README.md

metadata

license: cc-by-4.0
task_categories:
  - image-segmentation
tags:
  - medical
  - mri
  - lumbar-spine
  - segmentation
  - vertebrae
  - intervertebral-disc
  - spinal-canal
size_categories:
  - n<1K
configs:
  - config_name: default
    data_files:
      - split: training
        path: data/training-*
      - split: validation
        path: data/validation-*
dataset_info:
  features:
    - name: patient_id
      dtype: string
    - name: sequence
      dtype: string
    - name: series
      dtype: string
    - name: subset
      dtype: string
    - name: num_vertebrae
      dtype: int32
    - name: num_discs
      dtype: int32
    - name: sex
      dtype: string
    - name: manufacturer
      dtype: string
    - name: model
      dtype: string
    - name: magnetic_field_strength
      dtype: float32
    - name: num_slices
      dtype: int32
    - name: shape_zyx
      list: int32
    - name: num_label_classes
      dtype: int32
    - name: image_middle_slice
      dtype: image
    - name: mask_middle_slice
      dtype: image
    - name: overlay_middle_slice
      dtype: image
  splits:
    - name: training
      num_bytes: 35627666
      num_examples: 360
    - name: validation
      num_bytes: 10866295
      num_examples: 87
  download_size: 46479041
  dataset_size: 46493961

SPIDER — Lumbar Spine Segmentation in MR Images

Public release of the SPIDER dataset: 218 patients, 447 sagittal lumbar-spine MRI series with reference segmentations of vertebrae, intervertebral discs (IVDs), and spinal canal. Source: van der Graaf et al., Scientific Data 11:264 (2024).

Contents

Path	Description
images/images/	447 .mha MRI volumes — naming `{patient_id}_{t1
masks/masks/	447 .mha reference segmentations, one per image
overview.csv	Per-series metadata: subset (training/validation), sex, vendor, sequence parameters, etc.
radiological_gradings.csv	Per-IVD radiological gradings (Modic, Pfirrmann, herniation, etc.)

Splits

Split assignment is in overview.csv column subset:

• training — 360 series
• validation — 87 series

(The Grand Challenge hidden test set of 39 patients is not included; it is only accessible via https://spider.grand-challenge.org/.)

Label Scheme (masks)

• Vertebrae: sequential integer labels (numbered top-down)
• IVDs: labeled by the vertebra immediately above (lowest IVD = 201, next up = 202, ...)
• Spinal canal: 100
• Background: 0

Recommended Ground Truth

Use masks/masks/. The 41 T2-SPACE series are the most authoritative tier (fully manual annotation on near-isotropic, higher-resolution voxels). T1/T2 masks were produced by iterative semi-automatic refinement under MSK-radiologist supervision.

Source

• Paper: van der Graaf, J. W., et al. (2024). Lumbar spine segmentation in MR images: a dataset and a public benchmark. Sci Data 11, 264. doi:10.1038/s41597-024-03090-w
• Original Zenodo: https://zenodo.org/records/10159290
• Grand Challenge: https://spider.grand-challenge.org/
• License: CC-BY 4.0

This HuggingFace mirror reproduces the Zenodo v4 release (Nov 2023, modified Mar 2024) without modification.