Dataset Summary

The largest, fully-annotated abdominal CT dataset to date, including 9,262 CT volumes with annotations for 25 different anatomical structures.

Join the Touchstone Benchmarking Project

The Touchstone Project aims to compare diverse semantic segmentation and pre-training algorithms. We, the CCVL research group at Johns Hopkins University, invite creators of these algorithms to contribute to the initiative. With our support, contributors will train their methodologies on the largest fully-annotated abdominal CT datasets to date. Subsequently, we will evaluate the trained models using a large internal dataset at Johns Hopkins University. If you are the creator of a semantic segmentation or pre-training algorithm and wish to advance medical AI by participating in the Benchmark Project, please reach out to [email protected]. We will provide you further details on the project and explain your opportunities to collaborate in our future publications!

Note for Touchstone Benchmarking Project

This dataset should be only used for the second round of the Touchstone Project, and not to update first-round checkpoints. The first round dataset (5,195 annotated CT volumes, 9 annotated structures) is available at: AbdomenAtlas1.0Mini and AbdomenAtlas1.0MiniBeta

Downloading Instructions

1- Register at Huggingface, accept our terms and conditions, and create an access token:

Create a Huggingface account

Accept our terms and conditions for acessing this dataset: on the top of this page, click on "Expand to review and access", insert your data and click "Agree and access repository")

Create a Huggingface access token and copy it (you will use it in step 3, in paste_your_token_here)

2- Install the Hugging Face library:

pip install huggingface_hub[hf_transfer]==0.24.0
HF_HUB_ENABLE_HF_TRANSFER=1

[Optional] Alternative without HF Trasnsfer (slower)

pip install huggingface_hub==0.24.0

3- Download the dataset:

mkdir AbdomenAtlas
cd AbdomenAtlas
huggingface-cli download BodyMaps/AbdomenAtlas1.0Mini --token paste_your_token_here --repo-type dataset --local-dir .

[Optional] Resume downloading

In case you had a previous interrupted download, just run the huggingface-cli download command above again.

huggingface-cli download BodyMaps/AbdomenAtlas1.0Mini --token paste_your_token_here --repo-type dataset --local-dir .

4- Uncompress:

Uncompress:

bash unzip.sh

Check if the folder AbdomenAtlas/uncompressed contains all cases, from BDMAP_00000001 to BDMAP_00009262. If so, you can delete the original compressed files, running:

bash delete.sh

Paper

AbdomenAtlas-8K: Annotating 8,000 CT Volumes for Multi-Organ Segmentation in Three Weeks
Chongyu Qu¹, Tiezheng Zhang¹, Hualin Qiao², Jie Liu³, Yucheng Tang⁴, Alan L. Yuille¹, and Zongwei Zhou^1,*
¹Johns Hopkins University,
²Rutgers University,
³City University of Hong Kong,
⁴NVIDIA
NeurIPS 2023
paper | code | dataset | annotation | poster

How Well Do Supervised 3D Models Transfer to Medical Imaging Tasks?
Wenxuan Li, Alan Yuille, and Zongwei Zhou^*
Johns Hopkins University
International Conference on Learning Representations (ICLR) 2024 (oral; top 1.2%)
paper | code

Citation

@article{li2024abdomenatlas,
  title={AbdomenAtlas: A large-scale, detailed-annotated, \& multi-center dataset for efficient transfer learning and open algorithmic benchmarking},
  author={Li, Wenxuan and Qu, Chongyu and Chen, Xiaoxi and Bassi, Pedro RAS and Shi, Yijia and Lai, Yuxiang and Yu, Qian and Xue, Huimin and Chen, Yixiong and Lin, Xiaorui and others},
  journal={Medical Image Analysis},
  pages={103285},
  year={2024},
  publisher={Elsevier},
  url={https://github.com/MrGiovanni/AbdomenAtlas}
}

@article{bassi2024touchstone,
  title={Touchstone Benchmark: Are We on the Right Way for Evaluating AI Algorithms for Medical Segmentation?},
  author={Bassi, Pedro RAS and Li, Wenxuan and Tang, Yucheng and Isensee, Fabian and Wang, Zifu and Chen, Jieneng and Chou, Yu-Cheng and Kirchhoff, Yannick and Rokuss, Maximilian and Huang, Ziyan and others},
  journal={arXiv preprint arXiv:2411.03670},
  year={2024},
  url={https://github.com/MrGiovanni/RadGPT}
}

@inproceedings{li2024well,
  title={How Well Do Supervised Models Transfer to 3D Image Segmentation?},
  author={Li, Wenxuan and Yuille, Alan and Zhou, Zongwei},
  booktitle={The Twelfth International Conference on Learning Representations},
  year={2024},
  url={https://github.com/MrGiovanni/SuPReM}
}

@article{qu2023abdomenatlas,
  title={Abdomenatlas-8k: Annotating 8,000 CT volumes for multi-organ segmentation in three weeks},
  author={Qu, Chongyu and Zhang, Tiezheng and Qiao, Hualin and Tang, Yucheng and Yuille, Alan L and Zhou, Zongwei},
  journal={Advances in Neural Information Processing Systems},
  volume={36},
  year={2023},
  url={https://github.com/MrGiovanni/AbdomenAtlas}
}

Acknowledgements

This work was supported by the Lustgarten Foundation for Pancreatic Cancer Research and partially by the Patrick J. McGovern Foundation Award. We appreciate the effort of the MONAI Team to provide open-source code for the community.

License

AbdomenAtlas 1.1 is licensed under CC BY-NC-SA 4.0.

Uploading AbdomenAtlas to HuggingFace

The file AbdomenAtlasUploadMultipleFolders.ipynb has the code we used to upload AbdomenAtlas to Hugging Face. It may be ncessary to run the script multiple times, until it finishes without an uploading error. The uploading script requires PyTorch, huggingface_hub, and Jupyter Notebook.