May 2006
Volume 47, Issue 13
Free
ARVO Annual Meeting Abstract  |   May 2006
Automating Drusen Analysis With User–Friendly Graphical Interfaces With and Without Artifact Correction
Author Affiliations & Notes
  • K. Uy
    Dept of Ophthalmology, Columbia Univ–Harkness Eye Inst, New York, NY
  • R. Smith
    Dept of Ophthalmology, Columbia Univ–Harkness Eye Inst, New York, NY
  • M. Busuioc
    Dept of Ophthalmology, Columbia Univ–Harkness Eye Inst, New York, NY
  • C. Klaver
    Erasmus University, Rotterdam, The Netherlands
  • D. Despriet
    Erasmus University, Rotterdam, The Netherlands
  • Footnotes
    Commercial Relationships  K. Uy, None; R. Smith, None; M. Busuioc, None; C. Klaver, None; D. Despriet, None.
  • Footnotes
    Support  R01 EY015520 HIGHWIRE EXLINK_ID="47:5:5715:1" VALUE="EY015520" TYPEGUESS="GEN" /HIGHWIRE –01
Investigative Ophthalmology & Visual Science May 2006, Vol.47, 5715. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      K. Uy, R. Smith, M. Busuioc, C. Klaver, D. Despriet; Automating Drusen Analysis With User–Friendly Graphical Interfaces With and Without Artifact Correction . Invest. Ophthalmol. Vis. Sci. 2006;47(13):5715.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose: : Large scale digital analysis of macular drusen images is limited by time consuming pre–processing steps. Herein we report the automation of these steps combined with our published drusen identification algorithm and comparison with human graders. To our knowledge, this is the largest automated study of macular drusen images.

Methods: : 389 color slides graded by experts from the Netherlands genetic isolate study (Erasmus Rucphen Family) were digitized. The images were pre–processed (centration, cropping, color balancing, image sizing) using a Graphical User Interface (GUI). User interaction was limited to clicking on the foveal center and a peripapillary point. Images were then batch–processed for drusen identification(Smith et al, Arch Ophthal, 2005;123:200–206). Images with photographic artifacts (dust spots, flare, etc) were separated and re–processed with an artifact correction algorithm. The results for large drusen were analyzed by weighted kappas.

Results: : Using the GUI system the time spent pre–processing images was decreased by almost 60% with minimal human supervision. Before photographic artifact correction, agreements between human graders and the drusen identification algorithm numbered 189 and disagreements 200 (weighted kappa 0.23). With photographic artifact correction there were 204 agreements and 185 disagreements (weighted kappa 0.25).72 images needed re–processing with artifact correction , but this only yielded 15 more agreements (still 185 disagreements), indication more than one source of error in 57 of these 72 images. The most common remaining sources of disagreements were: borderline drusen size,prominent choroid,hypopigmentation, nerve fiber layer reflections, and confluent intermediate drusen identified as large drusen.

Conclusions: : An automated interface significantly reduces the time needed for digital analysis of macular drusen images and yields fair agreement with human graders. Systematic analysis and resolution of remaining sources of disagreement should improve concordance. A more detailed study on the cause of the discrepancies between human grading and the algorithm will shed light on the strengths and weaknesses of the program.

Keywords: image processing • age-related macular degeneration • drusen 
×
×

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.

×