README.md

metadata

license: cc0-1.0
task_categories:
  - image-segmentation
tags:
  - medical-imaging
  - brain-mri
  - parcellation
  - quality-control
  - neurips-2026
pretty_name: OpenBrain v1.0
configs:
  - config_name: sample_manifest
    data_files:
      - split: sample
        path: sample/sample_manifest.csv
    default: true

OpenBrain v1.0

OpenBrain: An Auditable Generated-Label Release for Whole-Brain MRI Parcellation
Accepted to NeurIPS 2026, Evaluations & Datasets Track. Paper on OpenReview

Qizhen Lan, Yu-Chun Hsu, Yuxiang Wei, Lijing Zhu, Zenan Sun, Liang He, Lishan Yu, Xiaoqian Jiang.

Maintained by Qizhen Lan (lanqz7766).

OpenBrain v1.0 is a public release of brain-extracted T1-weighted MRI images, SynthStrip-derived brain masks, and automated whole-brain segmentation labels.

The dataset viewer displays the 64-case sample manifest. The full release contains 35,838 cases under cases/.

Release Contents

• Cases: 35,838
• Source OpenNeuro datasets: 607
• License: CC0
• Artifacts per case:
  • image.nii.gz: revised brain-extracted T1w image
  • brain_mask.nii.gz: SynthStrip-derived brain mask
  • whole_brain_segmentation.nii.gz: automated whole-brain segmentation label

All released cases are derived from raw T1w images that were exactly matched to verified public OpenNeuro snapshots with CC0/public-domain-compatible license metadata. The segmentation labels are algorithmic labels, not manual ground truth.

Layout

cases/<case_id>/image.nii.gz
cases/<case_id>/brain_mask.nii.gz
cases/<case_id>/whole_brain_segmentation.nii.gz
metadata/openbrain_v1_0_metadata.csv
metadata/sha256_manifest.csv
sample/cases/<case_id>/...
sample/sample_manifest.csv
sample/README.md

QC Manifest Fields

The QC manifest uses Q1--Q7 for the raw OpenBrain QC-7 values and r1--r7 for the corresponding source-local percentile risk fields. Larger r_k values indicate higher risk relative to cases from the same OpenNeuro source dataset; they are not calibrated probabilities of segmentation correctness. R is the aggregate QC-7 risk score, and tier is the confidence tier derived from the fixed release operating points.

The public sample manifest under sample/sample_manifest.csv follows the same QC field convention for the 64-case sample subset.

Sample

The sample/ directory is a 64-case preview subset. Every sample case is drawn from the HighConf tier and includes the same three case artifacts as the full release: image.nii.gz, brain_mask.nii.gz, and whole_brain_segmentation.nii.gz. The sample is source-diverse (one case per source dataset) and selected by low combined risk R, with Q1/Q3/Q4 visual-cleanliness tie-breaks.

Use it for visual inspection, download testing, and pipeline smoke tests. The labels remain algorithmic pseudo-labels, not manual ground truth.

Exclusions

This release excludes cases with non-CC0 licenses, missing OpenNeuro provenance, manual-review license status, ambiguous raw BIDS matches, or missing raw T1w matches. OpenMind-derived quality scores and masks are not included.

Citation

@inproceedings{lan2026openbrain,
  title = {OpenBrain: An Auditable Generated-Label Release for Whole-Brain MRI Parcellation},
  author = {Lan, Qizhen and Hsu, Yu-Chun and Wei, Yuxiang and Zhu, Lijing and Sun, Zenan and He, Liang and Yu, Lishan and Jiang, Xiaoqian},
  booktitle = {Advances in Neural Information Processing Systems, Evaluations \& Datasets Track},
  year = {2026}
}

Contact

For questions about the release, contact Qizhen Lan at Qizhen.Lan@uth.tmc.edu.