Dataset Card for VertebralBodiesCT-Labels

This dataset contains labels for the thoracic and lumbar vertebral bodies from 1460 CT scans, designed for deep learning applications in anatomical landmark identification.

Dataset Details

VertebralBodiesCT-Labels is a dataset including segmentation labels for the vertebral bodies of the thoracic and lumbar spine, and the sacrum. Derived from 1460 CT scans originally published in the TotalSegmentator and VerSe datasets, labels were reviewed, corrected and adapted to focus exclusively on vertebral bodies, excluding the vertebral arch and spinous processes. These labels are intended for use in deep learning models for anatomical landmark identification, such as in body composition analysis and sarcopenia assessment. The dataset is available under the CC BY-SA 4.0 license and accompanied by model weights and a segmentation pipeline.

Uses

The labels are intended for use in deep learning models aimed at anatomical landmark identification, for instance in body composition analysis and sarcopenia assessment. They are well-suited for segmenting vertebral bodies in the thoracic and lumbar spine. However, because the labels exclude components like the spinous processes and vertebral arches, they are not suitable for whole spine segmentation tasks. Users should be aware that errors may persist, especially in cases involving transitional vertebrae. Models derived from this dataset are not intended for direct clinical application.

Dataset Structure

This dataset does not include imaging studies itself. These are available from the original TotalSegmentator and VerSe datasets (see Sources).

Instead, this dataset comprises 1460 NIfTI (*.nii.gz) files containing the multi-label segmentation of vertebral bodies. Files derived from CT scans in the TotalSegmentator dataset are prefixed with tseg- followed by the corresponding TotalSegmentator ID (e.g., tseg-s1317.nii.gz). Files derived from CT scans in the VerSe dataset are prefixed with verse- followed by the corresponding VerSe ID (e.g., verse-825.nii.gz). The renaming process was automated using bash scripts, which can be accessed here. Some imaging studies were excluded, for instance due to non-orthogonal direction matrices (see Exclusions).

After labeling, the dataset was divided into six parts: one for testing and five for use as folds in a cross-validation training scenario. The split was stratified based on the data source (TotalSegmentator and VerSe). CT scans that were included in the test sets of the original datasets were preferentially assigned to the test set of this dataset. Additionally, the split was stratified according to the labels. For detailed information on the splits, see here.

Dataset Creation

For details regarding on the dataset creation and labeling process, please refer to our paper under submission. This dataset is built upon CT scans and labels from the TotalSegmentator and VerSe datasets. We reviewed existing labels and corrected errors, and adapted the labels to focus exclusively on vertebral bodies, excluding the vertebral arch and spinous processes.

Sources

• TotalSegmentator dataset & model weights
  • Repository: https://github.com/wasserth/TotalSegmentator
  • Download: Zenodo (CT, labels), GitHub (model weights)
  • License: CC BY 4.0
  • Paper: Radiology AI
  • Exlusions: 140, see here
  • Usage: CT scans, spine labels (T1-T12, L1-L5, Sacrum), spinal_cord labels & model weights
• VerSe dataset
  • Repository: https://github.com/anjany/verse
  • Download: OSF (VerSe2019), OSF (VerSe2020)
  • License: CC BY-SA 4.0
  • Paper: Medical Image Analysis, Radiology AI, Scientific Data
  • Exlusions: 2, see here
  • Usage: CT scans, spine labels (T1-T13, L1-L6)
• Body-and-Organ-Analysis model weights
  • Repository: https://github.com/UMEssen/Body-and-Organ-Analysis
  • Download: Zenodo
  • License: MIT License
  • Paper: Investigative Radiology
  • Usage: nervous_system model weights

Labels

label	structure	comment
0	background
1	T1
2	T2
3	T3
4	T4
5	T5
6	T6
7	T7
8	T8
9	T9
10	T10
11	T11
12	T12
13	L1
14	L2
15	L3
16	L4
17	L5
18	L6	optional
19	Os sacrum, S1
20	Os coccygis	not included in the current dataset
21	T13	optional

We adhered to definitions from the VerSe dataset. Details and the labeling process are described in our paper.

