June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Comparison between a deep learning algorithm and human expertise predicting the progression of geographic atrophy (GA) secondary to age-related macular degeneration
Author Affiliations & Notes
  • Gregor Sebastian Reiter
    Christian Doppler Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medizinische Universitat Wien, Vienna, Vienna, Austria
  • Dmitrii Lachinov
    Christian Doppler Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medizinische Universitat Wien, Vienna, Vienna, Austria
  • Wolf Bühl
    Department of Ophthalmology and Optometry, Medizinische Universitat Wien, Wien, Wien, Austria
  • Günther Weigert
    Department of Ophthalmology and Optometry, Medizinische Universitat Wien, Wien, Wien, Austria
  • Christoph Grechenig
    Christian Doppler Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medizinische Universitat Wien, Vienna, Vienna, Austria
  • Hrvoje Bogunovic
    Christian Doppler Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medizinische Universitat Wien, Vienna, Vienna, Austria
  • Ursula Schmidt-Erfurth
    Christian Doppler Laboratory for Ophthalmic Image Analysis, Department of Ophthalmology and Optometry, Medizinische Universitat Wien, Vienna, Vienna, Austria
  • Footnotes
    Commercial Relationships   Gregor Reiter RetInSight, Code F (Financial Support); Dmitrii Lachinov None; Wolf Bühl None; Günther Weigert None; Christoph Grechenig None; Hrvoje Bogunovic Apellis, Code F (Financial Support); Ursula Schmidt-Erfurth Apellis, Code F (Financial Support)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 3857. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Gregor Sebastian Reiter, Dmitrii Lachinov, Wolf Bühl, Günther Weigert, Christoph Grechenig, Hrvoje Bogunovic, Ursula Schmidt-Erfurth; Comparison between a deep learning algorithm and human expertise predicting the progression of geographic atrophy (GA) secondary to age-related macular degeneration. Invest. Ophthalmol. Vis. Sci. 2022;63(7):3857.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Geographic atrophy (GA) secondary to age-related macular degeneration is a progressive disease with irreversible loss of visual function. Artificial intelligence (AI)-algorithms are able to monitor disease progression on an individual level. The aim of this study was to investigate the predictive power of AI for GA progression in comparison to retinal expert judgement.

Methods : The set-up of the study consisted of three rounds with two weeks of breaks to prevent memorization, each round entailing two tasks: 1) Experts had to sort GA progression into three subgroups based on anticipated growth rates: slow, intermediate and fast, with a square root growth of <0.2mm, 0.2-0.4mm and >0.4mm, respectively. 2) Experts selected the faster growing case out of two cases. The first round comprises baseline fundus autofluorescence (FAF), the second infrared (IR) and optical coherence tomography (OCT) and the third FAF, IR and OCT images. To compare AI prediction with human experts, FAF annotations of GA performed by a centralized reading center were employed as the reference. For the first task, accuracy and Kappa index between predictions and ground truth were evaluated. For the second task, the concordance index was calculated.

Results : A total of 134 eyes from 134 patients from a phase II clinical trial (FILLY, Apellis) were included, among those 53 were from sham arm and 81 from fellow eyes. Four retinal experts evaluated 120 eyes and 120 eye pairs for respective tasks in each round, resulting in 2880 human gradings. The experts on average reached accuracy of 0.37, 0.43, 0.41 and a Kappa index of 0.05, 0.14, 0.12 on FAF, IR+OCT and FAF+IR+OCT, respectively. On the second task experts achieved a concordance index of 0.62, 0.59 and 0.60. The AI was able to reach an accuracy of 0.48 and Kappa index of 0.21 on the first task, and concordance index of 0.69 on the second task solely utilizing OCT imaging.

Conclusions : Prediction of lesion growth will become an important task for future patient counselling, most importantly after treatments become available. Human gradings improved when the estimation was based on OCT features. However, AI performs this task in a superior manner compared to human experts. AI-supported decisions will guide future management and improve patient care in one of the leading causes for irreversible loss of vision.

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

 

×
×

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.

×