BraTS 2021
RSNA-ASNR-MICCAI Brain Tumor Segmentation Challenge 2021
Pre-operative multi-parametric brain MRI (T1, contrast-enhanced T1, T2, FLAIR) of 2,040 adult glioma patients from many institutions, with expert-approved tumor sub-region segmentations and MGMT promoter methylation labels.
Overview
BraTS 2021 was the tenth edition of the Brain Tumor Segmentation challenge and the first organized jointly by RSNA, ASNR and MICCAI. It grew the benchmark from 660 cases in 2020 to 2,040 adult glioma patients and added a second task: predicting MGMT promoter methylation from imaging. Its training and validation data were later reused unchanged in BraTS 2022 and as the BraTS 2023 Adult Glioma set.
Composition
Every case has four structural MRI volumes acquired before surgery: native T1, gadolinium-enhanced T1, T2 and T2-FLAIR. The challenge split the 2,040 patients into 1,251 training, 219 validation and 570 test cases. Labels were released for training cases only; the test set stays hidden. Inclusion required a pathologically confirmed diagnosis and a known MGMT status. TCIA hosts 1,480 of the subjects (training and validation of both tasks) together with a crosswalk to the source TCIA collections.
Acquisition
Scans are routine clinical exams from many institutions in North America, Europe and India, with different scanners and protocols, so image quality varies widely. Part of the cohort comes from TCIA collections (TCGA-GBM, TCGA-LGG, IvyGAP, CPTAC-GBM, ACRIN-FMISO-Brain); the rest from institutional contributions. For the segmentation task all volumes were converted to NIfTI, registered to the SRI24 template, resampled to 1 mm isotropic and skull-stripped. For the classification task, skull-stripped volumes were mapped back to patient space and saved as DICOM.
Annotations
Initial segmentations came from a STAPLE fusion of earlier top BraTS methods (nnU-Net, DeepScan, DeepMedic). Volunteer neuroradiologists refined them in ITK-SNAP, and senior board-certified neuroradiologists approved them, returning cases for correction where needed. Labels mark necrotic core (1), edema/infiltrated tissue (2) and enhancing tumor (4). MGMT status is a binary label from laboratory assays of the resected tissue.
Known limitations
- Each case was refined by a single annotator, so inter-rater agreement cannot be measured.
- MGMT status was measured with different assays and thresholds at each site and is only given as binary.
- Other abnormalities, such as white matter hyperintensities, were not labelled.
- Age, sex and scanner details are not published per case.
- The same images carry CC BY 4.0 on TCIA but CC BY-NC terms on Synapse; check which copy you use.
Cohort
Aggregate numbers from the sources below. Bars are relative to the 2,040 subjects.
Contrast combinations
How many subjects have exactly each set of contrasts.
| T1w | T1w_ce | T2w | FLAIR | Subjects with exactly this set |
|---|---|---|---|---|
2,040 |
Contrast / sequence
subjects, values can overlap
- T1-weighted 2,040 100%
- T1-weighted, contrast enhanced 2,040 100%
- T2-weighted 2,040 100%
- FLAIR 2,040 100%
Condition
subjects, values can overlap
- Glioma 2,040 100%
Split
subjects
- Training 1,251 61%
- Test 570 28%
- Validation 219 11%
License and access
Our reading of the license, not legal advice. Before you use the data, read the original license and confirm that your use is allowed. We take no responsibility for how you use a dataset. Full disclaimer
Download without an account
The same data is offered under 2 alternative licenses. Pick the copy whose terms fit; the summary on the right shows the friendliest answer per rule.
Creative Commons Attribution 4.0 International
Use, share and adapt the data for any purpose, including commercial use, as long as you credit the creators.
What you can do
- Yes
- Yes
- Yes
- Yes
What you can share
- Yes
- Yes
- Yes
What you must do
- Yes
- Share alike No
- No
- No
- Manuscript review No
- Release code No
- Return results No
- Delete after use No
Limits
- No
- Location limits No
Creative Commons Attribution-NonCommercial 4.0 International
Use, share and adapt the data with credit, but only for non-commercial purposes.
The Synapse terms state "The Data is subject to a CC-BY-NC license" and require an attribution statement naming Synapse ID syn51156910. The same images are thus offered under CC BY 4.0 on TCIA and CC BY-NC on Synapse. The original DICOM series in the source TCIA collections fall under the NIH Controlled Data Access Policy.
What you can do
- No
- Conditional
- Yes
- Yes
What you can share
- Conditional
- Conditional
- Conditional
What you must do
- Yes
- Share alike No
- No
- No
- Manuscript review No
- Release code No
- Return results No
- Delete after use No
Limits
- No
- Location limits No
Citation
Baid U, et al. The RSNA-ASNR-MICCAI BraTS 2021 Benchmark on Brain Tumor Segmentation and Radiogenomic Classification. arXiv:2107.02314 (2021). Also cite Menze et al., IEEE TMI 34(10) 2015 and Bakas et al., Sci Data 4, 170117 (2017), and the TCIA dataset DOI 10.7937/jc8x-9874.
All numbers
Every number on this page, as stored in stats.csv, with its source.
| Measure | Breakdown | Value | Source |
|---|---|---|---|
| Subjects | total | 2,040 | baid2021 Table 1 |
| Subjects | split=train | 1,251 | baid2021 Table 1 |
| Subjects | split=validation labels not released to participants | 219 | baid2021 Table 1 |
| Subjects | split=test held out | 570 | baid2021 Table 1 |
| Subjects | condition=glioma pre-operative baseline mpMRI of 2040 glioma patients | 2,040 | baid2021 Abstract (p. 3) |
| Subjects | contrast=T1w four structural scans (T1; T1Gd; T2; T2-FLAIR) per case | 2,040 | baid2021 Section 2.1.1 (p. 5) |
| Subjects | contrast=T1w_ce | 2,040 | baid2021 Section 2.1.1 (p. 5) |
| Subjects | contrast=T2w | 2,040 | baid2021 Section 2.1.1 (p. 5) |
| Subjects | contrast=FLAIR | 2,040 | baid2021 Section 2.1.1 (p. 5) |
| Subjects | contrast_set=FLAIR+T1w+T1w_ce+T2w "2,000 glioma cases = 8,000 mpMRI scans", four per case | 2,040 | baid2021 Section 2.1.3 (p. 9) |
| Scans | total "2,000 glioma cases = 8,000 mpMRI scans" | ~8,000 | baid2021 Section 2.1.3 (p. 9) |
Sources
The keys used in the table above.