Research map / Label-efficient and robust learning / Recognition and long-tail
OpenBrain: An Auditable Generated-Label Release for Whole-Brain MRI Parcellation
Qizhen Lan, Yu-Chun Hsu, YUXIANG WEI, Lijing Zhu, Zenan Sun, Liang He, Lishan Yu, Xiaoqian Jiang
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces OpenBrain, a confidence-aware, provenance-verified release of 35,838 generated whole-brain MRI parcellations from 607 OpenNeuro T1w datasets, organized as auditable case records.
Problem
Generated whole-brain MRI parcellation releases at cohort scale provide little evidence about which cases are reliable enough for reuse.
Approach
Releases generated labels as an auditable case-record collection with per-case confidence and provenance verification, rather than a directory of standalone segmentation files.
Key results
- Release of 35,838 provenance-verified OpenNeuro T1w cases from 607 source datasets.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| retain only lowest-risk 25% of cases by QC-7 risk | generated whole-brain labels · IXI | macro Dice: +1.3 pp | Improved |
| retain only lowest-risk 25% of cases by QC-7 risk | generated whole-brain labels · BrainLat | macro Dice: +3.7 pp | Improved |
| retain only lowest-risk 25% of cases by QC-7 risk | generated whole-brain labels · Mindboggle-101 (manual labels) | macro Dice: +1.5 pp | Improved |
| select lower-risk instead of random training cases (1,000 budget) | nnU-Net trained on OpenBrain · Mindboggle-101 | macro Dice: 0.710 -> 0.746 | Improved |
| select higher-risk instead of random training cases | nnU-Net trained on OpenBrain · Mindboggle-101 | macro Dice: 0.710 -> 0.456 | Hurt |
| evaluate on small-structure subset only | lower-risk training selection · BrainLat small structures | Def-A Dice: +0.4 pp, CI [-0.5, +1.2] | No clear effect |
| remove any single committee axis from QC-7 (ablation) | QC-7 aggregate · cross-cohort mean, K=25% | macro Dice gain: about 0.1 pp change | No clear effect |
| merge committee axes into one term (ablation) | QC-7 aggregate · cross-cohort mean, K=25% | macro Dice gain: 2.17 -> 2.04 | Hurt |
| add committee axes to r1-r4 (ablation) | QC-7 aggregate · cross-cohort mean, K=25% | macro Dice gain: 1.51 -> 2.17 | Improved |
| use 110-label instead of 96-label evaluation space (ablation) | QC-7 retention · IXI | macro Dice gain: +1.3 -> +1.6 | No clear effect |
Experimental setup
Datasets: OpenNeuro T1w datasets
Closest papers by meaning
- OmniBrainBench: A Comprehensive Multimodal Benchmark for Brain Imaging Analysis Across Multi-stage Clinical Tasks (CVPR 2026 · similarity 0.87)
- NOVA: A Benchmark for Rare Anomaly Localization and Clinical Reasoning in Brain MRI (NeurIPS 2025 · similarity 0.85)
- EMAD: Evidence-Centric Grounded Multimodal Diagnosis for Alzheimer's Disease (CVPR 2026 · similarity 0.84)
- CardioLens: Revealing the Clinical Reality Gap of MLLMs via Multi-Sequence Cardiac MRI Evaluations (NeurIPS 2026 · similarity 0.84)
- Enhancing vision-language models for medical imaging: bridging the 3D gap with innovative slice selection (NeurIPS 2024 · similarity 0.83)
- DDx-TRACE: A Benchmark for Medical Diagnostic Trajectories in VLMs (NeurIPS 2026 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
