Gazi Brains 2020
Gazi Brains 2020 Dataset (Turkish Brain Project)
Brain MRI of adults with high-grade glioma and of people without a brain tumor, built as a benchmark for tumor detection, classification and segmentation, with expert labels for tumor components, anatomy and non-tumoral findings. 100 subjects from 24 Turkish institutions.
Overview
Gazi Brains 2020 is a brain MRI benchmark from the Turkish Brain Project, prepared by medical experts at the Gazi University Faculty of Medicine with support from the Digital Transformation Office of the Presidency of Türkiye. It pairs 50 patients with histologically confirmed high-grade glioma and 50 subjects without a brain tumor, so that tumor detection and classification models can be trained on a balanced set. The data descriptor (Sagiroglu et al., PeerJ Computer Science 2025) reports baseline results of eight deep learning models for classification and detection, plus explainability maps.
Composition
There is one scan per subject. Every subject has axial T1-weighted, T2-weighted and FLAIR images. All glioma patients and 12 of the normal subjects also have a gadolinium-enhanced T1-weighted series. The paper counts about 1,000 to 1,060 2D slices per sequence and group. Mean age is 38.7 years in the normal group and 56.8 years in the glioma group; age and sex are released for every subject. In the Kaggle copy, subjects 1 to 50 are the glioma group and 51 to 100 the normal group.
Acquisition
Studies were collected in 2018 and 2019 at 24 institutions across Türkiye, on scanners from four vendors at 1.5 T or 3 T. Vendor and institution names are masked. Slice thickness, orientation and matrix size vary between sequences. For release, every sequence was registered to the subject's FLAIR image and the scans were defaced with hand-drawn masks that remove the nose, cheeks and teeth while keeping the orbits. The Synapse release is organized in BIDS with NIfTI images and includes MRIQC quality metrics.
Annotations
Six medical experts drew and cross-checked 12 labels covering anatomical structures, non-tumoral findings and tumor components. The paper shows tumor segmentations drawn on the post-contrast T1-weighted images.
Known limitations
- The cohort has 100 subjects from a single release; the paper says more patients will follow in a later version.
- The normal group is about 18 years younger on average than the glioma group, so age can act as a shortcut for classification.
- Post-contrast T1-weighted images are missing for 38 of the 50 normal subjects.
- Terms differ by copy: the descriptor paper excludes commercial use, while the Kaggle copy is labelled Apache 2.0. The Synapse page states no license.
- Both Synapse and Kaggle need a free account to download.
Cohort
Aggregate numbers from the sources below. Bars are relative to the 100 subjects.
Contrast / sequence
Groups can overlap
- FLAIR 100 100%
- T1-weighted 100 100%
- T2-weighted 100 100%
- T1-weighted, contrast enhanced 62 62%
Materials and Methods, The released dataset in Sagiroglu et al. 2025, PeerJ Computer Science
Condition
Groups can overlap
- Glioma 50 50%Age mean 56.8
- Healthy control 50 50%Age mean 38.7
Materials and Methods, The released dataset, Materials and Methods, The released dataset; Figure 1A in Sagiroglu et al. 2025, PeerJ Computer Science
Contrast / sequence by condition
Reported cross table. A dot marks a cell the source does not give.
| Glioma | Healthy control | |
|---|---|---|
| T1-weighted, contrast enhanced | 50 | 12 |
Materials and Methods, The released dataset in Sagiroglu et al. 2025, PeerJ Computer Science
Contrast combinations
How many subjects have exactly each set of contrasts.
| T1w | T1w_ce | T2w | FLAIR | Subjects with exactly this set |
|---|---|---|---|---|
62 | ||||
38 |
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
Different parts of the data carry different licenses. The summary on the right shows the most restrictive answer per rule.
Gazi Brains 2020 data availability statement (Sagiroglu et al., PeerJ Computer Science 2025)
Not a formal license. The creators' data descriptor states that the open-access data are available to researchers and experts except for commercial use, within the scope of the original license. The Synapse project page states no further terms.
What you can do
- No
- Not stated
- Not stated
- Create derived data Not stated
- Publish results Not stated
What you can share
- Share the data Not stated
- Share derived data Not stated
- Share trained models Not stated
What you must do
- Not stated
- Share alike No
- No
- No
- Manuscript review No
- Release code No
- Return results No
- Delete after use No
Limits
- No
- Location limits No
Commercial license: Not stated
Apache License, Version 2.0
A permissive software license, sometimes applied to data. It allows any use, copying, adaptation and redistribution, including commercial, as long as copies carry the license, keep the original notices and mark changed files.
What you can do
- Yes
- Yes
- Yes
- Yes
- Yes
What you can share
- Conditional
- Conditional
- Not stated
What you must do
- Conditional
- No
- No
- No
- Manuscript review No
- Release code No
- Return results No
- Delete after use No
Limits
- No
- Location limits No
Citation
Sagiroglu S, Terzi R, Celtikci E, Börcek AÖ, Atay Y, Arslan B, Sahin MC, Nernekli K, Demirezen U, Ozdemir OB, Özdem Karaca K, Azgınoğlu N. A novel brain tumor magnetic resonance imaging dataset (Gazi Brains 2020): initial benchmark results and comprehensive analysis. PeerJ Computer Science 11, e2920 (2025). doi:10.7717/peerj-cs.2920
Sources
Every number on this page comes from one of these documents. Each chart names the table or page it is taken from. The raw numbers are in stats.csv.