December 2002
Volume 43, Issue 13
Free
ARVO Annual Meeting Abstract  |   December 2002
Improved Blood Vessel Detection for Retinal GDx Images With a Novel Model Based Method
Author Affiliations & Notes
  • KA Vermeer
    Pattern Recognition Group Delft University of Technology Delft Netherlands
  • FM Vos
    Pattern Recognition Group Delft University of Technology Delft Netherlands
  • AM Vossepoel
    Pattern Recognition Group Delft University of Technology Delft Netherlands
  • HG Lemij
    Rotterdam Eye Hospital Rotterdam Netherlands
  • Footnotes
    Commercial Relationships    K.A. Vermeer, Laser Diagnostic Technologies, Inc. F; F.M. Vos, None; A.M. Vossepoel, None; H.G. Lemij, Laser Diagnostic Technologies, Inc. F.
Investigative Ophthalmology & Visual Science December 2002, Vol.43, 999. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      KA Vermeer, FM Vos, AM Vossepoel, HG Lemij; Improved Blood Vessel Detection for Retinal GDx Images With a Novel Model Based Method . Invest. Ophthalmol. Vis. Sci. 2002;43(13):999.

      Download citation file:


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

      ×
  • Supplements
Abstract

Abstract: : Purpose: The GDx (Laser Diagnostic Technologies, San Diego, CA) is a scanning laser polarimeter designed to measure retinal nerve fiber layer thickness in vivo. It has a built-in blood vessel detection algorithm that is used to disregard spurious readings on blood vessel locations, which would otherwise adversely affect the parameter calculations. Since the standard algorithm misses some vessels, comprising those with specular reflection, we designed a better, model-based method. Methods: Blood vessels show up as either single dark line-like structures or two parallel dark line-like structures separated by a specular reflex. We used a simple brightness gap to model the vessel or vessel walls, detect them with a Laplace and thresholding technique and later fill in the inner part. This approach allows for optimization of various parameters according to the specific image properties. We tested our algorithm by making a pixel-by-pixel comparison between the outcome of the algorithm and ten hand-labeled images. Results: Our algorithm yielded a sensitivity of 93% at a specificity of 91%. The sensitivity and specificity of the built-in GDx algorithm were 66% and 97.6%, respectively. Conclusions: Our algorithm outperformed the GDx on sensitivity, without compromising much on specificity. To reduce spurious readings over blood vessel, we think that the sensitivity should be higher than a good specificity. We therefore think that our method can help to improve GDx performance.

Keywords: 429 image processing • 484 nerve fiber layer • 554 retina 
×
×

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.

×