q3-MuPa: Quick, quiet, quantitative multi-parametric MRI using physics-informed diffusion models

MuPa-ZTE is a novel multi-parametric quantitative MRI protocol that enables fast and nearly silent scanning. However, accelerating multi-parametric acquisitions to meet clinical time constraints makes the reconstruction of accurate and clean qMRI maps increasingly challenging, particularly under severe undersampling and noise. In this work, we propose a physics-informed diffusion model for MuPa-ZTE qMRI mapping. A denoising diffusion probabilistic model is trained to map MuPa-ZTE weighted image series to T1, T2, and proton density maps, while the MuPa-ZTE forward model is incorporated as an explicit data consistency constraint during inference. The proposed method is trained entirely on synthetic data and evaluated on both synthetic data and real data under the nominal (∼4min) and fourfold-accelerated (∼1min) MuPa-ZTE acquisitions. Compared with dictionary matching and a purely data-driven diffusion model, the proposed approach yields accurate and less noisy 3D qMRI maps with improved structural fidelity. The integration of MuPa-ZTE acquisition with a physics-informed diffusion model, termed q3-MuPa, provides an acquisition-consistent framework for quick, quiet, and quantitative multi-parametric MRI.