MitakaKuma/SUMI-OpenCT

SUMI OpenCT

SUMI OpenCT is a curated multi-source chest CT collection prepared for research on distilling photon-counting CT image quality into routine energy-integrating CT. The release is associated with the paper:

Distilling Photon-Counting CT into Routine Chest CT through Clinically Validated Degradation Modeling, arXiv:2604.07329.

The target layout for every CT volume is:

data/
  DATASET_NAME_00000001/
    ct.nii.gz
  DATASET_NAME_00000002/
    ct.nii.gz

Each scan directory contains only ct.nii.gz. Provenance, source identifiers, checksums, and conversion metadata are stored in manifests/.

Safe Collection

Collection is fail-closed: commands are dry runs unless --execute is supplied, executed batches default to one case, and all collectors share a lock. Temporary data is capped at 20 GiB across .hf_tmp, while at least 100 GiB of filesystem free space is reserved and checked throughout every download and extraction. LUNA16 extracts one scan at a time instead of unpacking a complete subset. Converted scans are committed in bounded atomic batches with one manifest update, reducing Hub commit pressure without retaining source images.

Credentials are read only from supported Hugging Face environment variables. Never put API tokens in commands, files, logs, notebooks, or issue text.

See docs/collection_plan.md for source gates and safe dry-run commands. Run python scripts/collection_status.py for an offline readiness check.

Planned Public CT Sources

Dataset	Planned scans from paper	Status policy
DSB17	1,596	Source terms under review before mirroring converted scans
LIDC-IDRI	1,018	TCIA public collection; collect one CT series at a time
LNDb19	324	License conflict requires conservative review before derived NIfTI mirroring
LTRC	1,496	BioLINCC access-controlled; no public raw mirroring without permission
LUNA16	854	CC BY 4.0 source; collect from Zenodo/Grand Challenge
MIDRC	10,496	DUA prohibits redistribution without permission; manifest/instructions only
NLST	422	TCIA/IDC public collection; subset selection to be documented
RSNA-STR	1,110	Source terms prohibit redistribution; manifest/instructions only

Citation

If you use this collection or the associated enhancement outputs, cite the SUMI paper and the original dataset sources listed in LICENSES/source_terms.md.

@article{liu2026sumi,
  title={Distilling Photon-Counting CT into Routine Chest CT through Clinically Validated Degradation Modeling},
  author={Liu, Junqi and Zhou, Xinze and Li, Wenxuan and Ye, Scott and Sitek, Arkadiusz and Yang, Xiaofeng and Tang, Yucheng and Xu, Daguang and Ding, Kai and Wang, Kang and Yang, Yang and Yuille, Alan L. and Zhou, Zongwei},
  journal={arXiv preprint arXiv:2604.07329},
  year={2026}
}

License And Terms

This repository aggregates data derived from multiple public sources. The repository-level license is therefore marked as other; each source keeps its own license, attribution, and usage restrictions. See LICENSES/source_terms.md and the per-scan manifests before reuse.

LIDC-IDRI / TCIA

LIDC-IDRI images are redistributed as derived NIfTI volumes under the source collection's CC BY 3.0 license. Users must cite Armato III et al., Data From LIDC-IDRI, The Cancer Imaging Archive (2015), doi:10.7937/K9/TCIA.2015.LO9QL9SX, and follow the TCIA Data Usage Policies and Restrictions. Downstream mirrors and tools that provide direct access must preserve this dataset attribution, link the TCIA policy, and require their users to do the same. Do not attempt to identify or contact any participant.

LUNA16

LUNA16 volumes are redistributed from the official Grand Challenge release under CC BY 4.0. Cite van Ginneken and Jacobs, LUNA16 Part 1/2, Zenodo, doi:10.5281/zenodo.3723295, together with the LUNA16 challenge publication. Because the LUNA16 images are derived from LIDC-IDRI, preserve the LIDC-IDRI/TCIA attribution and usage obligations listed above as well.