Table 12.
Training configuration and hyperparameters.
| Parameter | Value |
|---|---|
| Input size | 240 × 240 × 155 |
| Modalities | FLAIR, T1, T1ce, T2 |
| Batch size (per GPU) | 1 |
| GPUs | 2 × RTX 4090 |
| Epochs | 300 |
| Optimizer | AdamW |
| Initial learning rate | 1e−4 |
| Weight decay | 1e−5 |
| LR scheduler | Cosine annealing |
| Gradient clipping | 1.0 |
| Dropout | 0.1 |
| Memory slots (L) | 16 |
| λ1 (alignment loss) | 0.5 |
| λ2 (temporal loss) | 0.3 |
| λs (smoothness loss) | 0.1 |
| Loss function | Dice loss + alignment + temporal |
| Early stopping | Patience = 30 epochs |
| Mixed precision | FP16 (Apex) |
| Random seeds | 3 runs averaged |