BraTS-GLI-Anatomy-Lesion is a labels-only derivative resource for WMH-aware anatomy-lesion segmentation on BraTS 2023-GLI cases. It provides 1,251 released unified segmentation labels and 116 image-repair labels derived from the SVID/WMH label resources associated with: Multi-Disease Segmentation of Gliomas and White Matter Hyperintensities in the BraTS Data Using a 3D Convolutional Neural Network.

The MRI images are not redistributed in this release. Users must obtain the corresponding BraTS 2023-GLI MRI data directly from the upstream Synapse project syn51156910: https://www.synapse.org/Synapse:syn51156910/wiki/627000.

Resource Scope

The released labels cover 1,251 BraTS 2023-GLI cases divided into a purified subset (394 cases) and an extended subset (857 cases). The main label for each case uses the 8-class anatomy-lesion schema: background, cortical gray matter, basal ganglia, white matter, lesion, ventricle, cerebellum and brainstem.

Each main label is stored under the BraTS-compatible case identifier so that users can align it with MRI files obtained from upstream BraTS 2023-GLI. For the 116 cases marked image_repair_required=repair_required, the repair_labels/modified/ directory provides 2-class repair labels whose integer values are 1 whole tumor and 2 WMH. These repair labels are not released segmentation targets; users must apply them to repair the corresponding upstream BraTS 2023-GLI MRI images themselves before reproducing the image inputs used for this release. The repair code is available in the public code repository at https://github.com/xyx200/brats-gli-anatomy-lesion-code. The original SVID/WMH source material is MIT-licensed by John Colby through johncolby/svid_paper; the MIT notice is preserved in LICENSE.txt.

Folder Structure

BraTS-GLI-Anatomy-Lesion-v1.0.0/
|-- LICENSE.txt
|-- participants.csv
|-- labels_description.json
|-- scripts/
|   `-- download.sh
|-- metadata/
|   |-- cases.csv
|   |-- data_dictionary.md
|   |-- id_mapping.csv
|   `-- release_manifest.csv
|-- data/
|   |-- purified/<case_id>/<case_id>-seg.nii.gz
|   `-- extended/<case_id>/<case_id>-seg.nii.gz
`-- repair_labels/
    `-- modified/<case_id>/<case_id>-repair-seg.nii.gz

Metadata And Provenance

participants.csv: compact released case/session table with subset and whether source-image repair is required.
labels_description.json: task description, label patterns and integer label mapping.
metadata/cases.csv: case-level label paths, upstream MRI reference, provenance, QC status, license and access metadata.
metadata/id_mapping.csv: mapping from original BraTS-compatible case identifiers to the four-digit internal IDs used during TumorSynth probability-map generation and entropy-weighted fusion.
metadata/release_manifest.csv: label-only file manifest with byte size and SHA-256 digest.
metadata/data_dictionary.md: field definitions and controlled provenance values.

WMH provenance is recorded per case in metadata/cases.csv: 52 purified cases are expert_negative, 342 purified cases are model_negative, and 857 extended cases are deepwmh_lst_intersection. Expert-negative cases do not have repair labels and are marked image_repair_required=no_repair_required. Tumor labels originate from BraTS 2023-GLI. Healthy tissue labels were generated through TumorSynth entropy-weighted fusion and mapped to the released 8-class anatomy-lesion schema.

For reproducing the healthy-tissue fusion curation, metadata/id_mapping.csv maps the four-digit IDs used in TumorSynth-derived modality probability maps back to the original BraTS-compatible case identifiers.

The image-repair labels for modified cases are derived from the SVID/WMH label resources associated with: Multi-Disease Segmentation of Gliomas and White Matter Hyperintensities in the BraTS Data Using a 3D Convolutional Neural Network, and the source repository johncolby/svid_paper.

Repository, Access, And Contact

Hugging Face gated dataset: https://huggingface.co/datasets/Yuxan222/BraTS-GLI-Anatomy-Lesion
Public landing page and Wiki: https://www.synapse.org/Synapse:syn75210889/wiki/640990
Synapse project ID: syn75210889
Controlled label release: syn75260521
Public code repository: https://github.com/xyx200/brats-gli-anatomy-lesion-code
Upstream BraTS 2023 project: syn51156910
Upstream BraTS access page: https://www.synapse.org/Synapse:syn51156910/wiki/627000
Official contact: xxy200200@stu.xjtu.edu.cn

Access Instructions

This resource is distributed through two access-controlled channels: the gated Hugging Face dataset and the controlled Synapse release. Both channels provide the same BraTS-GLI-Anatomy-Lesion v1.0.0 release, with identical labels, metadata, directory structure, release manifest and file checksums. There is no difference in dataset content between the two platforms. Users may request access through either platform and download the release from their preferred channel.

