CADS: A Comprehensive Anatomical Dataset and Segmentation for Whole-Body Anatomy in Computed Tomography

Overview

CADS is a robust, fully automated framework for segmenting 167 anatomical structures in Computed Tomography (CT), spanning from head to knee regions across diverse anatomical systems.

The framework consists of two main components:

1. CADS-dataset:

   • 22,022 CT volumes with complete annotations for 167 anatomical structures.
   • Most extensive whole-body CT dataset, exceeding current collections in both scale (18x more CT scans) and anatomical coverage (60% more distinct targets).
   • Data collected from publicly available datasets and private hospital data, spanning 100+ imaging centers across 16 countries.
   • Diverse coverage of clinical variability, protocols, and pathological conditions.
   • Built through an automated pipeline with pseudo-labeling and unsupervised quality control.

2. CADS-model:

   • An open-source model suite for automated whole-body segmentation.
   • Performance validated on both public challenges and real-world hospital cohorts.
   • Available as Python script run (this GitHub repo) for flexible command-line usage.
   • Also available as a user-friendly 3D Slicer plugin with UI interface, simple installation and one-click inference.

For more information on the dataset (data collection, labeling procedures, and model derivatives etc.), please refer to the CADS paper preprint.

Useful Links

• 📄 CADS Paper Preprint
• 🤗 CADS-dataset
• 📦 CADS-model Weights
• 🔧 CADS-model Codebase
• 🛠 CADS-model 3D Slicer Plugin

Format

All images and segmentations are provided in NIfTI format, organized by data source.

The directory structure is as follows:

root/
├── dataset_name/
│   ├── images/         # Original CT volumes
│   ├── segmentations/  # Segmentation masks (indexing see [model labelmap](https://github.com/murong-xu/CADS/blob/main/resources/info/labelmap.md))
│   └── README.md       # Dataset license, citation, and further details

Important Notice

• We are not the original owners of the CT images, except for the BrainCT-1mm and CT-TRI datasets newly released in this project.
• Users should review the corresponding README.md file in each dataset subdirectory before using the data and decide whether to include or exclude that dataset based on their intended use.

Dataset Sources Overview

The CADS-dataset comprises multiple publicly available and private-source datasets, each released under its own license.

The table below summarizes all included sources:

Citation

If you use any component of CADS (CADS-dataset, its curated segmentation masks, pretrained CADS-model, or the 3D Slicer extension), please cite:

@article{xu2025cads,
  title={CADS: A Comprehensive Anatomical Dataset and Segmentation for Whole-Body Anatomy in Computed Tomography},
  author={Xu, Murong and Amiranashvili, Tamaz and Navarro, Fernando and Fritsak, Maksym and Hamamci, Ibrahim Ethem and Shit, Suprosanna and Wittmann, Bastian and Er, Sezgin and Christ, Sebastian M. and de la Rosa, Ezequiel and Deseoe, Julian and Graf, Robert and Möller, Hendrik and Sekuboyina, Anjany and Peeken, Jan C. and Becker, Sven and Baldini, Giulia and Haubold, Johannes and Nensa, Felix and Hosch, René and Mirajkar, Nikhil and Khalid, Saad and Zachow, Stefan and Weber, Marc-André and Langs, Georg and Wasserthal, Jakob and Ozdemir, Mehmet Kemal and Fedorov, Andrey and Kikinis, Ron and Tanadini-Lang, Stephanie and Kirschke, Jan S. and Combs, Stephanie E. and Menze, Bjoern},
  journal={arXiv preprint arXiv:2507.22953},
  year={2025}
}