Reliability and reproducibility of corneal endothelial image analysis by in vivo confocal microscopy

Background: The aim of our study was to evaluate the reliability and reproducibility of in vivo slit-scanning corneal microscopy by endothelial analysis. Methods: We examined 12 normal eyes of 12 persons. We analysed intraobserver differences in group I and interobserver differences in group II by manual and automatic evaluation of endothelial images. We also compared the methods of analysis in the two groups. We observed endothelial density, mean cell area, coefficient of variation and number of cells within the frame. Results: In group I we found no significant difference between the observed parameters either by automatic or manual analysis. In group II no significant difference between the two observers was found using automatic analysis. On manual analysis, however, we observed a significant difference between endothelial density and the number of evaluated cells (P=0.0034 and P=0.0028). We observed a significant difference (P <0.001) between automatic and manual analysis concerning each parameter, with a rather tight linear correlation (Pearson correlation between 0.76 and 0.94). Conclusion: Automatic endothelial analysis was reliable and well reproducible in both - intraobserver and interobserver - groups. By manual evaluation, the clinical significance of interobserver differences can be disregarded. The differences between automatic and manual methods of analysis can be traced back to measurement technical reasons. We observed tight linear correlation between parameters. The data can be described well by linear regression. In vivo slit-scanning corneal microscopy may be an alternative to specular microscopic analyses for clinical use.