singapore.

Methods : The DLS was trained for referable DR, AMD and glaucoma on retinal images captured under the Singapore Integrated Diabetic Retinopathy Program (SiDRP; total images=148,266) between 2010-13, followed by validation on images from 2014-15. Referable DR was defined as moderate non-proliferative DR (NPDR) and above, while vision-threatening DR (VTDR) was defined as severe NDPR and above. AMD was defined using the Wisconsin AMD grading system, based on the presence of drusen, retinal pigmentary abnormalities, geographic atrophic atrophy and neovascular changes. Glaucoma suspect was based on a vertical cup-disc ratio of 0.7 and above, or the presence of any glaucomatous disc changes (e.g. disc hemorrhage, focal notching). We calculated the sensitivity and specificity of the DLS with reference to the grading done by professional graders. All retinal images were analyzed twice to evaluate the repeatability of DLS.

Results : Of the 148,266 images, 76,370 were used for training, and 71,896 were used for clinical validation. For DR, the sensitivity and specificity were 0.904 (95% CI=0.872-0.930) and 0.919 (95% CI=0.916-0.922) respectively for referable DR, and 1.000 (95% CI=0.940-1.000) and 0.911 (95% CI=0.908-0.914) respectively for VTDR. For AMD, the sensitivity and specificity were 0.816 (95% CI=0.710-0.896) and 0.817 (95% CI=0.813-0.821) respectively. For glaucoma suspect, the sensitivity and specificity were lower at 0.782 (95% CI=0.716-0.839) and 0.784 (95% CI=0.780-0.788) respectively.

Conclusions : DLS demonstrated promising results in detecting DR, AMD and possibly glaucoma suspects, using retinal photographs from patients with diabetes. Further large-scale testing and validation with additional cohorts from different sources will help in confirming the potential of the DLS for use for screening.

This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.

×
×

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.

×