Pediatric Brain Tumor Wsi Eval

pediatric-brain-tumor-wsi-eval

Clinically-Informed Modeling for Pediatric Brain Tumor Classification from Whole-Slide Histopathology Images — Joakim Nguyen et al. (2026) (arXiv:2604.21060, 2026)

What this evaluates

This benchmark evaluates the ability of weakly supervised multiple instance learning models to classify pediatric brain tumors from whole-slide histopathology images. It probes fine-grained diagnostic discrimination across varying class granularities (2 to 7 classes) under conditions of class imbalance and limited data.

Datasets

• Pediatric brain tumor WSI dataset — total ?; splits: train (-1), val (-1), test (-1)

Metrics

• Macro F1 (primary) — range: [0, 1]
  • Unweighted mean of per-class F1-scores. Calculated as 2 * (precision * recall) / (precision + recall) for each diagnostic class, then averaged across all classes regardless of their frequency.
• Weighted F1 — range: [0, 1]
  • F1-score averaged across classes weighted by their support (number of true instances), explicitly accounting for class imbalance in the dataset.

Input / output format

Input: Whole-slide histopathology images (WSIs) processed into image patches. Patient-level labels are assigned via majority voting of slide-level annotations.

Output: Predicted diagnostic class label (e.g., 'Ependymal tumors', 'High-grade brain tumors', 'Non-glial brain tumors', 'Non-neoplastic / normal brain', 'Other low-grade glial tumors', 'Pilocytic astrocytoma', 'DMG H3 mutated', or 'Tumor'/'Non-tumor') or class probabilities.

Scoring recipe

def compute_macro_f1(y_true, y_pred, classes):
    f1_scores = []
    for c in classes:
        tp = sum(1 for t, p in zip(y_true, y_pred) if t == c and p == c)
        fp = sum(1 for t, p in zip(y_true, y_pred) if t != c and p == c)
        fn = sum(1 for t, p in zip(y_true, y_pred) if t == c and p != c)
        prec = tp / (tp + fp) if (tp + fp) > 0 else 0
        rec = tp / (tp + fn) if (tp + fn) > 0 else 0
        f1 = 2 * prec * rec / (prec + rec) if (prec + rec) > 0 else 0
        f1_scores.append(f1)
    return sum(f1_scores) / len(f1_scores)

Common pitfalls

• Splitting data at the slide level instead of the patient level causes severe data leakage and overly optimistic results due to correlated slides from the same patient.
• Class imbalance is significant; using accuracy or unweighted metrics can mask poor performance on minority tumor subtypes.
• Patient-level labels are derived via majority voting of slide-level predictions, which can introduce noise if slide annotations are inconsistent.

Evidence (verbatim from paper)

Model performance was evaluated using comprehensive metrics appropriate for multi-class medical image classification: Per-Class Metrics: For each diagnostic class, we computed precision, recall, and F1-score to assess class-specific discrimination ability. Aggregate Metrics: Overall classification accuracy, macro-averaged precision, recall, and F1-score were calculated to summarize global performance across all classes. Additionally, weighted F1-score was computed to account for class imbalance.

Citation

@misc{nguyen2026clinically,
  title={Clinically-Informed Modeling for Pediatric Brain Tumor Classification from Whole-Slide Histopathology Images},
  author={Joakim Nguyen et al. (2026)},
  year={2026},
  note={arXiv:2604.21060}
}

• arXiv: 2604.21060