Table 1.
Diagnostic performance of deep-learning models for binary classification of colorectal lesions on colonoscopic photographs in the test dataset.
| Model | Diagnostic Performance,% (95% CI) | AUC (95% CI) | ||||
|---|---|---|---|---|---|---|
| Accuracy (%) | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | ||
| Neoplastic lesions vs. non-neoplastic lesions | ||||||
| ResNet-152 | 79.4 (78.5–80.3) | 95.4 (93.2–97.6) | 30.1 (25.5–34.7) | 80.8 (78.4–83.2) | 68.8 (58.4–79.2) | 0.821 (0.802–0.840) |
| Inception-ResNet-v2 | 79.5 (77.6–81.4) | 94.1 (92.5–95.7) | 34.1 (28.1–40.1) | 81.6 (80.6–82.6) | 65.0 (54.7–75.3) | 0.832 (0.810–0.854) |
| Advanced colorectal lesions vs. non-advanced colorectal lesions | ||||||
| ResNet-152 | 86.7 (84.9–88.5) | 80.0 (75.4–84.6) | 91.3 (90.8–91.8) | 86.0 (83.7–88.3) | 87.1 (85.1–89.1) | 0.929 (0.927–0.931) |
| Inception-ResNet-v2 | 87.1 (86.2–88.0) | 83.2 (81.5–84.9) | 89.7 (87.7–91.7) | 84.5 (81.0–88.0) | 88.7 (87.7–89.7) | 0.935 (0.929–0.941) |
CI, confidence interval; PPV, positive predictive value; NPV, negative predictive value; AUC, area under the curve.