Table 5.
Performance comparison among different data augmentation strategies for contrastive learning.
| Pathway | Model | AUC (%) | Acc (%) | Sen (%) | Spec (%) | Mcc |
|---|---|---|---|---|---|---|
| PD-1 | Original_aug | 80.06 | 76.04 | 96.43 | 67.65 | 0.585 |
| Patch shuffle | 72.06 | 73.96 | 78.57 | 72.06 | 0.482 | |
| CLNet (ours) | 86.56 | 84.38 | 92.86 | 80.88 | 0.688 | |
| PD-L1 | Original_aug | 78.57 | 77.08 | 77.50 | 76.79 | 0.554 |
| Patch shuffle | 71.25 | 72.92 | 75.00 | 71.43 | 0.506 | |
| CLNet (ours) | 83.93 | 83.33 | 85.00 | 82.14 | 0.671 |
Original aug denotes the regulate data augmentation strategies used in contrastive learning.
The highest value is in bold.