<div align="center"> <img src="document/fig_dataset_characteristics.png" alt="AbdomenAtlas2.0" width=100%> </div> <h1 align="center" style="font-size: 60px; margin-bottom: 4px">AbdomenAtlas2.0: The Multi-Tumor Segmentation Dataset</h1> <div align="center">

</div>

We present AbdomemAtlas2.0 (The Multi-Tumor Segmentation Dataset) recently created by JHU. It is a large-scale, multi-institutional dataset, containing 10,135 CT scans with 15,130 tumors annotated across six organs and 5,893 controls. The AI ranks first in Medical Segmentation Decathlon (MSD).

Paper

<b>Scaling Tumor Segmentation: Best Lessons from Real and Synthetic Data</b> <br/> Qi Chen, Xinze Zhou, ...,Yefeng Zheng, Ling Shao, Alan Yuille, Zongwei Zhou<sup>★</sup> <br/> Johns Hopkins University <br/> ICCV 2025 <br/>

AbdomenAtlas2.0 Dataset

git clone https://github.com/BodyMaps/AbdomenAtlas2.0.git
cd AbdomenAtlas2.0
cd data
bash download_AbdomenAtlas2.0_ct.sh # It needs ~400GB storage
bash download_AbdomenAtlas2.0_label.sh

Official training and Validation set

AbdomenAtlas2.0 (n=10,135)

External out-of-distribution test set

Proprietary JHH Pancreatic Dataset
3D-IRCADb Dataset
PANORAMA Dataset
Kipa Dataset

AbdomenAtlas2.0 Benchmark

[!NOTE] We will call for comprehensive baseline methods.

| model | paper | github | P-Sen<sup>†</sup> | T-Sen<sup>‡</sup> | Spe | AUC | DSC | |:---|:---|:---|:---:|:---:|:---:|:---:|:---:| | nnU-Net | | | SuPreM | | | Models Genesis | | | Universal Model | | | UNet++ | | | TransUNet | | | MedNeXt | | | MedFormer | | | UniSeg | | | LHU-Net | |

<sup>†</sup> Patient-wise sensitivity: A case is considered a true positive if the model detects one or more tumors in a patient who has any tumor, regardless of whether the predicted location is accurate. <br/> <sup>‡</sup> Tumor-wise sensitivity: A tumor is considered a true positive only if it is correctly localized. Patients with multiple tumors can contribute multiple true positives.

AbdomenAtlas2.0 Model

[!NOTE] We will release more checkpoints as we receive permission from the respective authors. Stay tuned!

Citation

@inproceedings{chen2025scaling,
  title={Scaling Tumor Segmentation: Best Lessons from Real and Synthetic Data},
  author={Chen, Qi and Zhou, Xinze and Liu, Chen and Chen, Hao and Li, Wenxuan and Jiang, Zekun and Huang, Ziyan and Zhao, Yuxuan and Yu, Dexin and He, Junjun and others},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={24001--24013},
  year={2025},
  url={https://github.com/BodyMaps/AbdomenAtlas2.0}
}

Acknowledgement

This work was supported by the Lustgarten Foundation for Pancreatic Cancer Research, the Patrick J. McGovern Foundation Award, and the National Institutes of Health (NIH) under Award Number R01EB037669. We would like to thank the Johns Hopkins Research IT team in IT@JH for their support and infrastructure resources where some of these analyses were conducted; especially DISCOVERY HPC. Paper content is covered by patents pending.

AbdomenAtlas2.0

Install / Use

README