Awesome

VerSe: Large Scale Vertebrae Segmentation Challenge

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

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

<details><summary>Table of Contents</summary>

What is VerSe?
Citing VerSe
Data
Download
License
Code
Contact
Other Related Work
Acknowledgements

</details>

What is VerSe?

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 355 patients. The challenge was held in two iterations in conjunction with MICCAI 2019 and 2020. The tasks evaluated for include: vertebral labelling and segmentation.

Citing VerSe

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

Sekuboyina A et al., VerSe: A Vertebrae Labelling and Segmentation Benchmark for Multi-detector CT Images, 2021. In Medical Image Analysis: https://doi.org/10.1016/j.media.2021.102166 Pre-print: https://arxiv.org/abs/2001.09193
Löffler M et al., A Vertebral Segmentation Dataset with Fracture Grading. Radiology: Artificial Intelligence, 2020. In Radiology AI: https://doi.org/10.1148/ryai.2020190138
Liebl H and Schinz D et al., A Computed Tomography Vertebral Segmentation Dataset with Anatomical Variations and Multi-Vendor Scanner Data, 2021. Pre-print: https://arxiv.org/pdf/2103.06360.pdf

Data

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

Download

WGET:

OSF Reporsitories:

VerSe'19: https://osf.io/nqjyw/
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

License

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.

Code

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

Contact

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

Other Related Work

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)

Acknowledgements

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.