Datasets:

YongchengYAO
/

VerSe-Lite

📝 Landmark annotations, visualization figures and the benchmark plan files live in 🔥MedVision🔥, where you can load the complete images and annotations from dataset configs.

Relation to the source dataset


In the source	374 CT series across VerSe'19 + VerSe'20
Excluded here	30 `sub-gl*` scans (CC BY-NC-ND) and 19 duplicate `_split-verse<NNN>` series
In this repo	325 `Images` + 325 `Masks`

The 30 sub-gl* scans are excluded: that imaging is released under CC BY-NC-ND, which forbids derivative works, so no derived annotation can be redistributed for them. A further 19 series are dropped by de-leaking — ~18 subjects were scanned as 2-3 separate _split-verse<NNN> series, and keeping them all would place the same spine in both the train and test split. 374 scans upstream -> 344 redistributable -> 325 shipped.

Why -Lite? The suffix marks this as a derived redistribution rather than a copy of the source. These are preprocessed volumes — every case has been format-converted, geometry-normalised and reoriented to RAS+ — and for some sources cases or modalities are excluded as well (see the table above). Use it to reproduce MedVision, not as a substitute for the original release. See Preprocessing below for exactly what was changed.

Preprocessing

Images and masks standardized to RAS+ orientation; masks cast to uint16 on the image grid.
Geometry is read via nibabel's .affine, which resolves from QFORM (VerSe files carry sform_code=0, qform_code=1, so reading SFORM directly would give a zero matrix).
Vertebral centroids from the challenge *_ctd.json files are voxel indices in the native orientation (not world-mm); they are mapped through the native->world->RAS+ chain to 0-based indices in the shipped volume.
macOS resource forks (__MACOSX, ._*) in the source archives are filtered out.

📝 Field of view varies from cervical-only to whole-body, so only 250 of the 325 scans contain all of L1-L5. Those form the Images-lumbar/ subset used by the biometry task.

Segmentation Labels

labels_map = {
    "1": "vertebra C1",
    "2": "vertebra C2",
    "3": "vertebra C3",
    "4": "vertebra C4",
    "5": "vertebra C5",
    "6": "vertebra C6",
    "7": "vertebra C7",
    "8": "vertebra T1",
    "9": "vertebra T2",
    "10": "vertebra T3",
    "11": "vertebra T4",
    "12": "vertebra T5",
    "13": "vertebra T6",
    "14": "vertebra T7",
    "15": "vertebra T8",
    "16": "vertebra T9",
    "17": "vertebra T10",
    "18": "vertebra T11",
    "19": "vertebra T12",
    "20": "vertebra L1",
    "21": "vertebra L2",
    "22": "vertebra L3",
    "23": "vertebra L4",
    "24": "vertebra L5",
    "25": "vertebra L6",
    "28": "vertebra T13"
}

Landmarks

landmarks_map = {
    "P1": "centroid of vertebra L1",
    "P2": "centroid of vertebra L2",
    "P3": "centroid of vertebra L3",
    "P4": "centroid of vertebra L4",
    "P5": "centroid of vertebra L5"
}

News

[25 Jul, 2026] Initial release. This dataset is integrated into 🔥MedVision🔥, where you can use these config names to load data in python:
- VerSe_BiometricsFromLandmarks_Task01_Sagittal_Test
- VerSe_BiometricsFromLandmarks_Task01_Sagittal_Train
- VerSe_BoxSize_Task01_Axial_Test
- VerSe_BoxSize_Task01_Axial_Train
- VerSe_BoxSize_Task01_Coronal_Test
- VerSe_BoxSize_Task01_Coronal_Train
- VerSe_BoxSize_Task01_Sagittal_Test
- VerSe_BoxSize_Task01_Sagittal_Train
- VerSe_MaskSize_Task01_Axial_Test
- VerSe_MaskSize_Task01_Axial_Train
- VerSe_MaskSize_Task01_Coronal_Test
- VerSe_MaskSize_Task01_Coronal_Train
- VerSe_MaskSize_Task01_Sagittal_Test
- VerSe_MaskSize_Task01_Sagittal_Train

Data Usage Agreement

By using the dataset, you agree to the terms as follow.

You must comply with the original CC BY-SA 4.0 license terms of the source dataset.
You are recommended to refer to the source of this dataset in any publication: https://huggingface.co/datasets/YongchengYAO/VerSe-Lite
You must cite the original publication(s):
- https://doi.org/10.1016/j.media.2021.102166

Official Release

For more information, please go to the official site: https://github.com/anjany/verse

Download from Huggingface

# python
from huggingface_hub import snapshot_download
snapshot_download(repo_id="YongchengYAO/VerSe-Lite", repo_type='dataset', local_dir="/your/local/folder")

Downloads last month: 17