Table 4. FOSCAL dataset average results for the end-to-end and embedding classification approache. The highest values for each metric across all experiments are highlighted in bold.
| Method | Configuration | Acc (%) | Pre (%) | Sens (%) | F1 (%) | AUC (%) |
|---|---|---|---|---|---|---|
| End-to-end | VGG16 | 96.99 ± 1.10 | 96.62 ± 1.21 | 96.61 ± 1.03 | 96.58 ± 1.11 | 99.50 |
| ResNet-152 | 95.57 ± 5.83 | 95.74 ± 5.53 | 95.79 ± 5.52 | 95.57 ± 5.82 | 98.87 | |
| InceptionV3 | 94.11 ± 4.45 | 94.10 ± 4.46 | 94.08 ± 4.46 | 94.07 ± 4.50 | 98.07 | |
| Embedding | ResNet-152 + RF | 95.11 ± 2.06 | 94.81 ± 3.56 | 95.42 ± 2.96 | 94.67 ± 2.05 | 96.06 |
| ResNet-152 + SVM | 96.00 ± 2.56 | 94.74 ± 2.51 | 96.00 ± 2.12 | 96.46 ± 1.84 | 94.15 |