BrainMetShare
Skull-stripped, co-registered 3D brain MRI (pre- and post-contrast T1, post-contrast FLAIR) of 156 Stanford patients with brain metastases, with radiologist lesion masks for 105 cases. Non-commercial research use only.
Overview
BrainMetShare is a brain MRI collection of patients with brain metastases, assembled at Stanford University and distributed by the Stanford AIMI Center. It was built to support research on automatic detection and segmentation of metastatic lesions, and it is the data behind the deep learning study by Grøvik et al. in JMRI.
Composition
The dataset covers 156 consecutive patients with at least one brain metastasis who had not yet received surgery or radiation therapy, imaged between June 2016 and June 2018 at a single center. Mean age was 63 years (SD 12, range 29 to 92). The paper reports 105 women and 51 men. The primary tumor was lung cancer in 99 patients, breast cancer in 33, melanoma in 7, genitourinary cancer in 7, gastrointestinal cancer in 5 and other cancers in 5. 64 patients had 1 to 3 metastases, 47 had 4 to 10 and 45 had more than 10. Lesions ranged from 2 mm to more than 4 cm.
The release splits the cases into 105 with lesion masks (training folder) and 51 without (test folder). The paper used a different split of 100 training, 5 development and 51 test cases.
Acquisition
Each case has four 3D axial sequences: T1-weighted fast spin echo (CUBE) before and after gadolinium, a post-gadolinium IR-prepped FSPGR (BRAVO) and a post-gadolinium CUBE FLAIR. Contrast was given at a standard dose of 0.1 mmol/kg. Scans came from GE 1.5 T (18 patients) and GE or Siemens 3 T (138 patients) systems. The sequences are co-registered, resampled to 256 x 256 pixels in plane (about 0.94 mm, 1.0 mm through plane) and skull-stripped with BET using a mask from the pre-contrast T1 series. A spreadsheet lists the primary cancer of each case.
Annotations
Two neuroradiologists outlined every enhancing metastasis slice by slice on the post-contrast FSPGR images, guided by the FLAIR and post-contrast spin echo images, and cross-checked each other. Masks are binary.
Known limitations
- Single center and retrospective; most patients have lung or breast cancer.
- Only the 105 training cases include masks.
- Images are already resampled and skull-stripped, so original resolution and non-brain tissue are not available.
- The research use agreement allows personal, non-commercial research only and forbids redistribution, derivative works and clinical use.
Cohort
Aggregate numbers from the sources below. Bars are relative to the 156 subjects.
Sex
- Female 105 67%
- Male 51 33%
Age
mean 63± 12 · range 29 to 92No age bins reported.
Contrast combinations
How many subjects have exactly each set of contrasts.
| T1w | T1w_ce | FLAIR | Subjects with exactly this set |
|---|---|---|---|
156 |
Contrast / sequence
subjects, values can overlap
- T1-weighted 156 100%
- T1-weighted, contrast enhanced 156 100%
- FLAIR 156 100%
Condition
subjects, values can overlap
- Brain metastases 156 100%
- Lung cancer 99 63%
- Breast cancer 33 21%
- Melanoma 7 4%
Field strength
subjects
- 3 T 138 88%
- 1.5 T 18 12%
Split
subjects
- Training 105 67%
- Test 51 33%
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
Create an account and accept terms online
Register and accept the Stanford research use agreement. The data are downloaded through Redivis.
Stanford University School of Medicine Dataset Research Use Agreement (AIMI)
Free for personal, non-commercial research only. You may not share the data or the download link, may not create derivative works, and must not try to re-identify patients. Stanford sells a separate commercial license.
The BrainMetShare page shows the Stanford University School of Medicine Research Use Agreement with "Brain Mets Dataset" as the dataset name. It also states that the dataset is for non-clinical research use only and must not be relied upon in patient care.
What you can do
- No
- Not stated
- No
- Not stated
What you can share
- No
- No
- Share trained models Not stated
What you must do
- Yes
- Share alike No
- Yes
- Ethics approval No
- Manuscript review No
- Release code No
- Return results No
- No
Limits
- Yes
- Location limits No
Commercial license: Yes Companies can apply for a one-year license per dataset, reviewed by a Stanford committee. The page lists an annual fee of USD 70,000 per dataset for agreements contracted in FY25. Details
Citation
Grøvik E, Yi D, Iv M, Tong E, Rubin D, Zaharchuk G. Deep Learning Enables Automatic Detection and Segmentation of Brain Metastases on Multisequence MRI. J Magn Reson Imaging 51(1):175-182 (2020). https://doi.org/10.1002/jmri.26766. Dataset DOI: https://doi.org/10.71718/z66c-qr59
All numbers
Every number on this page, as stored in stats.csv, with its source.
| Measure | Breakdown | Value | Source |
|---|---|---|---|
| Subjects | total consecutive patients imaged June 2016 to June 2018 | 156 | grovik2020 Patient Population |
| Studies | total one whole-brain MRI study per patient | 156 | aimi-brainmetshare Dataset Details |
| Subjects | split=train released with lesion masks | 105 | aimi-brainmetshare Assignment of Labels |
| Subjects | split=test released without masks | 51 | aimi-brainmetshare Assignment of Labels |
| Subjects | sex=female | 105 | grovik2020 Table 1 |
| Subjects | sex=male | 51 | grovik2020 Table 1 |
| Mean age | total | 63 | grovik2020 Patient Population |
| Age SD | total | 12 | grovik2020 Patient Population |
| Minimum age | total | 29 | grovik2020 Patient Population |
| Maximum age | total | 92 | grovik2020 Patient Population |
| Subjects | condition=brain_metastasis at least one metastasis per patient | 156 | grovik2020 Patient Population |
| Subjects | condition=lung_cancer primary cancer | 99 | grovik2020 Table 1 |
| Subjects | condition=breast_cancer primary cancer | 33 | grovik2020 Table 1 |
| Subjects | condition=melanoma primary cancer (skin/melanoma) | 7 | grovik2020 Table 1 |
| Subjects | contrast=T1w pre-contrast T1 spin echo (3D CUBE) | 156 | aimi-brainmetshare Dataset Details |
| Subjects | contrast=T1w_ce post-contrast T1 spin echo and post-contrast IR-prepped FSPGR | 156 | aimi-brainmetshare Dataset Details |
| Subjects | contrast=FLAIR post-contrast T2 FLAIR | 156 | aimi-brainmetshare Dataset Details |
| Subjects | contrast_set=FLAIR+T1w+T1w_ce all four sequences for every case | 156 | aimi-brainmetshare Dataset Details |
| Subjects | field_strength=1.5 | 18 | grovik2020 Patient Population |
| Subjects | field_strength=3 | 138 | grovik2020 Patient Population |
Sources
The keys used in the table above.
- grovik2020 Grøvik et al. 2020, Deep Learning Enables Automatic Detection and Segmentation of Brain Metastases on Multisequence MRI (JMRI 51(1)) paper
- aimi-brainmetshare Stanford AIMI BrainMetShare dataset page website
- datacite-brainmetshare DataCite record of the BrainMetShare dataset DOI (issued 2020) website