Overview

Schema Version

https://atlas.rsna.org/schemas/2025-11/dataset.json

Name

A Vertebral Segmentation Dataset with Fracture Grading

Link

https://osf.io/nqjyw/

Indexing

Keywords: Vertebral Segmentation, Spine CT, Vertebral Fractures, Bone Mineral Density, Deep Learning, Computer-Aided Diagnosis, Thoracolumbar Spine, Cervical Spine, NIfTI Format

Content: CT, MK, BQ

Author(s)

Maximilian T. Löffler

Anjany Sekuboyina

Alina Jacob

Anna-Lena Grau

Andreas Scharr

Malek El Husseini

Mareike Kallweit

Claus Zimmer

Thomas Baum

Jan S. Kirschke

Organization(s)

Department of Diagnostic and Interventional Neuroradiology, School of Medicine, Klinikum rechts der Isar, Technical University of Munich

Department of Informatics, Technical University of Munich

License

Text: CC BY-SA 4.0

URL: https://creativecommons.org/licenses/by-sa/4.0/

Contact

m_loeffler@web.de

Funding

Supported by the European Research Council (ERC) with starting grant no. 637164 “iBack” to J.S.K. and by the Deutsche Forschungforschungsgemeinschaft (DFG, German Research Foundation) with project no. 432290010 to J.S.K. and T.B.

Ethical review

The local institutional review board approved this retrospective evaluation of imaging data and waived written informed consent (proposal 27/19 S-SR).

Comments

This dataset provides vertebral segmentation masks for spine CT images and annotations of vertebral fractures or abnormalities per vertebral level. It was used for the VerSe 2019 challenge held during the 22nd International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI). It is intended for large-scale machine learning aimed at automated spine processing and fracture detection.

Dataset

Motivation

The dataset's purpose is to facilitate large-scale machine learning for automated spine processing and fracture detection, addressing the underdiagnosis of vertebral fractures on CT scans and the increasing workload for radiologists.

Sampling

160 CT image series from 141 patients were randomly selected. Inclusion criteria for patients were age older than 30 years and no history of bone metastases. Imaging requirements included 120-kVp acquisition with sagittal reformations reconstructed by filtered back projection (bone kernel) with a spatial resolution of at least 1 mm in the craniocaudal direction.

Partitioning scheme

The dataset is separated into training, public validation, and secret test data subsets for the VerSe 2019 challenge.

Missing information

The dataset only included patients older than 30 years, potentially limiting reliability for younger individuals. It does not cover many normal variants and vertebral abnormalities (e.g., bone metastasis and primary bone tumors were excluded). A rigorous evaluation and inclusion of all possible postoperative changes (including vertebral replacements) is missing. The study focused on edge-enhancing reconstructions, excluding soft-tissue kernels and iterative reconstruction algorithms. Isotropic resolution was not available in all scans, and spatial resolution was limited to 1 mm in each direction.

Relationships between instances

CT image series of one patient are contained within one dataset partition.

Noise

On low-quality scans with a lot of background noise, differentiating between fused vertebrae or low-density degenerative calcification and adjacent soft tissue can be difficult.

Confidentiality

CT data were anonymized by conversion to Neuroimaging Informatics Technology Initiative (NIfTI) format.

Re-identification

CT data were anonymized by conversion to Neuroimaging Informatics Technology Initiative (NIfTI) format.