HaN-Seg: The head and neck organ-at-risk CT & MR segmentation dataset

License

CC BY-NC-ND 4.0
Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License

We have obtained special permission from the original dataset authors to release our derivative segmentations on the images.

Citation

Paper BibTeX:

@article{podobnik2023han,
  title={HaN-Seg: The head and neck organ-at-risk CT and MR segmentation dataset},
  author={Podobnik, Ga{\v{s}}per and Strojan, Primo{\v{z}} and Peterlin, Primo{\v{z}} and Ibragimov, Bulat and Vrtovec, Toma{\v{z}}},
  journal={Medical physics},
  volume={50},
  number={3},
  pages={1917--1927},
  year={2023},
  publisher={Wiley Online Library}
}

Dataset description

HaN-Seg provides anonymized CT and T1-weighted MR scans of 42 head and neck cancer patients acquired for image-guided radiotherapy planning. Each CT scan includes expert-curated binary masks for 30 organs-at-risk (OARs), enabling research on multimodal image analysis and precise radiotherapy planning.

Challenge homepage: https://han-seg2023.grand-challenge.org/

Number of CT volumes: 42

Contrast: Contrast-enhanced

CT body coverage: Head and neck

Does the dataset include any ground truth annotations?: Yes

Original GT annotation targets: (30 OARs) arytenoids, brainstem, carotid artery, cervical esophagus, cochlea, cricopharyngeal inlet, lacrimal gland, larynx—glottis, larynx—supraglottic, lips, mandible, optic chiasm, optic nerve, oral cavity, parotid gland, pituitary gland, spinal cord, submandibular gland, and thyroid

Number of annotated CT volumes:42

Annotator: Human

Acquisition centers: Institute of Oncology Ljubljana, Slovenia

Pathology/Disease: All cases represent pathological conditions, predominantly head and neck tumors. Other pathologies may also be present, as the dataset was not systematically screened for them.

Original dataset download link: https://zenodo.org/records/7442914#.ZBtfBHbMJaQ

Original dataset format: nrrd

Note

We did not include the original GT annotations for spinal cord from HaN-Seg. Since other datasets in CADS cover a larger field of view where the spinal cord extends below the head-and-neck, we generated our own spinal cord segmentations on Han-Seg images to ensure consistency across CADS.