Automatic Estimation of Limbal Stem Cell Densities in Cultured Epithelial Cell Microscopy Imaging

Abstract

Limbal stem cell deficiency (LSCD) is a progressive corneal disease that renders the corneal epithelium unable to repair itself, which can lead to the eventual loss of vision. Advances in technology have allowed for the growth of limbal stem cells ex-vivo for the purposes of transplantation. One method used to evaluate the quality of these cultivated cells is cell density, which is typically calculated manually by experts, which is time-consuming and has high inter-rater variability. The goal of this project was to create a tool that automatically calculates cell density from digital images of the cultured cells. Results were compared against annotations from four experts with varying levels of experience. Cell counts had high correlation with expert annotations (r=0.64, p<0.01). When compared to human annotators with lower clinical experience, the algorithm achieved significantly better agreement with highly experienced annotators (r=0.75 vs r=0.19, p<0.01). These results suggest that the automated tool can provide meaningful cell density counts, which can potentially improve annotation consistency and reduce time required for evaluating LSCD cell cultures.