Investigative Ophthalmology & Visual Science Cover Image for Volume 65, Issue 7
June 2024
Volume 65, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2024
A deep learning model for identification of geographic atrophy using en-face OCT images
Author Affiliations & Notes
  • Reza Jafari
    Topcon Healthcare, Oakland, New Jersey, United States
  • Huiyuan Hou
    Topcon Healthcare, Oakland, New Jersey, United States
  • Tim Steffens
    Topcon Healthcare, Oakland, New Jersey, United States
  • Tony H Ko
    Topcon Healthcare, Oakland, New Jersey, United States
  • Mary Durbin
    Topcon Healthcare, Oakland, New Jersey, United States
  • Footnotes
    Commercial Relationships   Reza Jafari Topcon Healthcare, Code E (Employment); Huiyuan Hou Topcon Healthcare, Code E (Employment); Tim Steffens Topcon Healthcare, Code E (Employment); Tony Ko Topcon Healthcare, Code E (Employment); Mary Durbin Topcon Healthcare, Code E (Employment)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2024, Vol.65, 3745. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Reza Jafari, Huiyuan Hou, Tim Steffens, Tony H Ko, Mary Durbin; A deep learning model for identification of geographic atrophy using en-face OCT images. Invest. Ophthalmol. Vis. Sci. 2024;65(7):3745.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : To present a deep learning algorithm that can automatically detect geographic atrophy (GA). The goal is to enable detection of GA in primary eye care settings, leading to access to intervention and better patient care.

Methods : We trained a computer system with deep learning using spectral domain optical coherence tomography (SD-OCT) en-face images, which were generated by calculating the mean intensity of A-lines of the OCT from 218 eyes with GA and 172 normal eyes, each eye represented by an image from a single optometric practice. These images come from 6mmx6mm macula scans with the Maestro2 (Topcon Healthcare, Tokyo, Japan). A grader reviewed 548 macular OCT en-face images from subjects 75 and older from 6 other optometry practices. The algorithm was applied to this independent test set. The performance of the algorithm was evaluated by metrics such as precision (True Positive/(True Positive+False Positive)), F1 score (2xPrecisionxRecall/(Precision+Recall)), accuracy ((True Positive+True Negative)/Total number of Predictions), sensitivity (recall), and specificity.

Results : A total of 65 images were identified as having GA, 290 showed no evidence of GA, and the remaining images were considered as uncertain or of poor quality. The algorithm demonstrated the precision of 72%, F1 score 58%, and accuracy 87%. Additionally, it showed a sensitivity rate of 48% (31 out of 65) and a specificity of 96% (278 out of 290).

Conclusions : This deep learning-based automated algorithm for GA detection using en-face SD-OCT images was able to accurately identify GA with high specificity, indicating promise of applying artificial intelligence in assisting decision-making in optometric settings, enabling timely patient access to treatment options or clinical trials.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.

×
×

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.

×