Table 3.
Per-class performances of the deep learning models in the three-class classification.
| Model/Class | Sensitivity (%) | Specificity (%) | PPV (%) | NPV (%) | F1 Score | AUC (95% CI) |
|---|---|---|---|---|---|---|
| DenseNet-161 | ||||||
| CIN2-3 | 92.0 (86.9–97.1) | 92.4 (85.3–99.6) | 92.5 (87.0–98.0) | 93.4 (90.2–96.7) | 92.1 (88.9–95.3) | 0.981 (0.973–0.989) |
| CIN1 | 80.9 (70.9–90.8) | 96.0 (94.2–97.7) | 87.0 (84.0–89.9) | 94.5 (93.3–95.6) | 83.5 (77.6–89.4) | 0.974 (0.968–0.980) |
| Non-neoplasm | 97.8 (94.2–100.0) | 97.5 (95.6–99.5) | 94.4 (90.0–98.9) | 99.1 (97.6–100.0) | 95.9 (95.5–96.4) | 0.996 (0.992–0.999) |
| EfficientNet-B7 | ||||||
| CIN2-3 | 94.8 (92.8–96.7) | 93.4 (90.1–96.8) | 92.9 (90.3–95.6) | 95.1 (92.3–97.9) | 93.8 (91.7–96.0) | 0.982 (0.971–0.993) |
| CIN1 | 86.1 (82.4–89.7) | 96.4 (95.2–97.5) | 87.6 (81.2–94.0) | 95.6 (94.3–96.9) | 86.8 (82.1–91.4) | 0.979 (0.972–0.985) |
| Non-neoplasm | 94.7 (92.8–96.6) | 98.4 (97.0–99.7) | 96.0 (92.8–99.2) | 97.8 (97.1–98.6) | 95.3 (94.0–96.6) | 0.993 (0.985–1.000) |
PPV, positive predictive value; NPV, negative predictive value; AUC, area under the receiver operating characteristic curve; CI, confidence interval.