Table 3 Diagnostic performance of the artificial intelligence (AI) system and endoscopists for detecting early-stage colorectal cancer and precursor lesions.
All endoscopists n = 12 | Experts n = 3 | Fellows n = 5 | Beginners n = 4 | AI | |
|---|---|---|---|---|---|
Sensitivity | 87.40% | 87.40% | 87.40% | 87.10% | 97.30% |
median (range) | (78.9–90.5) | (84.9–90.5) | (86.4–89.9) | (78.9–88.4) | (95.9–98.3) |
Specificity | 96.40% | 97.30% | 96.40% | 93.20% | 99.00% |
median (range) | (89.1–98.2) | (96.4–98.2) | (93.6–98.2) | (89.1–98.2) | (98.6–99.2) |
Processing time | 2.4 sec/image | 2.7 sec/image | 2.2 sec/image | 2.4 sec/image | 0.022 sec/image |
median (range) | (1.5–12.9) | (2.1–4.7) | (1.5–8.7) | (1.7–2.9) |
