Table 2 Performance of classification algorithm in detection of visually significant cataracts
From: Detecting visually significant cataract using retinal photograph-based deep learning
Detection of visually significant cataractsa | |||
|---|---|---|---|
Testing sets | AUROC (%) (95% CI) | Sensitivity (%) (95% CI) | Specificity (%) (95% CI) |
Internal: | |||
SIMES (n = 72; N = 1,692) | 96.6 (95.5–97.7) | 95.7 (90.5–100.0) | 89.0 (84.7–93.5) |
External: | |||
SCES (n = 141; N = 5,747) | 96.5 (96.0–97.0) | 96.0 (93.1–98.9) | 88.1 (86.5–89.6) |
SINDI (n = 138; N = 5,626) | 96.3 (95.6–96.9) | 94.2 (91.1–97.6) | 90.3 (89.7–91.0) |
BES (n = 48; N = 4,632) | 91.6 (90.2–93.1) | 88.8 (79.5–97.7) | 81.1 (70.5–88.2) |
