June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Utilizing ocular images and artificial intelligence to screen for COVID-19
Author Affiliations & Notes
  • Joshua Hohlbein
    Ophthalmology, Brooke Army Medical Center, Fort Sam Houston, Texas, United States
  • Telyn Peterson
    Transitional Year Program, Brooke Army Medical Center, Fort Sam Houston, Texas, United States
  • Don Feeney
    Fortem Genus Inc., North Carolina, United States
  • Judy Feeney
    Fortem Genus Inc., North Carolina, United States
  • Fred Lewis
    Fortem Genus Inc., North Carolina, United States
  • Arshaan Nazir
    National Institute of Technology, Srinagar, India
  • Robert Enzenauer
    Ophthalmology, Pediatrics, University of Colorado Anschutz Medical Campus School of Medicine, Aurora, Colorado, United States
  • Footnotes
    Commercial Relationships   Joshua Hohlbein None; Telyn Peterson None; Don Feeney Fortem Genus Inc., Code O (Owner); Judy Feeney Fortem Genus Inc., Code O (Owner); Fred Lewis Fortem Genus Inc., Code C (Consultant/Contractor), iDetect, Code P (Patent); Arshaan Nazir None; Robert Enzenauer None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 1090. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Joshua Hohlbein, Telyn Peterson, Don Feeney, Judy Feeney, Fred Lewis, Arshaan Nazir, Robert Enzenauer; Utilizing ocular images and artificial intelligence to screen for COVID-19. Invest. Ophthalmol. Vis. Sci. 2023;64(8):1090.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Artificial Intelligence (AI), machine learning (ML), and deep learning (DL) have been a large focus in medical diagnostics research, particularly since the beginning of the COVID-19 pandemic. To date, reverse transcription-polymerase chain reaction (RT-PCR) and antigen testing remain the most sensitive and specific diagnostic tests for COVID-19, but they can often cause delays in proper treatment and can put healthcare workers at risk of exposure. While not yet fully validated, easily acquired bitmap images processed via AI algorithms show promise in aiding diagnosis.

Methods : The iDetect testing application has implemented ML through a Support Vector Machine (SVM) and DL through a Convolutional Neural Network (CNN) that can distinguish patterns for various diseases from eye photos. The accuracy of CNN and SVM for diagnosing COVID-19 was assessed in four phases. Phase 1 was conducted by collecting eye photos from subjects that were simultaneously PCR tested. The datasets were augmented and run through CNN models, with RT-PCR serving as a control. During phase 2, mobile eye imaging devices were utilized to photograph PCR-tested subjects’ eyes and upload them to the secure depository. Phase 3 was a study conducted overseas in collaboration with foreign government officials. It focused on the mass acquisition of photos from PCR-tested subjects by using mobile devices equipped with specialized eye imaging software. The CNN and SVM were trained on 7,288 COVID-19 positive images and 12,529 COVID-19 negative images. Phase 4 is currently ongoing.

Results : The data was analyzed with statical software using standard student t-testing. The performance of the iDetect SVM and algorithms showed an overall Area Under the ROC of 92%. The overall sensitivity of the iDetect is close to 90% in nearly all eye positions, except for up and to the right. The specificity of the iDetect is more than 93% in all eye positions.

Conclusions : This deep learning model shows rates of diagnostic accuracy for COVID-19 comparable to RT-PCR and antigen testing. While differentiating COVID-19 from other subgroups of SARS-CoV-2 hasn’t been examined in detail yet, iDetect’s accuracy and precision continues to improve. Using patient-driven image retrieval for the screening of COVID-19 requires no shipping, has less exposure risk than in-person testing, and can potentially be used anywhere that a camera and internet connection are available.

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

×
×

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.

×