ATM22 — Airway Tree Modeling Challenge 2022 (re-mirror)

Re-hosted mirror of the ATM22 challenge training set (Zhang et al., Medical Image Analysis 2023, arXiv:2303.05745), originally distributed through the ATM22 Grand Challenge. This mirror rebuilds the data from the authors' own Zenodo re-publication under CC BY 4.0, restructured into the same layout we use for KiTS23 / KiPA22 / AbdomenCT1K / etc. so a single Base3DDataset subclass can load it.

Composition

Split	Cases	With mask
train	299	yes

Original challenge composition: 300 train + 50 val (image-only) + 150 test (withheld). This mirror ships train only (300 cases minus 1 for ATM_164 whose label is misaligned with its corresponding image per challenge errata = 299 usable cases). The val and test sets have no public masks and so cannot be evaluated locally; refetch them from Zenodo imagesVal.rar if you want inference-only inputs.

File layout

dataset/ATM_001/
  imaging.nii.gz
  segmentation.nii.gz
...
dataset/ATM_500/
train.jsonl
README.md

train.jsonl lists one entry per case with image, mask, label, modality, dataset, official_split, patient_id keys. Image/mask paths are prefixed with data/nii/ATM22/ so they slot directly into the EasyMedSeg Base3DDataset HF_JSONL_PREFIX convention.

Mask labels

Binary airway mask:

Value	Class
0	background
1	airway

The airway annotation includes trachea, main bronchi, lobar bronchi, and segmental bronchi as one composite class (no branch-level subclasses in the public release).

License

CC BY 4.0, inherited from the upstream Zenodo mirror. The official challenge organizers' research-only terms apply on top. Cite the benchmark paper:

@article{zhang2023multi,
  title   = {Multi-site, multi-domain airway tree modeling (ATM'22):
             A public benchmark for pulmonary airway segmentation},
  author  = {Zhang, Minghui and Wu, Yangqian and Zhang, Hanxiao and others},
  journal = {Medical Image Analysis},
  year    = {2023},
  doi     = {10.1016/j.media.2023.102957}
}