PanTS: The Pancreatic Tumor Segmentation Dataset

We present PanTS (The Pancreatic Tumor Segmentation Dataset) recently created by JHU. It is a large-scale, multi-institutional dataset, containing 36,390 three-dimensional CT volumes from 145 medical centers, with expert-validated, voxel-wise annotations of over 993,000 anatomical structures, covering pancreatic tumors, pancreas head, body, and tail, and 24 surrounding anatomical structures such as vascular/skeletal structures and abdominal/thoracic organs.

As the largest and most comprehensive resource of its kind, PanTS offers a new benchmark for developing and evaluating AI models in pancreatic CT analysis.

Paper

PanTS: The Pancreatic Tumor Segmentation Dataset
Wenxuan Li, Xinze Zhou, Qi Chen, ..., Alan Yuille, Zongwei Zhou^★
Johns Hopkins University
NeurIPS 2025

PanTS Dataset

git clone https://github.com/MrGiovanni/PanTS.git
cd PanTS
bash download_PanTS_data.sh # It needs ~300GB storage
bash download_PanTS_label.sh http://www.cs.jhu.edu/~zongwei/dataset/PanTSMini_Label.tar.gz

Official training set

• PanTS-tr (n=9,000)

Official in-distribution test set

• PanTS-te (n=901)

External out-of-distribution test set

Note

To submit your model for evaluation, please email Dr. Zongwei Zhou ([email protected]) with:

• Your model checkpoint
• Testing script
• A brief README with usage instructions

The JHU Team will assess performance on three large out-of-distribution test sets and a RSNA Dataset, for which all pancreatic tumors have been re-annotated by our team.

• Proprietary UCSF Pancreatic Dataset (n=13,458)
• Proprietary Polish Pancreatic Dataset (n=5,259)
• Proprietary Peking University Dataset (n=3,066)
• RSNA Abdominal Trauma Detection Dataset (n=4,706)

PanTS Benchmark (official in-distribution test set)

Note

We are calling for more baseline methods.

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

^† 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.
^‡ 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.

PanTS Model

Note

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

Citation

@article{li2025pants,
  title={PanTS: The Pancreatic Tumor Segmentation Dataset},
  author={Li, Wenxuan and Zhou, Xinze and Chen, Qi and Lin, Tianyu and Bassi, Pedro RAS and Plotka, Szymon and Cwikla, Jaroslaw B and Chen, Xiaoxi and Ye, Chen and Zhu, Zheren and others},
  journal={arXiv preprint arXiv:2507.01291},
  year={2025},
  url={https://github.com/MrGiovanni/PanTS}
}

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.