Source-linked AI summary
MAMA-FLUX.2: Image-to-Image Synthesis of Post-Contrast Breast DCE-MRI for the MAMA-SYNTH Challenge
Kamil Kwarciak, Marek Wodzinski
TL;DR
Breast DCE-MRI synthesis could reduce reliance on gadolinium, but models must reproduce localized tumor enhancement while preserving patient-specific anatomy. The paper adapts FLUX.2-Klein-4B with LoRA and regional supervision, finding that moderate regional weighting gives the strongest reported trade-off in its development ablation. The study remains exploratory because hyperparameters were selected on only 10 development cases and the model is slice-wise.
Problem
Post-contrast breast MRI requires gadolinium, while synthesis must preserve anatomy and localized tumor enhancement for the MAMA-SYNTH task.
Method
The method uses conditional latent rectified-flow modeling with LoRA adaptation and a training objective combining global flow matching, tumor-region supervision, and stable-foreground preservation.
Results
Moderate regional supervision produced the most favorable trade-off between global image fidelity and tumor-focused performance, with λ_tumor = 0.25 and λ_stable = 0.1 selected in the development ablation.
Takeaways & Limitations
Conditional latent rectified-flow modeling combined with tumor-aware and anatomy-preserving supervision is a promising formulation for pre-to-post contrast breast MRI synthesis.
Takeaways & Limitations
The exploratory ablation used only 10 development cases, and the slice-wise method does not enforce volumetric consistency.
Abstract
from arXiv · showhide
Dynamic contrast-enhanced breast MRI is central to cancer diagnosis and monitoring, but requires gadolinium-based contrast agents. In this work, we address pre-to-post contrast breast MRI synthesis for the MAMA-SYNTH challenge. We propose MAMA-FLUX.2, a conditional latent flow-matching approach based on FLUX.2-Klein-4B. The pre-contrast image is encoded as spatial conditioning, while the model predicts the flow field associated with the post-contrast target latent. To adapt the pretrained model efficiently, we use LoRA fine-tuning and introduce a regional training objective combining global flow matching, tumor-region supervision, and stable foreground regularization. We further investigate LoRA rank, intensity windowing, and regional loss weights on axial slices, prioritizing clinically relevant tumor-focused metrics. Our ablation study shows that moderate tumor and stable-foreground weighting improves the trade-off between image fidelity and tumor-region accuracy. The final model achieves the best overall balance with LoRA rank/$α=64/64$, $\mathrm{MHA}_{\max}=25$, $λ_{\mathrm{tumor}}=0.25$, and $λ_{\mathrm{stable}}=0.1$. These results demonstrate that compact pretrained rectified-flow transformers can be adapted for contrast-enhanced MRI synthesis using parameter-efficient fine-tuning and task-aware regional losses.
1 Introduction
The paper targets virtual post-contrast breast MRI synthesis from pre-contrast acquisitions, preserving anatomy while reproducing clinically relevant enhancement. It adapts a compact latent rectified-flow transformer and adds tumor-aware, anatomy-preserving supervision, with ablations of key adaptation and loss settings.
- Virtual contrast enhancement synthesizes post-contrast breast MRI from pre-contrast acquisitions, potentially reducing gadolinium-related examination complexity and patient burden.
- MAMA-SYNTH evaluates peak-enhanced post-contrast image synthesis while requiring plausible enhancement and preservation of patient-specific anatomy.Tumor enhancement is often localized, making global image-level objectives potentially insufficient.
- Latent generative modeling, flow matching, and rectified-flow formulations provide an efficient framework for conditional medical image synthesis requiring realism and structural consistency.
- The proposed model adapts FLUX.2-Klein-4B by conditioning on the pre-contrast image and predicting the flow field for a noised post-contrast target latent.
- A three-component objective combines global latent flow matching, tumor-region supervision, and stable-foreground preservation to discourage unnecessary changes outside tumors.
- The study evaluates LoRA rank, LoRA scaling, and regional loss weighting using image-quality and tumor-focused metrics.
2 Methods
The method formulates pre-to-post contrast synthesis in VAE latent space using conditional rectified-flow modeling, then combines global, tumor-focused, and stable-foreground objectives.
- 2.1 Conditional Latent Flow Matching: Pre-contrast and post-contrast images are encoded into VAE latents for conditional latent rectified-flow synthesis.The model uses the pre-contrast latent as conditioning and reconstructs the synthetic post-contrast image through VAE decoding.
