Development of a Reliable Automated Algorithm for the Morphometric Analysis of Human Corneal Endothelium.

PURPOSE: Corneal images acquired by in vivo microscopy provide important clinical information on the health state of the corneal endothelium. However, the reliable estimation of the clinical morphometric parameters requires the accurate detection of cell contours in a large number of cells. Thus, for the practical application of this analysis in clinical settings, an automated method is needed.
METHODS: We propose the automatic segmentation of corneal endothelial cells contour through an innovative technique based on a genetic algorithm, which combines information about the typical regularity of endothelial cells shape with the pixels intensity of the actual image. The developed procedure is applied to 30 images acquired with the SP-3000P Topcon specular microscope. Automatic assessment of the clinical parameters is then performed by estimating endothelial cell density (ECD, number of cells per unit area), pleomorphism (fraction of hexagonal cells), and polymegethism (fractional standard deviation of cell areas). Ground truth values for these clinical parameters were obtained from cell contours manually drawn by 2 experts.
RESULTS: The mean percent absolute difference between the manual and the automated estimation was 0.6% for ECD, 3.1% for pleomorphism, and 5.3% for polymegethism. Comparable differences were obtained between the estimations provided by the 2 experts (0.5% for ECD, 2.6% for pleomorphism, and 2.9% for polymegethism). No statistically significant difference (P-value > 0.2) was found between automatic and manual assessments of each clinical parameter (power ≥ 77%).
CONCLUSIONS: The proposed totally automatic method seems capable of obtaining a reliable estimation of the relevant morphometric parameters used in clinical practice.