README.md

metadata

license: cc-by-4.0
task_categories:
  - image-segmentation
tags:
  - medical
  - ct
  - lung
  - nodule
  - luna16
pretty_name: LUNA16 — LUng Nodule Analysis 2016
size_categories:
  - 100<n<1K
configs:
  - config_name: default
    data_files:
      - split: preview
        path: data/preview-*
dataset_info:
  features:
    - name: patient_id
      dtype: string
    - name: subset
      dtype: string
    - name: num_slices
      dtype: int32
    - name: has_nodule_in_mid_slice
      dtype: bool
    - name: image
      dtype: image
    - name: lung_mask
      dtype: image
    - name: nodule_mask
      dtype: image
    - name: overlay
      dtype: image
  splits:
    - name: preview
      num_bytes: 317991662
      num_examples: 888
  download_size: 316899869
  dataset_size: 317991662

LUNA16

A mirror of the LUNA16 (LUng Nodule Analysis 2016) challenge data — 888 thoracic LDCT scans derived from LIDC-IDRI — repackaged for use in the EasyMedSeg medical segmentation framework.

Contents

• subset0/ … subset9/ — 888 CT scans in MetaImage format (.mhd + .raw)
• seg-lungs-LUNA16/ — official lung-field masks (.mhd + .zraw)
• nodule-masks-spheres/ — derived nodule masks built from annotations.csv by drawing a sphere of diameter_mm at each nodule centroid.
• annotations.csv — 1,186 nodule annotations (centroid + diameter)
• candidates.csv, candidates_V2.csv — false-positive-reduction candidates
• sampleSubmission.csv — challenge submission template

Mask sources

LUNA16 ships only centroid+diameter annotations for nodules — there are no official voxel-level nodule masks. Two derived mask sources are provided here:

1. Lung-field masks (seg-lungs-LUNA16/) — official, paired 1:1 with all 888 scans. Per the original release: "provided to aid nodule detection; NOT intended as the reference standard for any segmentation study."
2. Nodule sphere masks (nodule-masks-spheres/) — generated by the uploader from annotations.csv: for each scan, a binary uint8 volume with a sphere of diameter_mm placed at each nodule centroid. Use as an approximate target for nodule segmentation; not a true voxel-level GT.

Source

• Zenodo Part 1: https://zenodo.org/records/3723295 (subsets 0–6, masks, annotations)
• Zenodo Part 2: https://zenodo.org/records/4121926 (subsets 7–9)
• Challenge homepage: https://luna16.grand-challenge.org/

License

CC BY 4.0 — same as the original Zenodo release.

Citation

Setio, A.A.A., Traverso, A., de Bel, T., Berens, M.S.N., et al. (2017). Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images: The LUNA16 challenge. Medical Image Analysis 42: 1–13. doi:10.1016/j.media.2017.06.015