Free-breathing fast abdominal simultaneous T₁, T₂, PD and B₁ mapping via deep learning reconstruction

Multi-parametric quantitative MRI (mqMRI) enables non-invasive tissue characterization and holds significant clinical value in abdominal imaging. Conventional methods are restricted by long acquisition times and motion artifacts induced by physiological processes and irregular breathing, leading to registration errors between sequential acquisitions.

To develop and evaluate a framework for abdominal quantitative imaging under free-breathing conditions with reduced scan time and improved robustness to motion.

A novel framework integrating the longitudinal magnetization-controlled multiple overlapping-echo detachment (LMC-MOLED) sequence with synthetic data-driven deep learning reconstruction was proposed. To generate the synthetic dataset for training, a respiratory motion model combining global and local displacement vector fields was developed, ensuring temporal alignment with the inter-phase intervals of LMC-MOLED acquisitions. By applying interpolation to the original template, virtual imaging objects were generated to simulate respiratory-induced deformations while preserving anatomical textures, tissue boundaries, and inter-slice correlations. These objects were subsequently processed with MRI Bloch simulation to generate the final synthetic dataset, which was used to train a BGA-Multi-UNet architecture for motion correction and improved quantitative accuracy.

The proposed framework enabled simultaneous mapping of T₁, T₂, proton density, and B₁ under free-breathing conditions. With a 5-slice interleaved acquisition, the average acquisition time was reduced to approximately one second per slice without the need for respiratory gating. Both numerical and in vivo experiments demonstrated high accuracy and repeatability, with image quality comparable to conventional breath-hold acquisitions while mitigating motion-induced artifacts.

This framework offers a rapid and reliable solution for multi-parametric abdominal imaging under free-breathing conditions. The combination of synthetic motion modeling with deep learning reconstruction enables accurate quantitative mapping with effective motion correction, offering potential clinical benefits for efficient and robust abdominal magnetic resonance imaging.