May 2004
Volume 45, Issue 13
ARVO Annual Meeting Abstract  |   May 2004
Quantification of drusen pathology in color fundus images using image processing and color analysis algorithms
Author Affiliations & Notes
  • P. Lundh
    Visual Science, Institute of Ophthalmology, UCL, London, United Kingdom
  • I. Leung
    Reading Centre, Moorfields Eye Hospital, London, United Kingdom
  • L. Huang
    Colour and Imaging Institute, University of Derby, Derby, United Kingdom
  • T. Peto
    Reading Centre, Moorfields Eye Hospital, London, United Kingdom
  • Footnotes
    Commercial Relationships  P. Lundh, None; I. Leung, None; L. Huang, None; T. Peto, None.
  • Footnotes
    Support  none
Investigative Ophthalmology & Visual Science May 2004, Vol.45, 3114. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      P. Lundh, I. Leung, L. Huang, T. Peto; Quantification of drusen pathology in color fundus images using image processing and color analysis algorithms . Invest. Ophthalmol. Vis. Sci. 2004;45(13):3114.

      Download citation file:

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

  • Supplements

Abstract: : Purpose:Total drusen area is a risk factor for the development of visual loss and choroidal neovascularization (CNV) from age–related macular degeneration (AMD). In this study, we evaluate the suitability software developed to measure drusen–load in digitized fundus photographs. Methods:The software design was based on the drusen grading protocol from the International Classification System (ICS) (Bird et al, Surv. Opht. 1995) and it was tested on 50 randomly selected images from the Progression of AMD Study (Sallo et al, ARVO 2003). The images were of a variable, but gradable quality, and all had drusen as the main characteristic change. The segmentation algorithm computed the location, size and number of drusen through a six–step process: The image was first filtered for lighting artifacts and noise; Thereafter, a brightness threshold algorithm was applied locally to segment multiple regions of interest where the probability of a drusen deposit was high. Thereafter, a second hue–based descriptor was applied on these regions to eliminate tissue specular reflections. Local maximum within these regions were taken as seed points and grown with a contrast–driven region–growing algorithm. Degree of confluence was estimated through structural pattern recognition. Once drusen was located the number of, size and distributions were calculated and reported according to the ICS protocol. The results were compared with those of image graders from the Moorfields Eye Hospital Reading Centre using the same protocol. The digital display grading was performed on the higher quality frame from each baseline F2 color stereoscopic pair, which was scanned at 1000 dpi (Nikon Super CoolScan 4000) at 8 bits/channel and displayed on calibrated monitors with XGA resolution. Results:We report a general agreement between the software algorithm and the human graders with an overall correlation coefficient r=0.860 where all differences are statistically non–significant (P < 0.01). Least agreement is found for drusen <125μm due to artifacts generated during the digitalization process. Conclusion:Manual quantification of drusen load is time consuming. The study demonstrates that this semi–automatic software has the potential to assess the change of drusen area in the majority of high–risk patients with age–related maculopathy. Scanning of the slides at higher resolution might help improve on the agreement.

Keywords: drusen • imaging/image analysis: clinical • clinical research methodology 

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.