Automated morphometric description of human corneal endothelium from in-vivo specular and confocal microscopy

Corneal images acquired by in-vivo specular and confocal microscopy provide clinical information on the cornea endothelium health state. Indeed, the normal hexagonal shape of endothelial cells is usually affected by age and pathologies. At present, the analysis is based on manual or semi-automatic methods and the segmentation of a large number of endothelial cells is required for a meaningful estimation of the clinical parameters. We propose an automated method that detects the centers of endothelial cells by convolving the original image with customized two-dimensional kernels, derives a structure made by connected vertices from the recognized centers using the Euclidean distance, and fine-tunes the derived structure through a genetic algorithm, which combines information about the typical regularity of endothelial cells shape with the pixels intensity of the actual image. The final structure of connected vertices forms a set of polygons that fit the underlying cells contours. From these contours the morphometric parameters of clinical interest can be easily computed. The procedure was applied to 15 images acquired with the SP-3000P (Topcon, Japan) specular microscope and 15 images acquired with the Confoscan4 (Nidek Technologies, Italy) confocal microscope, from both healthy and pathological subjects. Ground truth values for the morphometric parameters were obtained from manually carefully drawn cell contours. Results show that the mean percent absolute difference between the automated and the manual estimate of the clinical parameters is between 2 and 6%, and no statistically significant difference was found between them. The proposed totally automatic method appears capable of detecting contour of hundreds of cells covering a large area, and of obtaining a reliable estimation of the relevant morphometric parameters used in clinical practice.