The Synapse project landing page and Wiki are publicly viewable. The versioned label release (syn75260521) remains access-controlled and requires a registered Synapse account. The Hugging Face repository is configured as a manually gated dataset and requires a registered Hugging Face account.

Before requesting access, users must:

Obtain access to the upstream BraTS 2023 data through Synapse ID syn51156910.
Accept the applicable BraTS 2023 post-Challenge Terms and Conditions.
Complete the corresponding email request below and attach evidence of approved upstream BraTS 2023-GLI access.
By submitting the request, accept the BraTS-GLI-Anatomy-Lesion resource terms stated on this public page.

Hugging Face Manual Access Request

Sign in to Hugging Face and submit a request on the gated dataset page: https://huggingface.co/datasets/Yuxan222/BraTS-GLI-Anatomy-Lesion.
Email the completed request below to xxy200200@stu.xjtu.edu.cn.
The maintainer manually verifies the request and, when approved, accepts the matching pending Hugging Face request.

Subject: Hugging Face access request for BraTS-GLI-Anatomy-Lesion v1.0.0

Dear BraTS-GLI-Anatomy-Lesion maintainers,

I request access to the BraTS-GLI-Anatomy-Lesion v1.0.0 gated dataset on Hugging Face.

- Hugging Face username:
- Synapse username:
- Email:

I confirm that I have already obtained access to the upstream BraTS 2023-GLI data through Synapse ID `syn51156910`.

Please attach one screenshot showing your approved access to upstream BraTS 2023-GLI (`syn51156910`). The screenshot should show your Synapse username/account and the approved or access-granted status. You may redact unrelated personal information.

By requesting access, I confirm that:

- I will use this resource only for non-commercial research.
- I have obtained access to the upstream BraTS 2023-GLI data through Synapse ID `syn51156910`.
- I will comply with the applicable upstream BraTS/Synapse terms.
- I will not share downloaded label files with anyone who has not been approved for access to this resource.
- I will preserve the required BraTS attribution statement.
- I will preserve the John Colby MIT notice for SVID/WMH repair-label source material.
- I will follow the current upstream BraTS citation requirements.
- I will cite the BraTS-GLI-Anatomy-Lesion resource article. Until a final peer-reviewed bibliographic citation is available, I will cite the current preprint: Xiang, X., Hao, S., Wang, F., Ma, J., and Lian, C. (2026). *GLI-AL: A Multi-Modal Glioma MRI Label Resource with Unified Anatomy-Lesion Labels*. arXiv:2607.22135. https://doi.org/10.48550/arXiv.2607.22135.
- I will retain the release version and provenance notices.
- I will not attempt re-identification.
- I will not use this resource for direct clinical diagnosis or treatment decisions.
- If I use cases marked `image_repair_required=repair_required`, I will use the provided repair labels and associated repair code to repair the corresponding upstream BraTS 2023-GLI MRI images before attempting to reproduce this release's image inputs.
- I will comply with applicable research ethics standards and upstream BraTS/Synapse data-use terms.

After both the Hugging Face request and the email have been submitted, the maintainer reviews them manually. The Hugging Face username in the email must match the pending request on the dataset page.

After access is granted, download the complete repository with huggingface_hub:

from huggingface_hub import snapshot_download

path = snapshot_download(
    repo_id="Yuxan222/BraTS-GLI-Anatomy-Lesion",
    repo_type="dataset",
    token=True,
)

Synapse Manual Access Request

To request access, email the following completed request to xxy200200@stu.xjtu.edu.cn. The maintainer manually verifies the requester's approved access to upstream BraTS 2023-GLI. After approval, the maintainer grants the requester's Synapse account read and download access to syn75260521.

Subject: Access request for BraTS-GLI-Anatomy-Lesion v1.0.0

Dear BraTS-GLI-Anatomy-Lesion maintainers,

I request access to the BraTS-GLI-Anatomy-Lesion v1.0.0 controlled label release (`syn75260521`) in Synapse project `syn75210889`.

- Synapse username:
- Email:

I confirm that I have already obtained access to the upstream BraTS 2023-GLI data through Synapse ID `syn51156910`.

Please attach one screenshot showing your approved access to upstream BraTS 2023-GLI (`syn51156910`). The screenshot should show your Synapse username/account and the approved or access-granted status. You may redact unrelated personal information.

