FLARE21 — MICCAI 2021 FLARE Abdominal Organ Segmentation (re-host)

Re-host of the FLARE21 labeled training set (361 cases) from the Zenodo record 5903672, restructured into the dataset/case_XXXXX/ + train.jsonl layout shared with KiTS19/KiTS23/SLIVER07 so a single Base3DDataset subclass can load it.

Composition

Split	Cases	With mask
train	361	yes

The FLARE21 validation/test sets (hidden, server-side scoring) are not included.

Mask labels

Value	Class
0	background
1	liver
2	kidney
3	spleen
4	pancreas

Files

dataset/case_00000/imaging.nii.gz        # abdominal CT
dataset/case_00000/segmentation.nii.gz   # 0..4 label volume
...
train.jsonl

patient_id is the new contiguous case_XXXXX; source_id keeps the upstream train_NNN id.

License

Per the FLARE21 challenge terms, treat as CC BY-NC-SA 4.0 (non-commercial). (The Zenodo record tags CC-BY-4.0; the challenge data page states CC-BY-NC-SA — the more restrictive non-commercial terms are used here.) Cite:

@article{ma2022flare,
  title   = {Fast and Low-GPU-memory abdomen CT organ segmentation: The FLARE challenge},
  author  = {Ma, Jun and Zhang, Yao and Gu, Song and others},
  journal = {Medical Image Analysis},
  volume  = {82},
  pages   = {102616},
  year    = {2022}
}