Look well to the spine for the cause of disease - Hippocrates

VerSe examples. Observe the variability in data: field-of-view, fractures, transitional vertebrae, etc.

Spine or vertebral segmentation is a crucial step in all applications regarding automated quantification of spinal morphology and pathology. With the advent of deep learning, for such a task on computed tomography (CT) scans, a big and varied data is a primary sought-after resource. However, a large-scale, public dataset is currently unavailable.

We believe VerSe can help here. VerSe is a large scale, multi-detector, multi-site, CT spine dataset consisting of 374 scans from 300 patients. The challenge was held in two iterations in conjunction with MICCAI 2019 and 2020. The tasks evaluated for include: vertebral labelling and segmentation.

• The dataset has four files corresponding to one data sample: image, segmentation mask, centroid annotations, a PNG overview of the annotations.

• Data structure

  • 01_training - Train data
  • 02_validation - (Formerly) PUBLIC test data
  • 03_test - (Formerly) HIDDEN test data

• Sub-directory-based arrangement for each patient. File names are constructed of entities, a suffix and a file extension following the conventions of the Brain Imaging Data Structure (BIDS; https://bids.neuroimaging.io/)

Example:
-------
training/rawdata/sub-verse000
    sub-verse000_dir-orient_ct.nii.gz - CT image series

training/derivatives/sub-verse000/
    sub-verse000_dir-orient_seg-vert_msk.nii.gz - Segmentation mask of the vertebrae
    sub-verse000_dir-orient_seg-subreg_ctd.json - Centroid coordinates in image space
    sub-verse000_dir-orient_seg-vert_snp.png - Preview reformations of the annotated CT data.

• Centroid coordinates of the subject based structure (.json file) are given in voxels in the image space. 'label' corresponds to the vertebral label:
  • 1-7: cervical spine: C1-C7
  • 8-19: thoracic spine: T1-T12
  • 20-25: lumbar spine: L1-L6
  • 26: sacrum - not labeled in this dataset
  • 27: cocygis - not labeled in this dataset
  • 28: additional 13th thoracic vertebra, T13

1. (VerSe'19) https://s3.bonescreen.de/public/VerSe-complete/dataset-verse19training.zip

2. (VerSe'19) https://s3.bonescreen.de/public/VerSe-complete/dataset-verse19validation.zip

3. (VerSe'19) https://s3.bonescreen.de/public/VerSe-complete/dataset-verse19test.zip

4. (VerSe'20) https://s3.bonescreen.de/public/VerSe-complete/dataset-verse20training.zip

5. (VerSe'20) https://s3.bonescreen.de/public/VerSe-complete/dataset-verse20validation.zip

6. (VerSe'20) https://s3.bonescreen.de/public/VerSe-complete/dataset-verse20test.zip

1. VerSe'19: https://osf.io/nqjyw/
2. VerSe'20: https://osf.io/t98fz/

Note: The annotation format of the complete VerSe data is NOT identical to the one used for the MICCAI challenges. The OSF repositories above also point to the MICCAI version of the data and annotations. Nonetheless, we recommend usage of the restructured data and annotations

If you use VerSe, we would appreciate references to the following papers.

Löffler M et al., A Vertebral Segmentation Dataset with Fracture Grading. Radiology: Artificial Intelligence, 2020. (https://doi.org/10.1148/ryai.2020190138)

Sekuboyina A et al., VerSe: A Vertebrae Labelling and Segmentation Benchmark for Multi-detector CT Images, 2020. (https://arxiv.org/abs/2001.09193)

The data is provided under the CC BY-SA 4.0 License, making it fully open-sourced.

The rest of this repository is under the MIT License.

We provide helper code and guiding notebooks.

• Data reading, standardising, and writing: Data utilities
• Evaluation (as employed in the 2020 challenge): Evaluation utilities
• Notebooks: Data preperation, Evaluation

For queries and issues not fit for a github issue, please email Anjany Sekuboyina or Jan Kirschke .

VerSe has resulted in numerous other publications. Below are a few selected ones.

Sekuboyina A. et al., Labelling Vertebrae with 2D Reformations of Multidetector CT Images: An Adversarial Approach for Incorporating Prior Knowledge of Spine Anatomy. Radiology: Artificial Intelligence, 2020. (https://doi.org/10.1148/ryai.2020190074)

• This work is supported by the European Research Council (ERC) under the European Union’s ‘Horizon2020’ research & innovation programme (GA637164–iBack–ERC–2014–STG).
• We thank NVIDIA for the support in challenge organisation with compute.
• README derived from PCam.