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

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging
• Computational Imaging

Background:

• Multi-parametric quantitative MRI (qMRI) protocols like MuPa-ZTE offer fast and nearly silent scanning.
• Accelerated acquisition in qMRI presents challenges in reconstructing accurate quantitative maps due to undersampling and noise.

Purpose of the Study:

• To develop a physics-informed diffusion model for robust qMRI mapping using the MuPa-ZTE protocol.
• To improve the accuracy and quality of T1, T2, and proton density maps reconstructed from accelerated MuPa-ZTE acquisitions.

Main Methods:

• A denoising diffusion probabilistic model was trained to generate qMRI maps from MuPa-ZTE weighted images.
• The MuPa-ZTE forward model was integrated as a data consistency constraint during inference.
• The model was trained on synthetic data and validated on both synthetic and real data.

Main Results:

• The proposed physics-informed diffusion model (q3-MuPa) produced accurate and less noisy 3D qMRI maps.
• Improved structural fidelity was observed compared to dictionary matching and purely data-driven methods.
• Effective performance was demonstrated under both nominal and accelerated (fourfold) MuPa-ZTE acquisition schemes.

Conclusions:

• The q3-MuPa framework enables efficient and high-quality quantitative multi-parametric MRI.
• Physics-informed deep learning models can overcome reconstruction challenges in accelerated qMRI.
• This approach facilitates quick, quiet, and quantitative multi-parametric MRI for clinical applications.