MAMA-MIAAI (Here We Go Again)
Bias-field-corrected breast DCE-MRI from the MAMA-MIA breast-cancer cohort, with automated and manual whole-breast masks
Open the source page ↗ See it in the directory
At a glance
- What it is
- Bias-field-corrected breast DCE-MRI from the MAMA-MIA breast-cancer cohort, with automated and manual whole-breast masks
- Area
- Breast
- Modality
- MR
- Format
- NIfTI
- Size
- 1,506 cases; 7,868 bias-field-corrected DCE-MRI images, 1,506 automated whole-breast masks, and 94 manual masks. 150.1 GB uncompressed (104.5 GB zip).
- Normals included
- No
- License
- CC BY-NC-SA 4.0
- Commercial use
- Not allowed under the free terms
- Access
- Open download, no account
- Status
- Verified: source page reviewed by Hounsfield Lab
- Added
- 2026-11-04
- Last checked
- 2026-11-11
- Source
- physionet.org/content/mama-miaai-herewegoagain/1.0.0/
How to access
Open the PhysioNet record for version 1.0.0 (published 13 October 2026, DOI 10.13026/w6wc-jh81). The page is marked Open Access: anyone can download the files if they follow the license, with no account and no credentialing. Use the zip, wget, or the PhysioNet Python package. The segmentation model and its weights are not in the zip. They are at https://github.com/FloXyPython/MRI_BreastSeg . A companion repository is https://github.com/FloXyPython/MAMA-MIAAI .
Limitations
NIfTI, not DICOM. These are N4 bias-field-corrected pre-treatment T1-weighted DCE-MRI scans derived from MAMA-MIA (TCIA collections DUKE, ISPY1, ISPY2, and NACT), not the original DICOM. PhysioNet's license field and license page are CC BY-NC-SA 4.0, so commercial use is not allowed and adaptations stay under the same license. The usage notes on the same page say the scans and masks are adapted material from MAMA-MIA and are released under CC BY-NC 4.0. LICENSE.txt in the file tree is empty. Read the license page before use. The abstract counts 7,868 images; the data description also says 1,506 bias-field-corrected sequences, one per case, with multiple contrast time points inside a case. Automated masks come from one model (MRI_BreastSeg; the DUKE automated masks used the earlier Breast U-Net). The authors note that residual segmentation errors can affect the bias correction. The 94 manual masks are a subset of ISPY1, ISPY2, and NACT. Two private Medical University of Vienna cohorts were used while training the model and are not in this release. Research use only, not for diagnosis. Breast-cancer cases only, not healthy controls.
Citation required?
Yes. Cite Schwarzhans et al., MAMA-MIAAI - Here We Go Again, PhysioNet version 1.0.0, DOI 10.13026/w6wc-jh81, and the MAMA-MIA paper named on that page (Garrucho et al., Scientific Data 2025). CC BY-NC-SA 4.0 requires attribution and share-alike.
Related open-source tools
- Pillar-0 (finetune code + HF weights): Open 3D radiology foundation model; finetuning framework, weights on Hugging Face
- nnU-Net: Self-configuring segmentation framework
- 3D Slicer: Open platform for visualization, segmentation and image computing
- TorchIO: Medical image loading, preprocessing and augmentation for PyTorch
- SimpleITK: Simplified interface to ITK for image I/O, filtering, segmentation and registration in Python, R, Java, C# and other languages
- ITK: C++/Python toolkit for segmentation and registration
More breast datasets
- Breast-Cancer-Screening-DBT (BCS-DBT): MG · CC BY-NC 4.0
- CBIS-DDSM: MG · CC BY 3.0
- CMMD: MG · CC BY 4.0
- Duke-Breast-Cancer-MRI: MR · CC BY-NC 4.0
- NCI Imaging Data Commons: CT, MR, PT, MG, CR · Mostly CC BY; a minority of files are CC-NC (each file is tagged)
- NYU fastMRI: MR · NYU fastMRI Dataset Sharing Agreement (internal research or education only)
Hounsfield Lab links to the source and does not host images. The license and access summary reflects the source page on the last-checked date; terms can change, so read the source before you download. Not legal advice.