Automated quantification of keratocyte density by using confocal microscopy in vivo

To compare keratocyte density determined by using confocal microscopy with keratocyte density determined in the same corneas by histology. Digital en face images of central corneas were recorded three times by using confocal microscopy in vivo in six New Zealand White rabbits. Bright objects (keratocyte nuclei) in the images were automatically identified by using a custom algorithm to estimate total and regional stromal keratocyte densities. The corneas were then excised, fixed, and sectioned in a sagittal plane for histology. Keratocyte nuclei were manually counted from digitized images of 50 histologic sections per cornea. Total and regional keratocyte densities were estimated from the histologic sections by using stereologic methods based on nuclei per unit area, mean nuclear diameter, and section thickness. Histologic cell densities were corrected for tissue shrinkage. By confocal microscopy, total keratocyte density was 39,000 +/- 1,200 cells/mm3 (mean +/- SE; n = 6); cell density was 47,100 +/- 1,300 cells/mm3 in the anterior stroma and decreased to 27,900 +/- 2,700 cells/mm3 in the posterior stroma (P = 0.004). Analysis of the three separate confocal images of each cornea produced repeatable total cell densities (mean coefficient of variation = 0.035). By histology, total keratocyte density was 37,800 +/- 1,100 cells/mm3, not significantly different from that estimated by confocal microscopy (P = 0.43); anterior cell density was 48,300 +/- 900 cells/mm3 and decreased to 29,400 +/- 900 cells/mm3 posteriorly (P < 0.001). Rabbit keratocyte density estimated by automated analysis of confocal microscopy images in vivo is repeatable and agrees with keratocyte density estimated from histologic sections.