June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Development of Image Recognition Technology to Identify Individuals with Trachomatous Trichiasis Who Need Surgery
Author Affiliations & Notes
  • Emily Gower
    Epidemiology, University of North Carolina at Chapel Hill Gillings School of Global Public Health, Chapel Hill, North Carolina, United States
  • Juan Carlos Prieto
    Psychiatry, University of North Carolina at Chapel Hill School of Medicine, Chapel Hill, North Carolina, United States
  • Hina Shah
    Psychiatry, University of North Carolina at Chapel Hill School of Medicine, Chapel Hill, North Carolina, United States
  • Kasey Jones
    RTI International, Research Triangle Park, North Carolina, United States
  • Rebecca Flueckiger
    RTI International, Research Triangle Park, North Carolina, United States
  • Jerusha Weaver
    Epidemiology, University of North Carolina at Chapel Hill Gillings School of Global Public Health, Chapel Hill, North Carolina, United States
  • Hashiya Kana
    National Eye Centre, Kaduna, Kaduna, Nigeria
  • Robert Chew
    RTI International, Research Triangle Park, North Carolina, United States
  • Footnotes
    Commercial Relationships   Emily Gower None; Juan Prieto None; Hina Shah None; Kasey Jones None; Rebecca Flueckiger None; Jerusha Weaver None; Hashiya Kana None; Robert Chew None
  • Footnotes
    Support  RTI International
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 3829. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Emily Gower, Juan Carlos Prieto, Hina Shah, Kasey Jones, Rebecca Flueckiger, Jerusha Weaver, Hashiya Kana, Robert Chew; Development of Image Recognition Technology to Identify Individuals with Trachomatous Trichiasis Who Need Surgery. Invest. Ophthalmol. Vis. Sci. 2022;63(7):3829.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Trachomatous trichiasis (TT) is the leading infectious cause of blindness, and an estimated 1.8 million currently need surgery to prevent blindness. The disease primarily affects individuals living in rural parts of developing countries, which can make case identification challenging. Surgery is provided primarily through surgical camps, making it necessary to identify cases in advance. Currently, community members receive a brief training on how to identify TT and then are tasks with going door-to-door to identify potential cases. However, their success is limited, with only 15%-30% of potential cases actually having TT, and many cases are not identified. The goal of this project was to develop an image recognition algorithm that can identify TT with at least 90% accuracy. Ultimately, the goal of this project is to provide local communities with the ability to identify TT with a high rate of accuracy, which will increase the efficiency of TT surgery programs.

Methods : We utilized images from an ongoing TT surgery clinical trial in Ethiopia to develop the algorithm. We designed segmentation and patching software to identify the regions of interest (the upper eyelid) and to indicate areas of the eyelid with and without TT. We trained neural networks to classify each eyelid as having TT present or absent. We then created a smartphone app that incorporates the algorithm that can be used by local community members to identify patients with TT. We conducted a field evaluation in Mozambique to determine how well individuals can take high-quality images using the app.

Results : We utilized over 5,000 images of eyes with and without TT to train and test the algorithm. The algorithm currently has a sensitivity of 92% and specificity of 87%. Six individuals were trained on use of the app and they tested the app over a 5-day period in 2 districts in Mozambique. They took 991 images with the app. The majority of the images were of acceptable quality for the algorithm to process. Over 98% of community members were willing to have their eyelids imaged in order to evaluate the algorithm and app.

Conclusions : This study demonstrated strong potential for machine learning to identify eyelids with TT. Future research should evaluate the effectiveness of this approach across multiple trachoma-endemic countries.

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.

×