June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Improving image-based identification of blinding trachoma for AI applications
Author Affiliations & Notes
  • Christopher J Brady
    Surgery/Ophthalmology, University of Vermont Larner College of Medicine, Burlington, Vermont, United States
  • Lindsay Aldrich
    University of Vermont Larner College of Medicine, Burlington, Vermont, United States
  • Damien Socia
    Surgery/Research, University of Vermont Larner College of Medicine, Burlington, Vermont, United States
  • Sheila K West
    Johns Hopkins Medicine Wilmer Eye Institute, Baltimore, Maryland, United States
  • R. Chase Cockrell
    Surgery/Research, University of Vermont Larner College of Medicine, Burlington, Vermont, United States
  • Footnotes
    Commercial Relationships   Christopher Brady None; Lindsay Aldrich None; Damien Socia None; Sheila West None; R. Chase Cockrell None
  • Footnotes
    Support  NIGMS grant P20GM125498
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 2395. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Christopher J Brady, Lindsay Aldrich, Damien Socia, Sheila K West, R. Chase Cockrell; Improving image-based identification of blinding trachoma for AI applications. Invest. Ophthalmol. Vis. Sci. 2023;64(8):2395.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Trachomatous inflammation—follicular (TF) is the key sign indicating active trachoma for decision making in elimination programs. In recent case studies in Oceania, TF was shown to be less pathognomonic for blinding disease than previously believed, resulting in unnecessary mass distribution of antibiotics. Since patterns of trachomatous scarring (TS) and trachomatous inflammation—intense (TI) were helpful in ruling out blinding trachoma, we sought to analyze two photographic datasets for TF, TI, and TS to develop a more robust characterization of at-risk communities with which to train AI algorithms.

Methods : Two datasets of eyelid photographs from 3 sub-Saharan African countries were used. Dataset 1 was from areas with known high prevalence with TF and TI supplied. Dataset 2 was from an area with low prevalence with TF supplied. Images were graded for the presence of TS (and TI when needed) to develop a ground truth training set for our AI classifier. Proportions of TF, TS, TI were compared between the high and low-prevalence datasets. These proportions were compared to publicly available data from the Solomon Islands. Tools to segment trachomatous features were built in Supervisely to further support training an AI classifier for blinding trachoma.

Results : Dataset frequencies of TF, TI and TS are shown in Fig 1. TI/TF and TS/TF ratios were nearly identical in the areas with known or recent blinding trachoma as compared with the Solomon Islands. Web-based tools allow for rapid consensus image annotation.

Conclusions : Imaging-based diagnostic tools have the promise to enhance the identification of communities at risk of blinding trachoma by rapidly characterizing patterns of clinical findings. Further validation of metrics like TI/TF and TS/TF ratios as well as age-related TS frequencies is needed using worldwide datsets. Exploration of such metrics prior to expensive and invasive interventions could prevent unnecessary treatment. When designing AI classifiers, applying the WHO simplified grading criteria to photographs merely to identify TF would be a missed opportunity. Since trachoma surveys are already often conducted in an mHealth paradigm supported by the Tropical Data program, collection of geocoded photographs would be a natural evolution.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

 

Frequencies of trachoma signs in each dataset

Frequencies of trachoma signs in each dataset

 

Screen capture of follicle annotation tool

Screen capture of follicle annotation tool

×
×

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.

×