Data

Overview

The MAMA-SYNTH challenge adopts a multi-center data design to promote algorithmic generalizability across scanners, acquisition settings, and patient populations.

The benchmark combines a large-scale training cohort with two independent external test cohorts. This design supports robust evaluation of virtual contrast-enhancement methods under clinically realistic distribution shifts.

Training cohort

The training cohort is based on the MAMA-MIA dataset, which contains pre-treatment breast DCE-MRI from 1,506 patients collected from more than 25 centers across the United States.

This dataset was also used in the first edition of the MAMA challenges for primary tumor segmentation and pathologic complete response prediction, and serves here as the development cohort for virtual post-contrast MRI synthesis.

External test cohorts

The test data are acquired from two independent external institutions:

  • Radboud University Medical Centre — The Netherlands
  • Instituto Alexander Fleming — Argentina

Each test case corresponds to a 2D slice extracted from a patient’s DCE-MRI examination. For each patient, the selected slice contains the largest malignant tumor area in the peak-enhancement phase.

All test images are fat-suppressed and acquired in the axial plane.

This external evaluation design is intended to assess the robustness of submitted methods across different scanners, imaging protocols, and patient populations.

Cohort summary

Attribute Training cohort (MAMA-MIA) Test cohort A (Radboud University Medical Centre) Test cohort B (Instituto Alexander Fleming)
Role Development / training cohort External test cohort External test cohort
Country / region United States The Netherlands Argentina
Number of cases 1,506 cases 200 cases 100 cases
Source Multi-center dataset Single external institution Single external institution
Centers 25+ centers 1 institution 1 institution
Imaging type Pre-treatment breast DCE-MRI 2D slice from breast DCE-MRI 2D slice from breast DCE-MRI
Slice selection - Largest malignant tumor area in the peak-enhancement phase Largest malignant tumor area in the peak-enhancement phase
Acquisition plane Axial (84.4%), Sagittal (15.6%) Axial Axial
Fat suppression - Yes Yes
Image dimension - 416 × 416 px 512 × 512 px
Magnetic field strength 1.5T (72.1%), 3T (27.9%) 3T 1.5T
Scanner manufacturer GE (64.1%), Siemens (27.3%), Philips (8.6%) Siemens GE
Contrast agents - DOTAREM (99%), GADOVIST (0.5%) DOTAREM, GADOVIST
Acquisition protocol - t1_fl3d_tra_Dixon_W -
TR - 5–10 ms 4.2 ms
TE - 2–5 ms 2 ms
Flip angle - 8–15° 12°
Slice thickness - Approximately 1 mm 1.1 mm
Pixel spacing - Range: 0.8654–0.9615 mm; Mean: 0.8689 mm; Median: 0.8654 mm; SD: 0.0157 mm -
Molecular subtypes - Luminal: 165 (85.7%); Triple Negative: 23 (9.4%); Other: 12 (4.9%) Luminal: 37 (37%); Triple Negative: 30 (30%); Other: 20 (20%)

Data access and usage

The challenge data are provided for the development and evaluation of methods within the scope of MAMA-SYNTH 2026.

Participants may train their methods using:

  • the data provided by the challenge, and/or
  • additional publicly available datasets

The use of private data is not allowed.

Any external dataset used in model development should be clearly documented as part of the submission.

Data policy

Participants must comply with the challenge governance requirements and with any organizer-defined restrictions regarding data usage, redistribution, and controlled-access repositories.

The use of non-public institutional data, private collections, or privately shared model-development resources is prohibited.

We further note that by participating in this challenge, participants agree to comply with EO 14117, 28 CFR Part 202, and Guide Notice NOT-OD-25-083 and acknowledge that the usage of NIH Controlled-access Data Repositories (CADRs) is prohibited in this challenge.

Further details regarding permitted data usage and submission requirements will be provided in the relevant challenge documentation.