SAROS: A dataset for whole-body region and organ segmentation in CT imaging

License

Mix of CC BY 3.0, CC BY 4.0, and CC BY-NC 3.0.

Citation

Paper BibTeX:

@article{koitka2024saros,
  title={SAROS: A dataset for whole-body region and organ segmentation in CT imaging},
  author={Koitka, Sven and Baldini, Giulia and Kroll, Lennard and van Landeghem, Natalie and Pollok, Olivia B and Haubold, Johannes and Pelka, Obioma and Kim, Moon and Kleesiek, Jens and Nensa, Felix and others},
  journal={Scientific Data},
  volume={11},
  number={1},
  pages={483},
  year={2024},
  publisher={Nature Publishing Group UK London}
}

Dataset:

Koitka, S., Baldini, G., Kroll, L., van Landeghem, N., Haubold, J., Sung Kim, M., Kleesiek, J., Nensa, F., & Hosch, R. (2023). SAROS – A large, heterogeneous, and sparsely annotated segmentation dataset on CT imaging data (SAROS) (Version 2) [Data set]. The Cancer Imaging Archive. https://doi.org/10.25737/SZ96-ZG60

Dataset description

Sparsely Annotated Region and Organ Segmentation (SAROS) is a large, heterogeneous CT segmentation dataset from TCIA, designed to support body composition analysis. Initial annotations were generated using in-house segmentation models and manually refined on every fifth axial slice, with remaining slices labeled as “ignore” (value 255).

It includes 900 CT series from 882 patients, sampled from multiple TCIA collections such as ACRIN-FLT-Breast, ACRIN-HNSCC-FDG-PET/CT, ACRIN-NSCLC-FDG-PET, C4KC-KiTS, LIDC-IDRI, Pancreas-CT, QIN-HEADNECK, and various TCGA cohorts.

Number of CT volumes: 900

Contrast: -

CT body coverage: Various

Does the dataset include any ground truth annotations?: Yes

Original GT annotation targets: Subcutaneous tissue, muscle, abdominal cavity, thoracic cavity, bones, gland structure, pericardium, prosthetic breast implant, mediastinum

Number of annotated CT volumes: 900

Annotator: In-house segmentation models + human refinement

Acquisition centers: -

Pathology/Disease: -

Original dataset download link: https://www.cancerimagingarchive.net/analysis-result/saros/

A script to download and resample the images in GitHub repository: https://github.com/UMEssen/saros-dataset

Original dataset format: DICOM