April 2009
Volume 50, Issue 13
Free
ARVO Annual Meeting Abstract  |   April 2009
User Interactive Retinal Image Analysis: Realizing the Practical Digital Promise
Author Affiliations & Notes
  • M. Busuioc
    Ophthalmology, Columbia Univ-Harkness Eye Inst, New York, New York
  • R. T. Smith
    Ophthalmology, Columbia Univ-Harkness Eye Inst, New York, New York
  • R. P. Post
    BME, Columbia University, New York, New York
  • J. Chen
    BME, Columbia University, New York, New York
  • N. Lee
    BME, Columbia University, New York, New York
  • J. Shi
    BME, Columbia University, New York, New York
  • A. Laine
    BME, Columbia University, New York, New York
  • Footnotes
    Commercial Relationships  M. Busuioc, None; R.T. Smith, None; R.P. Post, None; J. Chen, None; N. Lee, None; J. Shi, None; A. Laine, None.
  • Footnotes
    Support  New York Community Trust
Investigative Ophthalmology & Visual Science April 2009, Vol.50, 308. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      M. Busuioc, R. T. Smith, R. P. Post, J. Chen, N. Lee, J. Shi, A. Laine; User Interactive Retinal Image Analysis: Realizing the Practical Digital Promise. Invest. Ophthalmol. Vis. Sci. 2009;50(13):308.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose: : We propose that by taking advantage of the human visual system and expert knowledge, the promised efficiencies of digital methods can be achieved in practice as well as in theory. Thus, simple labeling of regions of interest that is easily performed in a few moments by the human can provide enormous advantage to already well-developed algorithms.

Methods: : Three of our analysis algorithms, drusen segmentation, image registration, and geographic atrophy (GA) segmentation in autofluorescence (AF) images, were integrated with user interactive tools into the Columbia Macular Genetics database of demographic, imaging and genetic information, a modified three tier model with an SQL 2005 Server back-end, with analysis results returned to the repository. The graphical interface has Tool Panels for each task, as well as I/O from the image database and to Excel. Drusen Tool: the user circles up to three regions (confluent drusen, many drusen and few drusen) as input to our established math model for background leveling [US Patent # 7,248,736 B2, 2007], producing an image in which drusen appear on a uniform background for global thresholding. Multimodal Image Registration: our main system for AF, infrared, angiographic or photographic data (J Chen et al. A novel registration method for retinal images. IEEE Internatl Sympos, Vancouver, 08/08) is completely automated. For highly discordant images the user restricts automated corner point matching to suitable vasculature. GA segmentation in AF images: the user indicates in a few broad strokes areas of GA and background for input into the watershed transform (N Lee et al, Interactive segmentation of GA. IEEE Internatl Sympos, Asilomar, 10/08). Each tool was tested on 20 images of fair to good quality.

Results: : Mean drusen segmentation time: user drawing(s) (45 +/- 23 sec), algorithm time (8 sec, vectorized version in Matlab). Mean registration time (4 discordant image pairs): user drawings, both images (39 +/- 9 sec), algorithm (45 sec); 16 image pairs, algorithm only. Mean GA in AF segmentation time: user drawing (34 +/- 15 sec), algorithm time (1 sec). Repeat or manual intervention was needed in 5 drusen images, no registration tasks, and in 3 GA images.

Keywords: imaging/image analysis: clinical • retinal degenerations: hereditary • 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.

×