July 2019
Volume 60, Issue 9
Open Access
ARVO Annual Meeting Abstract  |   July 2019
Robotics and Artificial Intelligence in the Management of Vision Threatening Disease
Author Affiliations & Notes
  • Ashley Ooms
    Rutgers New Jersey Medical School, Newark, New Jersey, United States
  • Andrew Caterfino
    Rutgers New Jersey Medical School, Newark, New Jersey, United States
  • Nithisha Prasad
    Rutgers New Jersey Medical School, Newark, New Jersey, United States
  • Peter Khouri
    Drexel University, Philadelphia, Pennsylvania, United States
  • Logan Wilson
    Northwestern University, Evanston, Illinois, United States
  • Ben Szirth
    Rutgers New Jersey Medical School, Newark, New Jersey, United States
  • Footnotes
    Commercial Relationships   Ashley Ooms, None; Andrew Caterfino, None; Nithisha Prasad, None; Peter Khouri, None; Logan Wilson, None; Ben Szirth, None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science July 2019, Vol.60, 1486. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Ashley Ooms, Andrew Caterfino, Nithisha Prasad, Peter Khouri, Logan Wilson, Ben Szirth; Robotics and Artificial Intelligence in the Management of Vision Threatening Disease. Invest. Ophthalmol. Vis. Sci. 2019;60(9):1486.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : 36 million Americans have a vision threatening disease (VTD). By 2030, this may reach 58 million. Yet, there are only 5.6 ophthalmologists per 100,000 Americans. Artificial intelligence (AI) software can be used to grade fundus images and determine when remote consultation with a specialist via a telepresence robot (TR) is needed. We aimed to demonstrate that AI can detect diabetic retinopathy (DR) and that TRs can provide direct specialist contact, suggesting together they may capture more VTDs and promptly facilitate their management, especially in remote areas.

Methods : Non-mydriatic 45° color retinal images were captured at a Type 1 Diabetes Mellitus (T1DM) conference (July 2018). Two certified readers (r1, r2) assessed images on gradeability, vertical cup-to-disc ratio, and DR. Images were uploaded to the cloud for grading via a secure AI software (Visulytix’s Pegasus, UK). A TR (DoubleRobotics, Burlingame, CA) was piloted in community screenings for interaction with remote specialists. 14 participants interfaced with the TR and r1 and took a 5-point survey (5PS) on perception of robotics in healthcare. 41 T1DM and 40 non-T1DM participants who only spoke with r1 took a 5PS on their comfort with TRs.

Results : AI graded 197 T1DM patients (age 25 +/- 17.1, T1DM duration 11.4 years +/- 10.7 years, 42% male). Sensitivity, specificity, positive and negative predictive values, and accuracy in DR detection were calculated for r1 and AI after grading 375 images (Fig. 1). DR capture rate as evaluated by r1 and r2 was 10.4%. Overall VTD capture rate was 30.6% (23 cataracts, 4 AMD, 2 glaucoma, 39 DR). A paired t-test did not suggest that those who interacted with the TR had a preference between the TR and r1 (4.57 vs 4.79, p=.19; Fig. 2). An unequal variances t-test suggested that interacting with the TR increased the likelihood of wanting robots involved in one’s healthcare (3.50 vs 2.48, p=.02).

Conclusions : The AI and TR combination presents a novel approach to improving access to ophthalmic care. The TR, having received positive feedback from participants, could facilitate direct interaction with specialists when AI detects DR. Further studies must be done for implementation in telemedicine.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.

 

AI and r1 disagreed 109 times. 81% of AI false positives were due to mislabeling retinal sheen as exudates or hemorrhages.

AI and r1 disagreed 109 times. 81% of AI false positives were due to mislabeling retinal sheen as exudates or hemorrhages.

 

5PS were given to participants > 12 years old. 1=very uncomfortable, 3=neutral, 5=very comfortable.

5PS were given to participants > 12 years old. 1=very uncomfortable, 3=neutral, 5=very comfortable.

×
×

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.

×