Dataset001_BrainMets
Task: Multi-class tumor segmentation (necrotic core, enhancing tumor, edema) on brain MRI (T1, T1c, T2, FLAIR) + primary tumor origin classification (binary: NSCLC vs Breast carcinoma)
Note: The classification head outputs 3 classes (0: NSCLC, 1: Breast carcinoma, 2: Other origin — present in the full training cohort but excluded from this public dataset). Since no label-2 cases appear in the public split, AUROC/AUPRC are computed as binary using the class-1 (Breast carcinoma) probability against class-0 (NSCLC).
Per-Fold Results — Segmentation
| Fold | DSC (necrotic) | DSC (enhancing) | DSC (edema) | DSC mean | NSD (necrotic) | NSD (enhancing) | NSD (edema) | NSD mean |
|---|---|---|---|---|---|---|---|---|
| 0 | 0.693 | 0.541 | 0.633 | 0.622 | 0.761 | 0.659 | 0.710 | 0.710 |
| 1 | 0.634 | 0.605 | 0.485 | 0.575 | 0.719 | 0.732 | 0.602 | 0.684 |
| 2 | 0.591 | 0.664 | 0.544 | 0.599 | 0.679 | 0.828 | 0.733 | 0.746 |
| 3 | 0.589 | 0.553 | 0.399 | 0.514 | 0.629 | 0.669 | 0.527 | 0.608 |
| 4 | 0.621 | 0.601 | 0.633 | 0.619 | 0.713 | 0.699 | 0.697 | 0.703 |
| Mean | 0.626 | 0.593 | 0.539 | 0.586 | 0.700 | 0.717 | 0.654 | 0.690 |
| Std | 0.042 | 0.049 | 0.100 | 0.045 | 0.050 | 0.068 | 0.087 | 0.051 |
Per-Fold Results — Classification
| Fold | AUROC | AUPRC | Bal Acc | F1 |
|---|---|---|---|---|
| 0 | 0.849 | 0.824 | 0.764 | 0.776 |
| 1 | 0.848 | 0.840 | 0.772 | 0.766 |
| 2 | 0.938 | 0.867 | 0.840 | 0.771 |
| 3 | 1.000 | 1.000 | 0.955 | 0.921 |
| 4 | 0.993 | 0.994 | 0.611 | 0.530 |
| Mean | 0.926 | 0.905 | 0.788 | 0.753 |
| Std | 0.074 | 0.085 | 0.125 | 0.141 |
Figures
