# Nnunet Medical Seg Eval

> Evaluates a self-adapting U-Net framework for medical image segmentation across multiple 3D and 2D tasks. It probes the model's ability to automatically adapt preprocessing, architecture, and training pipelines to achieve robust segmentation performance without manual tuning. Use when the user wants to benchmark on Medical Segmentation Decathlon (Phase 1), or asks about evaluating this task. Reports Dice score.

- Skill: `qhjqhj00/nnunet-medical-seg-eval` (Agent Skill)
- Install (CLI): `npx skillmds add qhjqhj00/nnunet-medical-seg-eval`
- Raw SKILL.md: https://api.skillmd.com/api/skills/qhjqhj00/nnunet-medical-seg-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/nnunet-medical-seg-eval

---


# nnunet-medical-seg-eval

> nnU-Net: Self-adapting Framework for U-Net-Based Medical Image Segmentation — Isensee et al. (2018) (arXiv:1809.10486, 2018)

## What this evaluates

Evaluates a self-adapting U-Net framework for medical image segmentation across multiple 3D and 2D tasks. It probes the model's ability to automatically adapt preprocessing, architecture, and training pipelines to achieve robust segmentation performance without manual tuning.

## Datasets

- **Medical Segmentation Decathlon (Phase 1)** — total ?; splits: train (-1), val (-1), test (-1)

## Metrics

- `Dice score` **(primary)** — range: [0, 1]
  - Computed per foreground class as 2 * |A ∩ B| / (|A| + |B|), where A and B are predicted and ground truth masks. Averages across classes are typically reported.

## Input / output format

**Input**: 3D or 2D medical image volumes with associated multi-class or binary segmentation masks.

**Output**: Pixel-wise segmentation mask predicting class labels for each voxel/pixel in the input volume.

## Scoring recipe

```python
def dice_score(pred, gt):
    pred = pred.flatten()
    gt = gt.flatten()
    intersection = np.sum(pred * gt)
    return (2.0 * intersection) / (np.sum(pred) + np.sum(gt))
```

## Common pitfalls

- Performance on held-out test sets may drop due to distribution shifts between training and testing data (e.g., BRATS dataset).
- Only a single test submission is permitted per model; ensembling or multiple submissions are explicitly discouraged.
- The U-Net Cascade architecture is only triggered when the 3D U-Net's input patch size cannot fully cover the original image volume.

## Evidence (verbatim from paper)

> As can be seen in Table 2 our phase 1 cross-validation results are robustly recovered on the held-out test set indicating a desired absence of over-fitting. The only dataset that suffers from a dip in performance on all of its foreground classes is BrainTumour.

## Citation

```bibtex
@misc{isensee2018nnunet,
  title={nnU-Net: Self-adapting Framework for U-Net-Based Medical Image Segmentation},
  author={Isensee et al. (2018)},
  year={2018},
  note={arXiv:1809.10486}
}
```

- arXiv: 1809.10486