MPFlow: Multi-modal Posterior-Guided Flow Matching for Zero-Shot MRI Reconstruction

Researchers developed MPFlow, a new zero-shot MRI reconstruction framework that uses multi-modal data and rectified flow to improve medical imaging quality. The system reduces tumor hallucinations by 15% while using 80% fewer sampling steps compared to existing diffusion methods, potentially advancing AI applications in medical diagnostics.

• →MPFlow achieves comparable image quality to diffusion baselines while using only 20% of the sampling steps, significantly improving efficiency.
• →The framework reduces tumor hallucinations by more than 15% as measured by segmentation dice scores on medical datasets.
• →Cross-modal guidance leverages complementary MRI acquisitions without requiring retraining of the generative prior.
• →PAMRI pretraining strategy enables shared representations across different MRI modalities for better anatomical fidelity.
• →Extensive testing on HCP and BraTS datasets demonstrates practical applicability for clinical workflows.