{"success":true,"data":{"author":"Anjany Sekuboyina et al., TU Munich, VerSe challenge (redistributed by Yongcheng Yao)","bucketAccessType":"public","bucketName":"YongchengYAO/VerSe-Lite","bucketPrefix":"","category":"Health & Biomedical","cloudProvider":"huggingface","createdAt":"2026-11-14T14:15:11.813Z","description":"VerSe is the MICCAI 2019 and 2020 Large Scale Vertebrae Segmentation Challenge dataset from TU Munich: multi-detector spine CT scans with per-vertebra voxel masks (labels C1 through L6 and transitional vertebrae) and vertebral centroid annotations. This repository is a derived redistribution by Yongcheng Yao for the MedVision benchmark. It holds 325 CT volumes and 325 masks: of the 374 upstream series, the 30 sub-gl scans released under CC BY-NC-ND were deliberately excluded and 19 duplicate split series were dropped to avoid subject leakage. Volumes are reoriented to RAS+ and masks cast to uint16 on the image grid; lumbar centroid landmarks are provided for the 250 scans that contain L1 to L5. As measured 2026-11-14 the data ship as five zip parts totalling about 48.8 GB; unzip and open the NIfTI (.nii.gz) files with nibabel or SimpleITK. Used for vertebra identification, spine segmentation, landmark detection and osteoporosis or fracture research. The VerSe organisers provide the data under CC BY-SA 4.0 (the excluded sub-gl scans aside); cite Sekuboyina et al., Medical Image Analysis 2021, and share derivatives alike.","displayTitle":"VerSe 2019 + 2020: vertebrae labelling and segmentation in spine CT (CC BY-SA subset)","id":"0feee524-3f22-5b4c-834b-c74c4f7f6dd7","license":"CC-BY-SA-4.0","licenseEvidenceUrl":"https://github.com/anjany/verse","objectCountEstimate":7,"probeStatus":"ok","referenceUrl":"https://doi.org/10.1016/j.media.2021.102166","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"YongchengYAO/VerSe-Lite\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"YongchengYAO/VerSe-Lite\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=YongchengYAO, license:cc-by-sa-4.0, size_categories:n<1K, downloads=77"],"format":"mixed","generatedAt":"2026-11-14T14:15:15.935Z","layout":{"example":"data-part001.zip","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"data-part001.zip","truncatedAt":512}],"sampleObjects":["data-part001.zip","data-part002.zip","data-part003.zip","data-part004.zip","data-part005.zip"],"sidecars":[{"bytes":"---\nlicense: cc-by-sa-4.0\ntask_categories:\n- image-segmentation\nlanguage:\n- en\ntags:\n- medical\n- image\n- ct\n- spine\n- vertebrae\n- segmentation\n- landmarks\npretty_name: 'verse-lite'\nsize_categories:\n- n<1K\n---\n\n\n## About\nThis is a preprocessed redistribution of [VerSe'19 + VerSe'20](https://github.com/anjany/verse), which is released under the `CC BY-SA 4.0` license.\n\n**Dataset summary:** 325 spine CT scans with per-vertebra segmentation masks, and lumbar (L1-L5) centroid landmarks for 250 of them.\n\n**Contents of this repository:**\n\n- `Images/` — 325 files\n- `Masks/` — 325 files\n\n📝 Landmark annotations, visualization figures and the benchmark plan files live in 🔥[MedVision](https://huggingface.co/datasets/YongchengYAO/MedVision)🔥, where you can load the complete images and annotations from dataset configs.\n\n\n## Relation to the source dataset\n\n| | |\n| --- | --- |\n| In the source | 374 CT series across VerSe'19 + VerSe'20 |\n| Excluded here | 30 `sub-gl*` scans (CC BY-NC-ND) and 19 duplicate `_split-verse<NNN>` series |\n| **In this repo** | **325 `Images` + 325 `Masks`** |\n\nThe 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**.\n\n**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](#preprocessing) below for exactly what was changed.\n\n\n## Preprocessing\n\n- Images and masks standardized to RAS+ orientation; masks cast to `uint16` on the image grid.\n\n- 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).\n\n- 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.\n\n- macOS resource forks (`__MACOSX`, `._*`) in the source archives are filtered out.\n\n📝 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.\n\n\n## Segmentation Labels\n\n```python\nlabels_map = {\n    \"1\": \"vertebra C1\",\n    \"2\": \"vertebra C2\",\n    \"3\": \"vertebra C3\",\n    \"4\": \"vertebra C4\",\n    \"5\": \"vertebra C5\",\n    \"6\": \"vertebra C6\",\n    \"7\": \"vertebra C7\",\n    \"8\": \"vertebra T1\",\n    \"9\": \"vertebra T2\",\n    \"10\": \"vertebra T3\",\n    \"11\": \"vertebra T4\",\n    \"12\": \"vertebra T5\",\n    \"13\": \"vertebra T6\",\n    \"14\": \"vertebra T7\",\n    \"15\": \"vertebra T8\",\n    \"16\": \"vertebra T9\",\n    \"17\": \"vertebra T10\",\n    \"18\": \"vertebra T11\",\n    \"19\": \"vertebra T12\",\n    \"20\": \"vertebra L1\",\n    \"21\": \"vertebra L2\",\n    \"22\": \"vertebra L3\",\n    \"23\": \"vertebra L4\",\n    \"24\": \"vertebra L5\",\n    \"25\": \"vertebra L6\",\n    \"28\": \"vertebra T13\"\n}\n```\n\n\n## Landmarks\n\n```python\nlandmarks_map = {\n    \"P1\": \"centroid of vertebra L1\",\n    \"P2\": \"centroid of vertebra L2\",\n    \"P3\": \"centroid of vertebra L3\",\n    \"P4\": \"centroid of vertebra L4\",\n    \"P5\": \"centroid of vertebra L5\"\n}\n```\n\n\n## News\n- [25 Jul, 2026] Initial release. This dataset is integrated into 🔥[MedVision](https://huggingface.co/datasets/YongchengYAO/MedVision)🔥, where you can use these config names to load data in python:\n\n  - `VerSe_BiometricsFromLandmarks_Task01_Sagittal_Test`\n  - `VerSe_BiometricsFromLandmarks_Task01_Sagittal_Train`\n  - `VerSe_BoxSize_Task01_Axial_Test`\n  - `VerSe_BoxSize_T","encoding":"utf8","key":"README.md","name":"README.md","size":5243,"truncated":true}],"sourceObjectKey":"data-part001.zip","starterCell":"from huggingface_hub import hf_hub_download\npath = hf_hub_download(repo_id=\"YongchengYAO/VerSe-Lite\", filename=\"data-part001.zip\", repo_type=\"dataset\")\nprint(path)","starterSamplePath":"data-part001.zip","totalObjects":7},"summary":"VerSe is the MICCAI 2019 and 2020 Large Scale Vertebrae Segmentation Challenge dataset from TU Munich: multi-detector spine CT scans with per-vertebra voxel masks (labels C1…","title":"VerSe 2019 + 2020: vertebrae labelling and segmentation in spine CT (CC BY-SA subset)","updatedAt":"2026-11-14T17:53:26.952Z","visibility":"public","websiteUrl":"https://github.com/anjany/verse"}}