Table 3.
Diagnostic performance of the established algorithm for the prediction of submucosal invasion in the external test dataset.
| Model | AUC | Accuracy (%) | Sensitivity (%) | Specificity (%) | Positive Predictive Value (%) | Negative Predictive Value (%) |
|---|---|---|---|---|---|---|
| Whole dataset | ||||||
| Inception-Resnet-v2 | 0.769 (0.755–0.783) | 74.1 (71.0–77.2) | 72.5 (72.5–72.5) | 74.3 (73.0–75.7) | 64.2 (62.9–65.5) | 81.0 (80.7–81.3) |
| DenseNet−161 | 0.887 (0.863–0.910) | 77.3 (75.4–79.3) | 80.4 (79.6–81.3) | 80.7 (78.5–83.0) | 72.6 (70.1–75.1) | 86.6 (85.9–87.4) |
| EGC (n = 60) | ||||||
| Inception-Resnet-v2 | 0.609 (0.572–0.647) | 65.0 (61.7–68.3) | 58.0 (55.1–60.8) | 62.2 (56.9–67.5) | 52.2 (40.8–63.6) | 70.4 (67.3–73.4) |
| DenseNet−161 | 0.747 (0.712–0.782) | 67.2 (64.4–70.1) | 65.2 (65.2–65.2) | 70.3 (67.2–73.4) | 57.8 (55.3–60.3) | 76.5 (75.7–77.2) |
AUC, area under the curve; EGC, early gastric cancer.