# Brats2020 Segmentation Eval

> This evaluation probes a model's ability to perform 3D medical image segmentation on brain tumors using MRI scans. It measures how well the architecture delineates tumor boundaries and classifies each voxel as tumor or background across multiple standard segmentation metrics. Use when the user wants to benchmark on BraTS 2020, or asks about evaluating this task. Reports Dice Coefficient.

- Skill: `qhjqhj00/brats2020-segmentation-eval` (Agent Skill)
- Install (CLI): `npx skillmds add qhjqhj00/brats2020-segmentation-eval`
- Raw SKILL.md: https://api.skillmd.com/api/skills/qhjqhj00/brats2020-segmentation-eval/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: qhjqhj00 (https://skillmd.com/u/qhjqhj00)
- Updated: 2026-09-08
- Page: https://skillmd.com/skills/qhjqhj00/brats2020-segmentation-eval

---


# brats2020-segmentation-eval

> Hybrid Multihead Attentive Unet-3D for Brain Tumor Segmentation — Butt et al. (2024) (arXiv:2405.13304, 2024)

## What this evaluates

This evaluation probes a model's ability to perform 3D medical image segmentation on brain tumors using MRI scans. It measures how well the architecture delineates tumor boundaries and classifies each voxel as tumor or background across multiple standard segmentation metrics.

## Datasets

- **BraTS 2020** — total ?; splits: train (-1), test (-1)

## Metrics

- `Dice Coefficient` **(primary)** — range: [0, 1]
  - 2 × Intersection / (Prediction + Ground Truth)
- `Mean IOU` — range: [0, 1]
  - Intersection / Union
- `Accuracy` — range: [0, 1]
  - (TP + TN) / (TP + TN + FP + FN)
- `Precision` — range: [0, 1]
  - TP / (TP + FP)
- `Sensitivity` — range: [0, 1]
  - TP / (TP + FN)
- `Specificity` — range: [0, 1]
  - TN / (TN + FP)

## Input / output format

**Input**: 3D MRI scan volumes

**Output**: Per-voxel predicted probability or binary segmentation mask (0 or 1)

## Scoring recipe

```python
def compute_metrics(pred_mask, gt_mask):
    tp = np.sum((pred_mask == 1) & (gt_mask == 1))
    fp = np.sum((pred_mask == 1) & (gt_mask == 0))
    fn = np.sum((pred_mask == 0) & (gt_mask == 1))
    tn = np.sum((pred_mask == 0) & (gt_mask == 0))
    dice = (2 * tp) / (2 * tp + fp + fn + 1e-6)
    iou = tp / (tp + fp + fn + 1e-6)
    accuracy = (tp + tn) / (tp + tn + fp + fn + 1e-6)
    precision = tp / (tp + fp + 1e-6)
    sensitivity = tp / (tp + fn + 1e-6)
    specificity = tn / (tn + fp + 1e-6)
    return dice, iou, accuracy, precision, sensitivity, specificity
```

## Common pitfalls

- The paper treats segmentation as a binary task (tumor vs. background), whereas standard BraTS benchmarks require region-specific metrics (enhancing tumor, tumor core, whole tumor).
- Pixel-level accuracy is reported, which is heavily skewed by the large background class in 3D MRI volumes and does not reflect boundary delineation quality.

## Evidence (verbatim from paper)

> The Dice Coefficient is another metric that quantifies the similarity between the predicted and ground truth masks. It is defined as: Dice = (2 × Intersection) / (Prediction + Ground Truth)

## Citation

```bibtex
@misc{butt2024hybrid,
  title={Hybrid Multihead Attentive Unet-3D for Brain Tumor Segmentation},
  author={Butt et al. (2024)},
  year={2024},
  note={arXiv:2405.13304}
}
```

- arXiv: 2405.13304