December 2002
Volume 43, Issue 13
ARVO Annual Meeting Abstract  |   December 2002
An Improved Automated Method for Estimating Ketatocyte Density in Confocal Microscopy
Author Affiliations & Notes
  • JW McLaren
    Ophthalmology Mayo Clinic Rochester MN
  • WM Bourne
    Ophthalmology Mayo Clinic Rochester MN
  • Footnotes
    Commercial Relationships   J.W. McLaren, None; W.M. Bourne, None. Grant Identification: Support: NIH Grant EY02037
Investigative Ophthalmology & Visual Science December 2002, Vol.43, 1714. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      JW McLaren, WM Bourne; An Improved Automated Method for Estimating Ketatocyte Density in Confocal Microscopy . Invest. Ophthalmol. Vis. Sci. 2002;43(13):1714.

      Download citation file:

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

  • Supplements

Abstract: : Purpose: Manual assessment of keratocyte density in confocal images of the corneal stroma is labor-intensive, time consuming, and in many cases is poorly reproducible. In this study we developed and refined a computerized algorithm to identify and count bright objects, (keratocytes) in confocal images of the corneal stroma. Methods: Confocal examinations (Tandem Scanning, Reston, VA) from five normal subjects were assessed manually and by an automated algorithm. Ten clear frames, approximately uniformly distributed through the stroma, were selected from each exam. Cell density in each frame was assessed manually by three observers on two occasions. An algorithm previously developed for identification of cells was refined to identify all bright objects in each confocal field. The cells identified manually in at least 4 of the 6 assessments were matched to the objects identified by the algorithm. Image variables such as object size, contrast, brightness, and brightness-area-product (mean brightness above background multiplied by the object area) of each object were used to discriminate objects that corresponded to cells identified manually from those that did not. Results: In 50 frames, the algorithm identified 2969 objects and the observers identified 1559 ± 77 cells (mean ± SD). The brightness-area-product was the most selective variable to match bright objects to cells identified manually, although densities of cells that appeared tightly packed in the anterior-most frames of the stroma were under-estimated. Excluding these frames, the average standard deviation of differences between manual and automated assessments of density was 2275 cells/mm3 and was similar to that of differences between human observers. Conclusion: The automated identification and selection of cells by using image parameters such as brightness-area-product will provide a more efficient and repeatable means of assessing cell density in confocal images of the stroma. However, discrimination thresholds for these variables may change depending on the characteristics of the cells in the image and the specific confocal microscope and its operating parameters.

Keywords: 374 cornea: stroma and keratocytes • 471 microscopy: confocal/tunneling • 429 image processing 

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.