By requesting access, I confirm that:

- I will use this resource only for non-commercial research.
- I have obtained access to the upstream BraTS 2023-GLI data through Synapse ID `syn51156910`.
- I will comply with the applicable upstream BraTS/Synapse terms.
- I will not share downloaded label files with anyone who has not been approved for access to this resource.
- I will preserve the required BraTS attribution statement.
- I will preserve the John Colby MIT notice for SVID/WMH repair-label source material.
- I will follow the current upstream BraTS citation requirements.
- I will cite the BraTS-GLI-Anatomy-Lesion resource article. Until a final peer-reviewed bibliographic citation is available, I will cite the current preprint: Xiang, X., Hao, S., Wang, F., Ma, J., and Lian, C. (2026). *GLI-AL: A Multi-Modal Glioma MRI Label Resource with Unified Anatomy-Lesion Labels*. arXiv:2607.22135. https://doi.org/10.48550/arXiv.2607.22135.
- I will retain the release version and provenance notices.
- I will not attempt re-identification.
- I will not use this resource for direct clinical diagnosis or treatment decisions.
- If I use cases marked `image_repair_required=repair_required`, I will use the provided repair labels and associated repair code to repair the corresponding upstream BraTS 2023-GLI MRI images before attempting to reproduce this release's image inputs.
- I will comply with applicable research ethics standards and upstream BraTS/Synapse data-use terms.

After access is granted, authenticate with the Synapse command-line client and download the versioned release:

synapse get -r syn75260521 --downloadLocation <output-directory>

Alternatively, if scripts/download.sh is already available locally, run:

bash scripts/download.sh syn75260521 <output-directory>

MRI images are not included and must be downloaded separately from the upstream BraTS 2023-GLI project syn51156910. For cases marked image_repair_required=repair_required, users must repair the corresponding upstream MRI images themselves using the provided repair labels and public repair code before attempting to reproduce this release's image inputs.

Terms Of Use

The derivative release package prepared by us is released under Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0): https://creativecommons.org/licenses/by-nc/4.0/. This license does not remove or replace the original MIT notice for johncolby/svid_paper, and it does not override applicable upstream BraTS data-use terms, including BraTS 2023 access terms for MRI data and BraTS 2018/SVID source-material terms where applicable.

Users must:

Use the resource for non-commercial research only.
Obtain upstream BraTS 2023-GLI MRI access through syn51156910.
Not share downloaded label files with anyone who has not been approved for access to this resource.
Preserve the required BraTS attribution statement.
Preserve the John Colby MIT notice for SVID/WMH repair-label source material.
Follow the current upstream BraTS citation requirements.
Cite the BraTS-GLI-Anatomy-Lesion resource article. Until a final peer-reviewed bibliographic citation is available, cite the current preprint: Xiang, X., Hao, S., Wang, F., Ma, J., and Lian, C. (2026). GLI-AL: A Multi-Modal Glioma MRI Label Resource with Unified Anatomy-Lesion Labels. arXiv:2607.22135. https://doi.org/10.48550/arXiv.2607.22135.
Retain the release version and provenance notices.
Comply with the applicable upstream BraTS/Synapse terms.

Responsible Use Notice

Users must not attempt re-identification. This derivative label resource is intended for research use and does not support direct clinical diagnosis or treatment decisions.

How To Cite This Resource

Publications using this resource should cite the BraTS-GLI-Anatomy-Lesion resource article. Until a final peer-reviewed bibliographic citation is available, the current citation is: Xiang, X., Hao, S., Wang, F., Ma, J., and Lian, C. (2026). GLI-AL: A Multi-Modal Glioma MRI Label Resource with Unified Anatomy-Lesion Labels. arXiv:2607.22135. https://doi.org/10.48550/arXiv.2607.22135. The associated data resource may additionally be cited alongside the resource article using its DOI: https://doi.org/10.7303/SYN75210889. Publications must continue to follow the current upstream BraTS citation requirements.

Work that uses the image-repair labels should also acknowledge: Multi-Disease Segmentation of Gliomas and White Matter Hyperintensities in the BraTS Data Using a 3D Convolutional Neural Network, and the source repository johncolby/svid_paper.

Release Integrity And QC Summary

Full labels-only release QC has been completed on the metadata and label manifest.

Cases listed: 1,251
Main unified segmentation labels: 1,251
Image-repair labels: 116
Released NIfTI label files: 1,367
MRI image files redistributed here: 0
participants.csv, metadata/cases.csv and metadata/release_manifest.csv are aligned at case and path level.

Required BraTS Attribution

Data used in this publication were obtained as part of the Brain Tumor Segmentation (BraTS) Challenge project through Synapse ID: syn51156910.

See LICENSE.txt for the license notice, upstream data-use conditions, attribution statement and the link to the official BraTS page for current upstream citation requirements.

Downloads last month: 13

Total file size:

313 MB

Paper for Yuxan222/BraTS-GLI-Anatomy-Lesion

GLI-AL: A Multi-Modal Glioma MRI Label Resource with Unified Anatomy-Lesion Labels

Paper • 2607.22135 • Published 10 days ago • 5