MAMA-SYNTH 2026 β BreizhCT (SAM Synth)
Weights of our submission to the MAMA-SYNTH 2026 challenge: virtual peak-enhancement breast DCE-MRI synthesized from a 2D pre-contrast T1-weighted slice.
CV_0.pt...CV_4.pt: five 2D U-Net++ (ResNet-34 encoder) models, one per fold of a patient-level split of MAMA-MIA, trained with KonfAI using an MAE loss and a SAM 2.1 perceptual loss.- Code, inference instructions and Grand Challenge container: https://github.com/vboussot/mama-synth
Hidden test set (300 cases): 12th of 30 entries.
Trained on MAMA-MIA (CC BY-NC 4.0); non-commercial use.
Inference Providers NEW
This model isn't deployed by any Inference Provider. π Ask for provider support