Abstract
Purpose :
To investigate the inter-grader and intra-grader reliabilities of graders and an automated algorithm, for cup-disc ratio (CDR) grading using fundus photographs.
Methods :
200 fundus photographs were selected from a database of 3000 photographs taken from patients screened in an eye clinic of a tertiary referral centre. Graders involved include glaucoma specialists(n=2), general ophthalmologists(n=2), optometrists(n=2), family physicians(n=2) and an Automated Algorithm(AA). In total, 2 rounds of CDR grading were held for each grader on 2 different dates, with the photographs presented in random order. CDR values were graded as 0.1-0.9, or as pale disc or ungradable. The grading results of the senior glaucoma specialist were used as the reference benchmark for comparison.
Results :
Statistical analysis results show varying levels of both inter-grader and intra-grader reliabilities among all graders, with no obvious correlation between their level of reliability and their experiences in grading CDR. Intra-Class Coefficient(ICC) values for inter-grader reliabilities range from 0.37 to 0.74, while ICC for intra-grader reliabilities range from 0.47 to 0.87. Receiver Operating Characteristic(ROC) curves were used to assess the capability of the AA in CDR grading. Area under curve(AUC) value of the AA was 0.829(66.3% sensitivity, 89.2% specificity), being very similar to the AUC 0.823 of the junior glaucoma specialist(94.4% sensitivity, 72.1% specificity).
Conclusions :
The AUC values suggest that the Automated Algorithm has the potential to have better performance over human graders in CDR grading of fundus photographs. This may have potential applications as a screening tool for large numbers of people in the community, to pick up asymptomatic glaucoma suspect patients early.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.