Abstract
Purpose :
Precise quantification of corneal epithelial defects is traditionally time-consuming and reliant on highly trained graders. Current literature lacks modern technologies using computer image analysis for reliable grading of corneal epithelial defects. Herein, we describe, validate and compare fully manual and semi-automated methods of quantifying corneal epithelial defects.
Methods :
Primary gaze corneal photographs utilized beam splitter attachments to the slit lamp camera after fluorescein dye was applied to the ocular surface. Photographs utilized cobalt blue light and a Wratten deep yellow 15 filter and were reviewed retrospectively. Manual grading was performed using Image J software by two masked, independent, trained graders to demarcate and quantitate corneal surface areas encompassed by the epithelial defect as well as the corneal limbal circumference. The semi-automated method utilized Python and the OpenCV library. The epithelial defect area is expressed as a percent of the total surface area. Statistical analysis including inter and intraclass correlation coefficients (ICC), ANOVA and Bland Altman plots were calculated using SPSS software.
Results :
Eighty four eye images were subjected to analysis. There was no statistically significant difference between the corneal epithelial defect areas measured by the two methods (manual and semi-automated p ≈ 0.278). The intra-grader ICCs for percent of epithelial defect area to corneal surface area for the manual and semi-automated method was 0.991 and 0.973 respectively with 95% confidence interval of [0.989-0.995]. The inter-grader ICCs for percent of the epithelial defect area to corneal surface area for the manual and semi-automated method was 0.990 and 0.980 respectively with 95% confidence interval of [0.984-0.994].
Conclusions :
We present two reliable and reproducible methods for quantifying corneal epithelial defects using slit lamp photographs with fluorescein staining. Two grading methodologies for measuring epithelial corneal defects relative to corneal surface area are described and validated. The reproducibility of the manual and semi-automated methods are both high. Both methods may be used for standardized evaluation of epithelial corneal defects in clinical research and trials. The semi-automated quantification allows for reduction in grading speed while maintaining excellent reliability.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.