# CT-LLM dataset download entry points

This page records the official dataset download sources used by the CT-LLM /
Radiology VQA work. It is not a redistributed copy of the data.

Download helper:

```bash
bash download_datasets.sh --dry-run all
bash download_datasets.sh --root /storage/research/artorg_mia/Radiology ctrate radgenome 3drad
```

Most Hugging Face datasets require:

```bash
pip install -U huggingface_hub
huggingface-cli login
```

Kaggle-hosted datasets require the Kaggle API and a configured
`~/.kaggle/kaggle.json`.

| Dataset | Official source | License / note |
|---|---|---|
| CT-RATE / CT-CHAT | <https://huggingface.co/datasets/ibrahimhamamci/CT-RATE> | CC BY-NC-SA 4.0; non-commercial research/education |
| RadGenome-ChestCT | <https://huggingface.co/datasets/RadGenome/RadGenome-ChestCT> | CC BY 4.0 dataset card; CT volumes are CT-RATE-derived |
| 3D-RAD | <https://huggingface.co/datasets/Tang-xiaoxiao/3D-RAD>, <https://github.com/Tang-xiaoxiao/3D-RAD> | Apache-2.0 annotations/code metadata; CT images are CT-RATE-derived |
| M3D-VQA | <https://huggingface.co/datasets/GoodBaiBai88/M3D-VQA> | Apache-2.0 metadata; academic/non-commercial caveats in README |
| CT-Bench | <https://github.com/CelestialShine/CT-Bench>, <https://www.kaggle.com/datasets/zhuqingqingqq/ct-bench> | License not clearly specified in local copy; verify upstream terms |
| RadImageNet-VQA | <https://huggingface.co/datasets/raidium/RadImageNet-VQA> | Apache-2.0 metadata plus RadImageNet Research Use Agreement restrictions |
| SLAKE | <https://www.med-vqa.com/slake/> | No explicit standard license found; use with citation and verify before redistribution |
| VQA-RAD | <https://osf.io/89kps/> | CC0 reported by common mirrors; cite the original VQA-RAD paper |
| VQA-MED / ImageCLEF | <https://www.imageclef.org/2021/medical/vqa> | Challenge-specific terms; VQA-Med 2021 helper repo uses CC BY 4.0 |

Release rule: share code, derived metadata, and benchmark indices where allowed;
do not redistribute raw medical images or volumes, especially CT-RATE-derived,
RadImageNet, M3D/Radiopaedia, or license-unclear sources.