- 2.1 Conditional Latent Flow Matching: At sampled noise level t, the method constructs a noised target latent and trains a transformer to predict its corresponding flow field.The conditioning includes the pre-contrast latent, text embedding, and timestep embedding.
- 2.2 Tumor-Aware and Anatomy-Preserving Objective: The training objective combines standard global flow matching with tumor-region supervision and stable-foreground preservation.The global loss averages squared flow-matching errors over all latent elements, while regional terms target clinically relevant structures.
- 2.2 Tumor-Aware and Anatomy-Preserving Objective: Tumor supervision resizes and broadcasts the tumor mask to latent-error resolution, emphasizing synthesis accuracy within the lesion region.The tumor-region loss is computed using the masked latent errors.
- 2.2 Tumor-Aware and Anatomy-Preserving Objective: Stable-foreground preservation targets anatomy outside the tumor using an image-space mask that suppresses background and excludes strongly enhancing tissue.The thresholds are set empirically to τfg = −0.45 and τenh = 0.5 before resizing and broadcasting the mask.
- 2.2 Tumor-Aware and Anatomy-Preserving Objective: The final objective weights the tumor-region and stable-foreground losses relative to the global loss using λtumor and λstable.These coefficients control the relative contributions of the two regional objectives.
3 Experiments
The experiments evaluate MAMA-FLUX.2 as a plane-aware, parameter-efficient latent rectified-flow system for synthesizing post-contrast breast MRI from pre-contrast slices. The setup uses paired 2D data, frozen pretrained components, LoRA adaptation, and ablations of model and preprocessing choices.
- Data and preprocessing: Paired pre-contrast and peak-enhanced post-contrast 2D MHA slices from MAMA-SYNTH support regional supervision and evaluation with tumor masks.The preprocessing follows the official protocol and preserves spatial metadata for evaluation.
- Anatomical plane routing: A binary classifier routes each slice to an axial or sagittal LoRA adapter before conditional synthesis with the frozen FLUX.2-Klein-4B backbone.Routing uses image dimensions, intensity statistics, foreground descriptors, and occupancy profiles.
- Model and fine-tuning: The pipeline encodes the pre-contrast slice into VAE latent space as spatial conditioning while optimizing only attention-projection LoRA adapters.The VAE, text encoder, and pretrained transformer weights remain frozen.
- Ablation design: The main axial-slice search varies LoRA rank and scaling, intensity-window upper bounds, and regional loss weights before transferring the selected configuration to sagittal slices.The reported final window uses a z-score range of [−0.51, 25].
- Evaluation protocol: Inference uses 50 rectified-flow sampling steps with guidance scale 4.0, while Table 1 reports complete ablation configurations as mean ± standard deviation.Runtime profiling covers spatial sizes from 256×256 to 512×512.
4 Results
The axial-slice ablation finds that moderate regional supervision gives the most favorable balance between image fidelity and tumor-focused performance. Qualitative examples preserve anatomy and aligned enhancement patterns, but challenging enhancement patterns still produce concentrated errors.
- Quantitative ablation: λtumor = 0.25 and λstable = 0.1 achieve the highest DSC and lowest HD95 in the 10-case development ablation.The analysis evaluates DSC, HD95, tumor SSIM, MSE, and LPIPS.
- Quantitative ablation: Increasing stable-foreground weighting improves MSE and LPIPS but degrades segmentation performance when λtumor = 0.25.
- Quantitative ablation: Stronger tumor-region weighting improves tumor SSIM but does not provide the best overall balance across metrics.
- Qualitative results: The selected model preserves main anatomical structure and produces spatially aligned enhancement patterns in the qualitative examples.The selected configuration uses LoRA rank/α = 64/64, MHAmax = 25, λtumor = 0.25, and λstable = 0.1.
- Qualitative results: A challenging case shows higher reconstruction error, with discrepancies concentrated around enhancing tissue and tumor-adjacent regions rather than the background.
5 Limitations
The study’s conclusions are limited by exploratory tuning on a small development subset and by slice-wise modeling without volumetric consistency. The selected design choices require validation on larger, independent datasets and across acquisition protocols.
- Scope and generalization: The ablation uses only 10 development cases, so selected hyperparameters may not generalize to the hidden test set or external cohorts.
- Method constraints: Slice-wise training does not enforce volumetric consistency, and intensity windows, regional loss weights, and plane routing were selected empirically.
- Scope and generalization: The design choices should be validated on larger independent datasets and across different acquisition protocols.