Table 2 Performance of GTRNet in predicting pathologic T stage across cohorts
From: Interpretable deep learning for multicenter gastric cancer T staging from CT images
Cohort | Accuracy (%) | AUC (95% CI) | Sensitivity | Specificity | PPV | NPV |
|---|---|---|---|---|---|---|
(T1/T2/T3/T4)% | ||||||
Training (n = 953) | 90.1 | 0.98 (0.98-0.99) | 93.8 / 83.8 / 84.9 / 97.9 | 97.7 / 94.6 / 95.2 / 99.3 | 93.4 / 83.5 / 85.6 / 97.9 | 97.9 / 94.7 / 95.0 / 99.3 |
Internal Test (n = 239) | 89.9 | 0.97 (0.95-0.98) | 86.7 / 98.3 / 83.3 / 91.7 | 99.9 /94.4 /94.4/ 97.8 | 99.9 / 85.3 / 83.3 / 93.2 | 95.7 / 99.4 / 94.4 / 97.2 |
External test set 1 (n = 360) | 93.6 | 0.95 (0.93–0.97) | 99.9 /97.8 /83.3 / 93.3 | 97.8 / 94.8 / 99.9 / 98.9 | 93.8 / 86.3 / 99.9 / 96.6 | 99.9 / 99.2 / 94.7 / 97.8 |
External test set 2 (n = 240) | 86.7 | 0.91 (0.88-0.95) | 85.0 / 93.3 / 86.7 /81.7 | 96.7 /93.9 / 95.0 /96.7 | 89.5 / 83.6 / 85.3 / 89.0 | 95.1 / 97.7 / 95.5 / 94.1 |
