June 2015
Volume 56, Issue 7
ARVO Annual Meeting Abstract  |   June 2015
Diabetic Retinopathy Teleretinal Screening: Computer Algorithm versus Reading Center
Author Affiliations & Notes
  • Otis Bennett Walton
    Ophthalmology, Baylor College of Medicine, Houston, TX
  • Jacob Gross
    Medical School, University of Texas at Houston, Houston, TX
  • Alex Young
    Ophthalmology, Baylor College of Medicine, Houston, TX
  • Kathryn Camero
    Ophthalmology, Baylor College of Medicine, Houston, TX
  • Yvonne I Chu
    Ophthalmology, Baylor College of Medicine, Houston, TX
    Harris Health System, Houston, TX
  • Ricci Sanchez
    Harris Health System, Houston, TX
  • Footnotes
    Commercial Relationships Otis Bennett Walton, None; Jacob Gross, None; Alex Young, None; Kathryn Camero, None; Yvonne Chu, None; Ricci Sanchez, None
  • Footnotes
    Support None
Investigative Ophthalmology & Visual Science June 2015, Vol.56, 1421. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Otis Bennett Walton, Jacob Gross, Alex Young, Kathryn Camero, Yvonne I Chu, Ricci Sanchez; Diabetic Retinopathy Teleretinal Screening: Computer Algorithm versus Reading Center. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):1421.

      Download citation file:

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

  • Supplements

Purpose: Use of diabetic teleretinal screening to assess for sight threatening diabetic eye disease (STDED) is increasingly common, but little research has been done to compare the screening effectiveness of computer automated interpretation to manual interpretation. We performed a retrospective chart review for diabetic patients who received diabetic teleretinal imaging, and computer algorithm results were compared with results from an image reading center of ophthalmologists and optometrists.

Methods: In order to increase screening rates among asymptomatic diabetics and direct patients with advanced disease to the eye clinics, Harris Health System (Houston, TX) patients received diabetic teleretinal screening in primary care clinics using non-mydriatic fundus photography. These images were analyzed by a computer algorithm, which classified patients into a binary referral or observation category. Each image was also reviewed by a trained optometrist or ophthalmologist, and the level of diabetic retinopathy was manually assigned. Results from 15,015 diabetic patients were evaluated. Additionally, a supplemental, retrospective chart review was performed for 384 patients determined by the reading center to have STDED, defined by a diagnosis of either severe non-proliferative DR or proliferative DR, so that the clinical result could be compared to both the computer based process and the reading center. Sensitivity and false negative rates were calculated.

Results: Based upon the reading center diagnosis, the sensitivity of the algorithm in detecting STDED was 66.4% (95% confidence interval [CI] 62.8% - 69.9%), with a false negative rate of 2%. When compared to the clinical fundus exam cohort with STDED, the algorithm demonstrated 75% sensitivity (CI, 66.8% - 81.8%), with a false negative rate of 1%. The positive predictive value of the manual reading center in detecting clinically evident STDED was 62% (CI 55.4% - 68.4%).

Conclusions: In light of its low false negative rate, computer algorithm-based diabetic teleretinal screening may be an effective alternative to manual image interpretation in screening for one of the leading causes of preventable blindness.


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.