May 2004
Volume 45, Issue 13
Free
ARVO Annual Meeting Abstract  |   May 2004
A Computer–Based Retinal Photo Screening System For Screening Diabetic Retinopathy Using Morphology And Genetic Algorithms
Author Affiliations & Notes
  • B. Raman
    Biomedical Division, Kestrel Corporation, Albuquerque, NM
  • E.S. Bursell
    Joslin Vision Network, Joslin Diabetes Center, Boston, MA
  • M. Wilson
    Biomedical Division, Kestrel Corporation, Albuquerque, NM
  • S.C. Nemeth
    Biomedical Division, Kestrel Corporation, Albuquerque, NM
  • P. Soliz
    Biomedical Division, Kestrel Corporation, Albuquerque, NM
  • Footnotes
    Commercial Relationships  B. Raman, None; E.S. Bursell, Joslin Diabetes Center F; M. Wilson, Kestrel Corporation E; S.C. Nemeth, Kestrel Corporation E; P. Soliz, Kestrel Corporation I, E, P.
  • Footnotes
    Support  NEI Grant # 1R43EY014493
Investigative Ophthalmology & Visual Science May 2004, Vol.45, 4125. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      B. Raman, E.S. Bursell, M. Wilson, S.C. Nemeth, P. Soliz; A Computer–Based Retinal Photo Screening System For Screening Diabetic Retinopathy Using Morphology And Genetic Algorithms . Invest. Ophthalmol. Vis. Sci. 2004;45(13):4125.

      Download citation file:


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

      ×
  • Supplements
Abstract

Abstract: : Purpose: To establish the sensitivity and specificity of a computer–based retinal photo screening system for diabetic retinopathy. Methods: Images from two different protocols and imaging modalities were used to develop and evaluate the Computer–based Screening System. Using a lower resolution camera and without dilation drugs, a total of 188 subjects (376 eyes, 69% with no diabetic retinopathy (NDR) and 31% with diabetic retinopathy (DR)) were studied. The data came from the Joslin Vision Network system that uses a non–mydriatic digital–video color fundus camera to acquire three stereo 45o retinal fields. Digitized 35mm film was used to represent higher resolution images taken with a mydriatic camera. A total of 50 subjects were used with an even distribution of controls and diabetics. Images were first screened automatically for image quality using power spectral density. A trained retinal reader provided ground truth, i.e true positive, false positive (FP), and false negative. FPswere removed using MA characteristics such as aspect ratio and edge sharpness. Information from the three retinal fields was combined using a genetic algorithm to yield a clinical classification. Results: Training was performed on retinal images (non–mydriatic) from 97 subjects (74 with NDR and 23 with DR) producing a sensitivity of 83% and a specificity of 69% for MA detection. A validation was performed on an independent data set (non–mydriatic) of 91 subjects (56 NDR and 35 DR) that produced a sensitivity of 69% and a specificity of 66% for MA detection. Using the high quality, digitized 35mm data, a sensitivity and specificity of greater than 85% was achieved. Conclusions: This project demonstrated that a computer based retinal image screening system has potential for broad–scale screening of diabetic patients. Improved image quality will further increase the sensitivity and specificity of the screening system. Acknowledgments: NEI Grant# 1R43EY014493. Rosansky Foundation, DAMD 17–03–2–0062. The Osiason Educational Foundation.

Keywords: diabetic retinopathy 
×
×

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.

×