May 2008
Volume 49, Issue 13
Free
ARVO Annual Meeting Abstract  |   May 2008
Automatic Estimation of Keratocyte Density in Confocal Microscopy of the Cornea
Author Affiliations & Notes
  • A. Ruggeri
    Dept of Information Engineering, University of Padova, Padova, Italy
  • F. Scarpa
    Dept of Information Engineering, University of Padova, Padova, Italy
  • Footnotes
    Commercial Relationships  A. Ruggeri, Nidek Technologies, C; F. Scarpa, Nidek Technologies, F.
  • Footnotes
    Support  None.
Investigative Ophthalmology & Visual Science May 2008, Vol.49, 3929. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      A. Ruggeri, F. Scarpa; Automatic Estimation of Keratocyte Density in Confocal Microscopy of the Cornea. Invest. Ophthalmol. Vis. Sci. 2008;49(13):3929.

      Download citation file:


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

      ×
  • Supplements
Abstract
 
Purpose:
 

To develop an algorithm for the automatic estimation of keratocyte volumetric density in images of corneal stroma acquired with confocal microscopy.

 
Methods:
 

Corneal image sequences were acquired with a clinical confocal microscope (ConfoScan4; Nidek Technologies, Srl., Padova, Italy) in 3 normal subjects. All the images in each sequence were registered to compensate for possible x-y shifts, in order to obtain a 3D stack of z-aligned images. After image enhancement and ROI selection, a custom segmentation procedure, based on a modified Otsu technique, was applied to each image to detect 2D contour and center of keratocytes. Only images from the stromal part of the sequence, i.e., containing no epithelial nor endothelial cells, were used. 101 stroma images were analyzed in subject nr. 1, 68 in nr. 2, and 103 in nr. 3. A clustering procedure was then applied to the images of each stack, so as to identify the 3D contour and center of keratocytes. Each stack of images was then partitioned into 5 adjacent layers: anterior peripheral stroma (0%-10% of stromal depth); anterior (11%-33%), central (34%-66%) and posterior (67%-90%) mid stroma; and posterior peripheral stroma (91%-100%). Keratocytes centers were counted in each layer, to eventually estimate their volumetric density. The entire procedure is fully automatic and requires no user input. To obtain ground truth values of densities to compare with, manual detection of keratocytes on each 2D image and then on the 3D stacks of images was also performed.

 
Results:
 

The percent differences between automatic keratocyte densities and the corresponding manual ones are reported in Table 1 for the 5 layers in each subject.

 
Conclusions:
 

In this limited data set of normal images, very good average automatic vs. manual differences were obtained (range: -2.9%; 6.5%). The least accurate estimations were obtained in the peripheral layers of stroma, with a maximum individual difference of 15%. An extensive evaluation will be performed in a larger set of CS4 sequences, including also pathological subjects, albeit obtaining reliable manual estimations of volumetric densities is quite difficult and time consuming.  

 
Keywords: cornea: stroma and keratocytes • imaging/image analysis: clinical 
×
×

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.

×