TotalSegmentator

License

CC BY 4.0
Creative Commons Attribution 4.0 International License

Citation

Paper BibTeX:

@article{wasserthal2023totalsegmentator,
  title={TotalSegmentator: robust segmentation of 104 anatomic structures in CT images},
  author={Wasserthal, Jakob and Breit, Hanns-Christian and Meyer, Manfred T and Pradella, Maurice and Hinck, Daniel and Sauter, Alexander W and Heye, Tobias and Boll, Daniel T and Cyriac, Joshy and Yang, Shan and others},
  journal={Radiology: Artificial Intelligence},
  volume={5},
  number={5},
  pages={e230024},
  year={2023},
  publisher={Radiological Society of North America}
}

Dataset description

The TotalSegmentator dataset comprises 1204 CT images with expert-refined annotations for 104 anatomical structures, including organs, bones, muscles, and vessels. Images were sampled from routine clinical practice, encompassing a variety of pathologies, scanner types, acquisition phases, and institutions, ensuring strong generalizability to real-world applications.

Number of CT volumes: 1203

Contrast: Multiple contrast phases (native, arterial, portal venous, late phase, others); includes dual-energy CT

CT body coverage: Various

Does the dataset include any ground truth annotations?: Yes

Original GT annotation targets: 104 structures (27 organs, 59 bones, 10 muscles, 8 vessels)

Number of annotated CT volumes: 1203

Annotator: AI + human refinement

Acquisition centers: University Hospital Basel

Pathology/Disease: 404 normal patients; 645 with abnormalities (tumor, vascular, trauma, inflammation, bleeding, others)

Original dataset download link: https://zenodo.org/records/6802614

Original dataset format: nifti

Note

This work uses TotalSegmentator dataset version 1.0. We began with v1 early in the project, and it became the basis for subsequent developments. Switching entirely to v2 would require substantial rework; although v2 contains additional images and structures, we consider our current use and planned label release of v1 valid. For model comparisons shown in our paper, we employ the latest TS model to ensure a fair and up-to-date evaluation of strengths and limitations.