Table 3 Referable Glaucoma AI performance when compared against image grading using FOP and Kowa fundus camera images.
Image grading using FOP fundus camera (n = 229) | Image grading using Kowa fundus Camera (n = 233) | |||||||
|---|---|---|---|---|---|---|---|---|
Likely Glaucoma | Disc Suspect | Unlikely Glaucoma | Likely Glaucoma | Disc Suspect | Unlikely Glaucoma | |||
AI Diagnosis | Referable Glaucoma | 67 (29%) | 26 (11%) | 21 (9%) | 77 (33%) | 24 (10%) | 17 (7%) | |
No Referable Glaucoma | Disc Suspect | 0 | 13 (6%) | 25 (11%) | 0 | 12 (5%) | 25 (11%) | |
No Glaucoma | 0 | 6 (3%) | 71 (31%) | 0 | 5 (2%) | 73 (31%) | ||
(b) AI performance in the detection of Referable Glaucoma (consensus image grading) | ||
|---|---|---|
Image grading using FOP fundus camera | Image grading using Kowa camera | |
Sensitivity | 100 % (95% CI: 94.6–100%) | 100% (95% CI: 95.2–100%) |
Specificity | 71.0% (95% CI: 63.6–77.4%) | 73.7% (95% CI: 66.3–80.0%) |
Accuracy | 79.5% (95% CI: 73.7–84.5%) | 82.4% (95% CI: 76.9–87.1%) |
Positive likelihood ratio | 3.45 (95% CI: 2.71–4.39) | 3.80 (95% CI: 2.93–4.95) |
Negative likelihood ratio | 0.00 | 0.00 |
Recall- no glaucoma | 82.1 (95% CI: 74.1–88.0%) | 85.2% (95% CI: 77.6–90.6%) |
