Brats 2024 Post Treatment Eval

---
name: brats-2024-post-treatment-eval
description: Evaluates 3D medical image segmentation models on post-treatment glioma MRI scans, testing their ability to delineate four clinically relevant tumor sub-regions (enhancing tissue, non-enhancing tumor core, surrounding FLAIR hyperintensity, and resection cavity) under treatment-induced anatomical variability and imaging artifacts. Use when the user wants to benchmark on BraTS 2024 Post-Treatment Glioma, or asks about evaluating this task. Reports Dice Similarity Coefficient (DSC).
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2405.18368
  bibtex_key: correia2024brats
  confidence: high
---

# brats-2024-post-treatment-eval

> The 2024 Brain Tumor Segmentation (BraTS) Challenge: Glioma Segmentation on Post-treatment MRI — Maria Correia de Verdier et al. (arXiv:2405.18368, 2024)

## What this evaluates

Evaluates 3D medical image segmentation models on post-treatment glioma MRI scans, testing their ability to delineate four clinically relevant tumor sub-regions (enhancing tissue, non-enhancing tumor core, surrounding FLAIR hyperintensity, and resection cavity) under treatment-induced anatomical variability and imaging artifacts.

## Datasets

- **BraTS 2024 Post-Treatment Glioma** — total 2200; splits: train (-1), val (-1), test (-1)

## Metrics

- `Dice Similarity Coefficient (DSC)` **(primary)** — range: [0, 1]
  - Measures spatial overlap between prediction and ground truth: DSC = 2|A∩B| / (|A|+|B|). Computed per sub-region and averaged across the four classes (ET, NETC, SNFH, RC).
- `Hausdorff Distance 95 (HD95)` — range: mm
  - 95th percentile of the maximum surface distances between prediction and ground truth masks, measuring boundary accuracy.

## Input / output format

**Input**: Four-channel 3D NIfTI volumes corresponding to T1, T1-Gd, T2, and FLAIR MRI sequences, preprocessed with brain extraction and affine registration to the MNI atlas.

**Output**: Four-channel 3D segmentation mask with discrete integer labels for Enhancing Tissue (ET), Non-enhancing Tumor Core (NETC), Surrounding Non-enhancing FLAIR Hyperintensity (SNFH), and Resection Cavity (RC).

## Scoring recipe

```python
def compute_dice(pred, gt):
    intersection = np.sum(pred * gt)
    return 2.0 * intersection / (np.sum(pred) + np.sum(gt) + 1e-8)
# Compute DSC for each of the 4 sub-regions (ET, NETC, SNFH, RC)
# Average the four scores to get the aggregate metric.
# HD95 computed as 95th percentile of pairwise surface distances.
```

## Common pitfalls

- Vessels and choroid plexus are frequently misclassified as Enhancing Tissue (ET).
- Intrinsic T1 hyperintensity in non-enhancing lesions is often incorrectly labeled as ET.
- Subtle SNFH or ET regions are frequently undersegmented.
- White matter changes from microvascular disease unrelated to treatment are mislabeled as SNFH.
- Resection Cavity (RC) and Non-enhancing Tumor Core (NETC) are often confused.
- RC masks incorrectly extend into ventricles or extra-axial spaces.

## Evidence (verbatim from paper)

> The data considered in the 2024 BraTS post-treatment glioma challenge was similar to the paradigm of the BraTS 2021-2023 challenge data, though with modifications specific to the post-treatment setting. The annotation of these data followed a pre-defined clinically approved annotation protocol (defined by expert neuroradiologists and radiation oncologists).

## Citation

```bibtex
@misc{correia2024brats,
  title={The 2024 Brain Tumor Segmentation (BraTS) Challenge: Glioma Segmentation on Post-treatment MRI},
  author={Maria Correia de Verdier et al.},
  year={2024},
  note={arXiv:2405.18368}
}
```

- arXiv: 2405.18368