Brats Peds 2023 Eval

brats-peds-2023-eval

BraTS-PEDs: Results of the Multi-Consortium International Pediatric Brain Tumor Segmentation Challenge 2023 — Fathi Kazerooni et al. (2024) (arXiv:2407.08855, 2024)

What this evaluates

Volumetric segmentation of pediatric brain gliomas using multi-institutional MRI data. It probes a model's ability to accurately delineate tumor sub-regions (enhancing tumor, peritumoral edema, necrotic/cystic core) in 3D MRI scans.

Datasets

• BraTS-PEDs 2023 — total ?; splits: train (-1), val (-1), test (-1)

Metrics

• Dice Score (primary) — range: [0, 1]
  • 2 * |A ∩ B| / (|A| + |B|), measuring voxel-wise overlap between prediction and ground truth for each tumor sub-region.
• Hausdorff Distance 95% (HD95) — range: mm
  • 95th percentile of the maximum surface distance between prediction and ground truth boundaries, measuring spatial boundary accuracy.

Input / output format

Input: Multi-sequence 3D volumetric MRI scans (T1, T1ce, T2, FLAIR) with corresponding tumor segmentation masks.

Output: 3D segmentation mask predicting tumor sub-regions (enhancing tumor, peritumoral edema, necrotic/cystic core).

Scoring recipe

import numpy as np
def dice_score(pred, gt):
    intersection = np.sum(pred * gt)
    return 2.0 * intersection / (np.sum(pred) + np.sum(gt))
def hd95(pred, gt):
    from scipy.spatial.distance import cdist
    surfaces_pred = np.array(np.where(pred > 0)).T
    surfaces_gt = np.array(np.where(gt > 0)).T
    if len(surfaces_pred) == 0 or len(surfaces_gt) == 0:
        return float('inf')
    dists = cdist(surfaces_pred, surfaces_gt)
    return np.percentile(dists, 95)

Common pitfalls

• Confusing pediatric BraTS-PEDs with adult BraTS datasets, which have different tumor distributions and scanner protocols.
• Failing to preprocess MRI sequences (e.g., N4 bias field correction, normalization) consistently across institutions, leading to domain shift.
• Evaluating on 2D slices instead of full 3D volumes, which violates the challenge's volumetric metric requirements.

Evidence (verbatim from paper)

The challenge establishes standardized evaluation metrics including the Dice Score across diverse clinical datasets, enabling reproducible, scalable volumetric analysis critical for clinical trial response assessment in pediatric neuro-oncology.

Citation

@misc{fathikazerooni2024bratspeds,
  title={BraTS-PEDs: Results of the Multi-Consortium International Pediatric Brain Tumor Segmentation Challenge 2023},
  author={Fathi Kazerooni et al. (2024)},
  year={2024},
  note={arXiv:2407.08855}
}

• arXiv: 2407.08855