May 2004
Volume 45, Issue 13
Free
ARVO Annual Meeting Abstract  |   May 2004
Automated detection of split bundles in polarimetric nerve fiber layer images.
Author Affiliations & Notes
  • K.A. Vermeer
    Pattern Recognition Group, Delft University of Technology, Delft, The Netherlands
    Rotterdam Eye Hospital, Rotterdam, The Netherlands
  • N.J. Reus
    Rotterdam Eye Hospital, Rotterdam, The Netherlands
  • F.M. Vos
    Pattern Recognition Group, Delft University of Technology, Delft, The Netherlands
    Department of Radiology, Academic Medical Center, Amsterdam, The Netherlands
  • A.M. Vossepoel
    Pattern Recognition Group, Delft University of Technology, Delft, The Netherlands
  • H.G. Lemij
    Rotterdam Eye Hospital, Rotterdam, The Netherlands
  • Footnotes
    Commercial Relationships  K.A. Vermeer, Laser Diagnostic Technologies, Inc. F; N.J. Reus, Laser Diagnostic Technologies, Inc. F; F.M. Vos, None; A.M. Vossepoel, None; H.G. Lemij, Laser Diagnostic Technologies, Inc. F, C.
  • Footnotes
    Support  none
Investigative Ophthalmology & Visual Science May 2004, Vol.45, 3309. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      K.A. Vermeer, N.J. Reus, F.M. Vos, A.M. Vossepoel, H.G. Lemij; Automated detection of split bundles in polarimetric nerve fiber layer images. . Invest. Ophthalmol. Vis. Sci. 2004;45(13):3309.

      Download citation file:


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

      ×
  • Supplements
Abstract

Abstract: : Purpose: A so–called split bundle is a common variant of nerve fiber layer thickness.1 Split bundles may occur both superiorly and, less frequently, inferiorly. Automated detection of split bundles will facilitate the use of normative maps with smaller variance. The purpose of this research is to develop and test an automated split bundle detection method. Methods: In each image, acquired with the GDx VCC (Laser Diagnostic Technologies, Inc., San Diego, CA), a 10 pixel wide band is extracted superiorly and inferiorly to obtain sectorial TSNIT–plots. These profiles are then approximated by a non–split model consisting of a Gaussian and an offset, and by a split model of two Gaussians and an offset. Statistics derived from these fits were used to classify the bundle as split or non–split using a Parzen density based classifier. The performance of the classifier was estimated by repeated 5–fold cross–validation. For training, 402 images of normal (healthy) eyes were used. An expert classified each bundle in these images as split or non–split, or rejected it in case of doubt. Results: 210 superior bundles were classified by the expert as split, 99 as non–split. Of the inferior bundles, 282 were classified as non–split, 42 as split. Our classifier, assuming equal costs for false positives and false negatives, showed an overall accuracy for superior bundles of 88% (sensitivity 93%, specificity 75%, prevalence 68%). For the inferior bundle, the overall accuracy was 88% (sensitivity 29%, specificity 96%, prevalence 13%). Conclusions: Our automated split bundle detection was able to correctly classify 88% of both superior and inferior bundles. For inferior bundles, this is not much better than simply assuming all bundles are non–split. For superior bundles, however, our detection method allows to apply different non–split and split normal deviation maps with increased specificity. 1. Colen TP, Lemij HG. Prevalence of split nerve fiber layer bundles in healthy eyes imaged with scanning laser polarimetry. Ophthalmology. 2001;108:151–156.

Keywords: image processing • nerve fiber layer • computational modeling 
×
×

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.

×