Abstract
Purpose: :
To develop a strategy for evaluation of semiautomatic methods for morphometry of corneal endothelium from specular microscopy images.
Methods: :
A simulator that randomly generates corneal endothelial specular microscopy images was developed. The simulator allowed setting of expected values for cell density, coefficient of variation and percentage of hexagonality. Each image was then generated with randomization such that each image is unique while storing the coordinates for the cell borders of each cell, so that real cell density, coefficient of variation of cell size, and percentage of hexagonal cells could be calculated. Filters were then applied so that the final image has the appearance and quality of a specular microscopy image. A total of 500 images were generated within a clinical range of cell density. Out of these, 12 images were selected and transferred to Imagenet-640 software for operator assisted semiautomatic image analysis. The outcome of the semiautomatic analysis was compared with known real data by linear regression for cell density, coefficient of variation of cell size and percentage of hexagonal cells plotting semiautomatic estimates as a function of known values.
Results: :
A 95 % confidence interval for the inclination coefficient was estimated for cell density to 0.98 ± 0.03 (d.f. = 11) for coefficient of variation to 0.57 ± 0.39 (d.f.=11), and for percentage of hexagonal cells to 0.63 ± 0.39 (d.f. = 11) indicating that the semiautomatic method with the Imagenet-640 software is associated with no scaling error for cell density estimates while there is scaling error for estimates of coefficient of variation and percentage of hexagonal cells. The residual standard deviation related to the average was for cell density 0.7-1.7 %, for coefficient of variation estimates 1.5-3.6 % and for percentage of hexagonal cells 4.6- 11 %.
Conclusions: :
The here developed software for simulation of corneal endothelial cells allows objective evaluation of semiautomatic strategies used for morphometry of corneal endothelium from specular microscopy images.
Keywords: cornea: endothelium • computational modeling • microscopy: light/fluorescence/immunohistochemistry