Table 1 Referable Glaucoma AI performance when compared against final diagnosis following comprehensive glaucoma evaluation.
Glaucoma specialist diagnosis (n = 243) | |||||
|---|---|---|---|---|---|
Confirmed Glaucoma | Glaucoma Suspects | Normal | |||
(a) Confusion matrix—AI system versus final diagnosis by Glaucoma specialists | |||||
AI Diagnosis | Referable Glaucoma | 104 (43%) | 15 (6%) | 4 (2%) | |
No Referable Glaucoma | Disc Suspect | 3 (1%) | 19 (8%) | 18 (7%) | |
No Glaucoma | 4 (2%) | 22 (9%) | 54 (22%) | ||
Total | 111 | 56 | 76 | ||
(b) Confusion matrix—AI system versus final diagnosis based on Glaucoma severity (HAP criteria [15]) by the specialists (N = 111 confirmed glaucoma) | |||||
Glaucoma severity diagnosis by specialists | |||||
Early | Moderate | Advanced | |||
AI Diagnosis | Referable Glaucoma | 26 | 22 | 56 | |
No Referable Glaucoma | Disc Suspect | 2 | 1 | ||
No Glaucoma | 2 | 2 | |||
(c) AI performance in the detection of Referable Glaucoma (Final diagnosis) | |||||
Sensitivity | 93.7% (95% CI: 87.6–96.9%) | ||||
Specificity | 85.6% (95% CI: 78.6–90.6%) | ||||
Accuracy | 89.3% (95% CI: 84.7–92.9%) | ||||
Positive likelihood ratio | 6.51 (95% CI: 4.28–9.90) | ||||
Negative likelihood ratio | 0.07 (95% CI: 0.04–0.15) | ||||
Recall- No glaucoma | 94.7% (95% CI: 87.2–97.9%) | ||||
