Table 1 Summary of hyperparameters and training details for Stage I and Stage II models.
Stage | Model | Dataset | Loss functions | Batch size | Optimizer | Learning rate \(\alpha\) | Convergence epochs |
|---|---|---|---|---|---|---|---|
Stage I (Pre-training) | TILSeg-MobileViT | PanopTILs | Focal+Dice(Main Head), MSE (Distance Head) | 8 | Adam | 0.001 | 30 |
Stage I (Fine-Tuning) | TILSeg-MobileViT | OSCC\(_{tcga}\) | Focal+Dice(Main Head), MSE (Distance Head) | 16 | Adam | 0.001 | 20 |
Stage II | Only Image(\(\phi _{img}\)) | Raw Images (OSCC\(_{tcga}\)) | Cross-Entropy | 16 | Adam | 0.001 | 16 |
0.0001 | 6 | ||||||
SGD | 0.001 | 9 | |||||
0.0001 | 30 | ||||||
Stage II | Only Cell density map(\(\phi _{cell}\)) | Cellular Density Maps (OSCC\(_{tcga}\)) | Cross-Entropy | 16 | Adam | 0.001 | 12 |
0.0001 | 8 | ||||||
SGD | 0.001 | 10 | |||||
0.0001 | 8 | ||||||
Stage II | OralTILs-ViT | Combined Features (Raw Images + Cellular Density Maps OSCC\(_{tcga}\)) | Cross-Entropy | 16 | Adam | 0.001 | 10 |
0.0001 | 7 | ||||||
SGD | 0.001 | 8 | |||||
0.0001 | 10 |
