---
license: cc-by-nc-sa-4.0
task_categories:
  - image-segmentation
tags:
  - medical
  - ct
  - kidney
  - tumor
  - cyst
  - segmentation
  - kits21
  - 3d
size_categories:
  - n<1K
---

# KiTS21 — 2021 Kidney Tumor Segmentation Challenge (re-host)

Re-host of the **KiTS21** public training set (300 cases) in the `dataset/case_XXXXX/` +
`train.jsonl` layout shared with KiTS19 / KiTS23 / SLIVER07. Imaging from
[neheller/KiTS-Challenge-Imaging](https://huggingface.co/datasets/neheller/KiTS-Challenge-Imaging);
ground-truth masks from the official [neheller/kits21](https://github.com/neheller/kits21) repo.

## Ground truth

KiTS21 collected three independent annotations per region; the GT here is the voxel-wise
**majority vote** (`aggregated_MAJ_seg.nii.gz`), the same target used by the official nnU-Net
baseline. AND/OR aggregations are not included.

## Composition

| Split | Cases | With mask |
|-------|------:|----------:|
| train | 300   | yes       |

The held-out test set (100 cases) has withheld masks and is not mirrored.

## Mask labels

| Value | Class      |
|-------|------------|
| 0     | background |
| 1     | kidney     |
| 2     | tumor      |
| 3     | cyst       |

## License

Data: **CC BY-NC-SA 4.0** (non-commercial), per KiTS challenge terms (repo code is MIT). Cite:

```bibtex
@article{heller2023kits21,
  title   = {The KiTS21 Challenge: Automatic segmentation of kidneys, renal tumors,
             and renal cysts in corticomedullary-phase CT},
  author  = {Heller, Nicholas and others},
  journal = {arXiv preprint arXiv:2307.01984},
  year    = {2023}
}
```