Defining an Optimal Sample Size for Corneal Epithelial Immune Cell Analysis Using in vivo Confocal Microscopy Images

Abstract

Purpose: In vivo confocal microscopy (IVCM) images are frequently used to quantify corneal epithelial immune cell (IC) density in clinical studies. There is currently limited evidence to inform the selection of a representative image sample size to yield a reliable IC density estimate, and arbitrary numbers of images are often used. The primary aim of this study was to determine the number of randomly selected, unique IVCM images required to achieve an acceptable level of accuracy when quantifying epithelial IC density, in both the central and peripheral cornea. The secondary aim was to evaluate the consistency and precision of an image selection approach where corneal epithelial IC density was quantified from “three representative images” selected independently by three experienced observers. Methods: All combinations of two to 15 non-overlapping IVCM images were used for deriving IC density estimates, for both the central and peripheral cornea, in 20 healthy participants; the density value from averaging quantifications in the 16 images was defined as the “true mean”. IC density estimates were compared with the true mean in each corneal region using a mean ratio. Intraclass correlation coefficients (ICCs) were used to evaluate the consistency of the mean ratios of IC density estimates derived from the method involving the manual selection of “three representative images” by the observers. The precision of the IC density estimates was compared to a scenario involving three randomly selected images. Results: A total of 12 randomly selected, non-overlapping IVCM images were found to be required to produce a corneal epithelial IC density estimate that was within 30% of the true mean, 95% of the time, for the central cornea; seven such images produced an equivalent level of precision in the peripheral cornea. Mean ratios of corneal IC density estimates derived from “three representative images” methods had poor consistency between observers (ICC estimates <0.5) and similar levels of precision when compared with using three randomly selected images (p > 0.05 for all comparisons), in both the central and peripheral cornea. Conclusions: Data presented in this study can inform image selection methods, and the sample size required for a preferred level of accuracy, when quantifying IC densities in the central and peripheral corneal epithelium using IVCM images.