Morphometry of Cells and Guttae in Subjects With Normal or Guttate Endothelium With a Contour Detection Algorithm

Abstract

This detection algorithm is repeatable and reproducible, and it generates a cell border overlay useful in analyzing the morphology of cells and guttae. The analysis of corneal guttae could become a useful follow-up procedure to discriminate between patients with corneal guttata and Fuchs dystrophy.

Further analysis of assessments of the coefficient of variation of corneal endothelial cell areas from specular microscopic images

“…The outcome of the present study is slightly smaller (with the SD about the group mean estimate being 3.2 per cent) but not much different, although it is a comparison of two independent assessments of 100 cells from the same images using different instruments. The result is consistent with those from several other recent studies, where similar discrepancies in COV data have been reported 13,14,25 . The discrepancy is not so much in the mean value of COV obtained by comparing two instruments or a repeated analysis on a single instrument by the same operator but in the size of the range of the difference in the COV values.…”

“…Such an assessment can be made by comparing the image to a set of reference grids or by actually counting the number of cells included within a specified region of interest (ROI), that is, a manual count of endothelial cell density. With the introduction of newer instruments over the past decade or so, there have been numerous comparisons made to determine the accuracy and/or reliability (repeatability) of various image analysis software systems that could replace a manual counting of the endothelial cells to provide an estimates of the ECD for a patient at a particular assessment 4–30 . Overall, most systems seem to provide reasonable estimates of ECD both when inter‐instrument and/or inter‐operator differences are taken into account.…”

“…This overall outcome probably (that is, that the COV assessment is unpredictable) reflects the substantial sensitivity of the COV calculation to just a few cells in an image being assigned different area values 35,36 . It could well be argued that further studies are required to define the repeatability of COV estimates from the same images and using the same method, although this is indirectly addressed in various other studies, where it was found to be poor 13,14,25 . It should be stressed that the errors or differences are likely to arise from border assignment for a cell domain and not from the actual measurement process that determines the number of pixels within a cell domain.…”

“…The result is consistent with those from several other recent studies, where similar discrepancies in COV data have been reported. 13,14,25 The discrepancy is not so much in the mean value of COV obtained by comparing two instruments or a repeated analysis on a single instrument by the same operator but in the size of the range of the difference in the COV values. With COV values for human endothelia often being in the range of 25 to 50 per cent, a discrepancy of 10 per cent (in the actual calculated value) represents a sizeable error in proportion to the absolute values for the COV.…”

“…There have been many assessments of the reliability of endothelial cell analysis systems to provide estimates of the endothelial cell density (ECD) 4–30 . While some of these reports provide data on the reliability of COV assessments, 11–14,16–18,23,25,26,29,30 these are reported in rather different ways such that comparisons are very difficult. Overall, while modern day endothelial image analysis systems should be able to provide the user‐operator with reasonable estimates of ECD, relatively little attention has been given to the source or cause of differences in COV estimates.…”