PanTS - The Pancreatic Tumor Segmentation Dataset

[Zongwei Zhou]

Dear Colleagues,

 

BraTS, LiTS, and KiTS have made a big impact on cancer research. Today, PanTS joins this family of open datasets!

 

PanTS (Pancreatic Tumor Segmentation Dataset) is a large-scale, multi-institutional dataset. It includes 36,390 3D CT scans from 145 medical centers, with expert-verified voxel-wise annotations of over 993,000 anatomical structures. These include pancreatic tumors, the pancreas (head, body, and tail), and 24 surrounding organs and structures such as vessels and bones. PanTS is the largest and most detailed resource of its kind and sets a new benchmark for AI development in pancreatic CT analysis.

 

You can download the CT scans and annotations by following the instructions in our GitHub:

 

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

 

Note that the PanTSMini_Label.tar.gz file is provided for early access only. Please do not share it publicly until the official release (e.g., upon paper acceptance).

 

Before the full release, we are inviting feedback from leading researchers like you—especially on the quality of annotations, usefulness in clinical workflows, and value for AI research.

 

One small favor: If you find PanTS useful, please help us spread the word by sharing the GitHub link with colleagues who might be interested in medical image segmentation.

 

 

Thanks and regards,

Zongwei Zhou, PhD

Computer Science, Johns Hopkins University

P: 1-(480)738-2575 | E: zzh...@jh.edu

https://www.zongweiz.com/

Scholar | GitHub | YouTube | Calendly