Table 7.
Results of Public versus RWD Models for the Glaucoma Classification Task
| Data Set | Evaluation Metrics on Test Set (%) | ||||||
|---|---|---|---|---|---|---|---|
| Predictor | Train | Test | Accuracy | SEN | PPV | F 1 | AUROC (95% CI) |
| ResNet-50 | RWD | RWD | 80.6 | 76.9 | 75.4 | 76.1 | 86.1 (84.7–88.1) |
| ResNet-50 | Public | 72.1 | 50.8 | 72.5 | 59.7 | 76.6 (74.5–79.0) | |
| ResNet-50 | RWD + Public | 80.0 | 80.7 | 72.4 | 76.3 | 85.2 (83.7–87.5) | |
| Physician | — | 77.6 | 88.7 | 71.4 | 79.1 | — | |
| ResNet-50 | Public | Public | 87.6 | 90.6 | 76.4 | 82.9 | 95.0 (94.3–96.5) |
| ResNet-50 | RWD | 80.8 | 74.4 | 68.0 | 71.1 | 86.1 (84.5–88.5) | |
| ResNet-50 | RWD + Public | 84.1 | 88.3 | 73.0 | 80.0 | 92.6 (91.7–94.7) | |
| Physician | — | 79.1 | 88.2 | 77.9 | 82.7 | — | |