May 2005
Volume 46, Issue 13
ARVO Annual Meeting Abstract  |   May 2005
Automated High–Resolution Early Drusen Detection and Measurement
Author Affiliations & Notes
  • T. Smith
    Ophthalmology, Columbia Univ, New York, NY
  • J. Chan
    Ophthalmology, Columbia Univ, New York, NY
  • I. Barbazetto
    Ophthalmology, Columbia Univ, New York, NY
  • D. Despriet
    Ophthalmology, Erasmus Univ, Rotterdam, The Netherlands
  • C. Klaver
    Ophthalmology, Ersamus Univ, Rotterdam, The Netherlands
  • Footnotes
    Commercial Relationships  T. Smith, None; J. Chan, None; I. Barbazetto, None; D. Despriet, None; C. Klaver, None.
  • Footnotes
    Support  New York Community Trust
Investigative Ophthalmology & Visual Science May 2005, Vol.46, 4296. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      T. Smith, J. Chan, I. Barbazetto, D. Despriet, C. Klaver; Automated High–Resolution Early Drusen Detection and Measurement . Invest. Ophthalmol. Vis. Sci. 2005;46(13):4296.

      Download citation file:

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

  • Supplements

Abstract: : Purpose:To develop automated image analysis suitable for measurement and early detection of age–related maculopathy. Methods:We previously developed an accurate, reproducible digital technique (background leveling with a mathematical model) of segmenting large soft drusen in fundus images. A modified approach was used to detect small (less than 65 micron diameter), intermediate or occasional large drusen (over 125 µ) in the 6000– µ region. The primary outcome variables were the areas of drusen detected in the central, middle and outer subfields (1000, 3000, and 6000 micron diameter respectively). The slides were also independently graded by stereo viewing for the same outcomes, estimated as <1% or <10%. High–resolution scanned images (6 microns per pixel) were obtained of 16 color slides (7 subjects) from the Netherlands genetic isolate study. The image was divided into 155 zones. The local background was leveled in each zone by the quadratic polynomial model to a mean gray level of 125. The threshold of 135 (ten gray levels above background) was initially applied to find drusen candidates, then region growing and morphological criteria were optimized for closest agreement with the expert gradings. Because small drusen may be identified only as a few pixels, we allowed candidate lesions to dilate by one pixel in radius, preserving pixels with gray levels 133 and above. We removed any candidate still smaller than 20 µ. The bounding ellipse for remaining candidates was calculated in Matlab 7.0, and we eliminated asymmetric candidates with eccentricity greater than 0.75. For remaining candidates, the area and diameter were taken to be the area and major axis of the bounding ellipse. Results:For the 48 subfields in total there was a 100% agreement with the human grading for drusen area <10%. There was 77% (37/48) agreement with the human grading for drusen area <1%. Disagreement was in all cases due to larger area estimation by the human grader. Conclusions:The automated system may be more precise in estimation of small drusen areas. Because ARM affects a substantial percentage of our population and can be treated with antioxidants, development of automated early detection systems for patients at risk could provide significant public health benefits.

Keywords: imaging/image analysis: clinical • 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.