Automatic characterization of Meibomian gland morphology

Abstract

Aims/Purpose: To characterize the Meibomian gland (MG) morphology in healthy subjects with an automatic software.Methods: An automatic software was developed for image analysis in R code, consisting of 4 stages: (1) selection of the region of interest, (2) image preprocessing to obtain a binary and a skeleton image, (3) modelling of each MG to fit each MG to a piecewise linear model, and (4) quantification of the following MG features: tortuosity, MG shortening, number of MGs, deviation from a vertical line, and MG area. Meibography images of the upper eyelid were acquired using the Easy Tear View+®. A group of 156 healthy subjects were included, with a mean age and confidence interval (CI) of 43.88 (95% CI: 41.38–46.37) years, range 16–76 years. Reference intervals (RI) were calculated for each MG feature to be used as a diagnostic tool to classify a specific value as normal or non‐normal. Generalized Additive Models for Location, Scale and Shape were used to estimate the percentile curve for each morphologic feature of MG, considering the influence of age and sex.Results: Age had an influence on tortuosity, mean = 0.3 (95% CI: 0.3–0.31), RI: 0.19–0.4 <20 years, and 0.24–0.39 at 50–60 years. Age and sex had an influence in the MG shortening, mean = 63% (95% CI: 62–64%), mean RI: 58–74% <20 years; 59–84% >75 years; women RI: 56–71% <20 years; 57–82% >75 years. The rest of the variables were not affected by age or sex: number of MGs = 16.05 (95% CI: 15.28–16.82), RI: 8–27; vertical deviation = 0.37 (95% CI: 0.36–0.38), RI: 0.29–0.45, and MG area = 14% (95% CI: 14–15%), RI: 8.17–19.81%.Conclusions: Tortuosity variability decreases between 45 and 60 years of age. The MG shortening increases with age and is higher in men. These values could be useful to classify a meibography image as normal or non‐normal, helping to diagnose MG alterations in an objective way.