Abstract
Purpose: :
Photo–grading systems exist for many retinal diseases, however, there is less experience with grading corneal photographs. In the Steroid Corneal Ulcer Trial (SCUT), we are randomizing bacterial keratitis patients to treatment with antibiotics and placebo or prednisolone phosphate. We will document the ulcers using digital photography and analyze these images to confirm effective randomization and as a secondary measure of treatment outcomes. We present an analysis of pilot data using a system that we developed for photo–analysis of corneal ulcers.
Methods: :
High resolution digital images were obtained at Aravind Eye Hospital. These images were standardized by normalizing the size and constructing a corneal grading grid, with representation of the limbus, optical zone and clock hours. The images were then graded by 3 expert and 6 non expert graders with respect to; lesion location, lesion area, lesion density, clock hours of corneal neovascularization, clock hours of conjunctival inflammation and hypopion. Image J software was used to determine the area of each lesion and the height of the hypopyon. Each grader evaluated 40 images; 17 were presented once and 23 were presented in both the original and flipped orientation.
Results: :
Experts graded with consistency greater than 90%. Non expert grader consistency varied from 75% to 90%. Inter–observer variability was greater for the non expert graders than the experts. For example, in expert evaluation of ulcer location, variance was detected in 8 of 40 photographs. Twice as much variance was observed among non experts. These trends were similar for all the variables analyzed. On average grading time per photo was 2.75 minutes for experts and 1.7 minutes for non experts. Pooled data revealed that ulcers involved the central 3 mm of the cornea in 87% of the photographs graded and the mean area of lesions was 18 mm2.
Conclusions: :
Expert graders had less intra–observer and inter–observer variability when grading digital corneal photographs of corneal ulcers than did non expert graders. However, individual non expert graders performed with results similar to experts. Expert graders required more time to grade photographs than non expert graders. A high percentage of the ulcers involved the central 3 mm of the cornea. These results will be used to refine our protocol for analysis of digital images for SCUT.
Keywords: cornea: clinical science • bacterial disease