q3-MuPa: Quick, Quiet, Quantitative Multi-Parametric MRI using Physics-Informed Diffusion Models

TL;DR

q3-MuPa introduces a physics-informed diffusion framework that maps MuPa-ZTE five-weighted MRI images to quantitative T1, T2, and PD maps. The method trains on synthetic data and enforces data consistency with the MuPa-ZTE forward model during inference, enabling high-quality 3D qMRI from either a 4:33 or a 1:09 MuPa-ZTE scan. Key contributions include a synthetic data pipeline, an explicit data-consistency mechanism, and demonstrating robust 3D qMRI performance with substantial acceleration while preserving anatomical detail. The approach shows strong potential for clinical translation by reducing scan time and maintaining quantitative accuracy across healthy and pathological brains.

Abstract

The 3D fast silent multi-parametric mapping sequence with zero echo time (MuPa-ZTE) is a novel quantitative MRI (qMRI) acquisition that enables nearly silent scanning by using a 3D phyllotaxis sampling scheme. MuPa-ZTE improves patient comfort and motion robustness, and generates quantitative maps of T1, T2, and proton density using the acquired weighted image series. In this work, we propose a diffusion model-based qMRI mapping method that leverages both a deep generative model and physics-based data consistency to further improve the mapping performance. Furthermore, our method enables additional acquisition acceleration, allowing high-quality qMRI mapping from a fourfold-accelerated MuPa-ZTE scan (approximately 1 minute). Specifically, we trained a denoising diffusion probabilistic model (DDPM) to map MuPa-ZTE image series to qMRI maps, and we incorporated the MuPa-ZTE forward signal model as an explicit data consistency (DC) constraint during inference. We compared our mapping method against a baseline dictionary matching approach and a purely data-driven diffusion model. The diffusion models were trained entirely on synthetic data generated from digital brain phantoms, eliminating the need for large real-scan datasets. We evaluated on synthetic data, a NISM/ISMRM phantom, healthy volunteers, and a patient with brain metastases. The results demonstrated that our method produces 3D qMRI maps with high accuracy, reduced noise and better preservation of structural details. Notably, it generalised well to real scans despite training on synthetic data alone. The combination of the MuPa-ZTE acquisition and our physics-informed diffusion model is termed q3-MuPa, a quick, quiet, and quantitative multi-parametric mapping framework, and our findings highlight its strong clinical potential.

Figures (13)

• Figure 1: Schematic of a MuPa-ZTE acquisition. The PD-weighted ZTE image is acquired upfront, followed by magnetisation-prepared segmented ZTE readouts for T1-weighted and T2-weighted images. The synthetic images at the bottom illustrate example contrasts for each segment.
• Figure 2: A schematic of the inference workflow of the proposed method. With a trained noise-prediction model, the qMRI maps corresponding to the provided weighted image series ($y$) are generated via $T$ steps. Explicit data consistency (DC) is incorporated at certain time steps in pre-defined $\mathbf{t_{DC}}$ to improve the mapping accuracy. The data consistency operation is conducted via a gradient descent optimisation, in which the physics forward model of MuPa-ZTE acquisition is encoded.
• Figure 3: Results of experiments on synthetic test datasets. The inferences were conducted on 30 patches of size $40^3$ drawn from synthetic (a) long and (b) short MuPa-ZTE datasets. The hyperparameter search tests all combinations of start time steps $t_{start} \in \{200, 400, 600\}$ and loss thresholds $\tau \in \{0.005, 0.003, 0.001, 0.0005\}$. MAE, RMSE and SSIM were used as the quantitative metrics and were computed for each parameter (PD, T1, T2) separately using the synthetic ground truth maps as reference. The units of MAE and RMSE for PD, T1, T2 are a.u., seconds, seconds respectively.
• Figure 4: Quantitative evaluation of T1 and T2 mapping using the NIST/ISMRM phantom, comparing DictMatch, DL-Diffusion, and DL-Diffusion-DC. All results were produced based on long MuPa-ZTE acquisition. Each plotted point corresponds to the mean value within one phantom sphere. The true reference value is on the x-axis and the measured value on the y-axis. Error bars represent $\pm$1 standard deviation. Data points of T1 mapping results are horizontally offset for clarity.
• Figure 5: Example of mapping results (PD in a.u., T1 and T2 in ms) from all three methods: DictMatch, DL-Diffusion, and DL-Diffusion-DC, on one healthy volunteer. Two scenarios were evaluated: (a) long MuPa-ZTE (acquisition time: 4 min 33 s) and (b) short MuPa-ZTE (acquisition time: 1 min 9 s).