• Vertebrae labeled: T1-T13, L1-L6, Sacrum (S1), with optional additional vertebrae (e.g., L6, T13)
• L1: First vertebra without ribs or with rib remnants <4 cm
• Thoracic Vertebrae: Ribs >4 cm; if T1 not visible, 12 thoracic vertebrae are assumed
• Transitional Vertebrae: Only "free” transitional vertebrae (Castellvi grade I or II) considered as lumbar; fused vertebrae (grade III or IV) labeled as sacrum
• Vertebral Bodies Only: only vertebral portions anterior to the spinal cord included (exemption: sacrum), also see here

Bias, Risks, and Limitations

• Data Source Bias: Despite being heterogenous, the dataset may not fully represent diverse populations, imaging techniques, or scanner types.
• Misfunction in Pathologies: Trained models may perform poorly in cases with uncommon or complex spinal pathologies not well-represented in the dataset.
• Persistent Labeling Errors: Despite corrections and manual reviews, some labeling errors may still be present. Found one? Contact us!
• Exclusion of Cervical Spine: The dataset only includes thoracic and lumbar vertebrae.
• Exclusion of Vertebral Arches and Spinous Processes: The labels focus solely on vertebral bodies, excluding arches and spinous processes.
• No Clinical Usage: This dataset is intended for research purposes only.

Citation

We highly appreciate the foundational work of the TotalSegmentator and VerSe authors. This project wouldn’t have been possible without their open-source contributions. Thus, if you use this dataset or any derived model, please cite the following publications:

Wasserthal, J., Breit, H.-C., Meyer, M.T., Pradella, M., Hinck, D., Sauter, A.W., Heye, T., Boll, D., Cyriac, J., Yang, S., Bach, M., Segeroth, M. (2023). TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images. Radiology: Artificial Intelligence. https://doi.org/10.1148/ryai.230024

Sekuboyina A., Husseini M.E., Bayat A., Löffler M., Liebl H., Li H., Tetteh G., Kukačka J., Payer C., Štern D., Urschler M., Chen M., Cheng D., Lessmann N., Hu Y., Wang T., Yang D., Xu D., Ambellan F., Amiranashvili T., Ehlke M., Lamecker H., Lehnert S., Lirio M., Pérez de Olaguer N., Ramm H., Sahu M., Tack A., Zachow S., Jiang T., Ma X., Angerman C., Wang X., Brown K., Kirszenberg A., Puybareau É., Chen D., Bai Y., Rapazzo B. H., Yeah T., Zhang A., Xu S., Hou F., He Z., Zeng C., Xiangshang Z., Liming X., Netherton T.J., Mumme R.P., Court L.E., Huang Z., He C., Wang L.-W., Ling S.H., Huỳnh L.D., Boutry N., Jakubicek R., Chmelik J., Mulay S., Sivaprakasam M., Paetzold J.C., Shit S., Ezhov I., Wiestler B., Glocker B., Valentinitsch A., Rempfler M., Menze B.H., Kirschke J.S. (2021). VerSe: A Vertebrae labelling and segmentation benchmark for multi-detector CT images. Medical Image Analysis. https://doi.org/10.1016/j.media.2021.102166

Haubold, J., Baldini, G., Parmar, V., Schaarschmidt, B.M., Koitka, S., Kroll, L., van Landeghem, N., Umutlu, L., Forsting, M., Nensa, F., & Hosch, R. (2023). BOA: A CT-Based Body and Organ Analysis for Radiologists at the Point of Care. Investigative Radiology. https://doi.org/10.1097/RLI.0000000000001040

Please cite our dataset:

Hofmann F.O. et al. Thoracic & lumbar vertebral body labels corresponding to 1460 public CT scans. https://huggingface.co/datasets/fhofmann/VertebralBodiesCT-Labels/

Contact

Any feedback, questions or recommendations? Feel free to contact us! 🤗