Liver Proton Density Fat Fraction (PDFF) at 0.55T Using Locally Low-Rank Deep Learning (LLR-DL) Reconstruction

Motivation: New medications like Resmetirom have increased the demand to assess fatty liver diseases with Proton Density Fat Fraction (PDFF) MRI. Imaging at low-field can help meet this demand but is challenging due to low signal-to-noise ratio (SNR). Goal(s): Provide a framework to reconstruct high SNR images and precise PDFF and R2* maps. Approach: Develop and compare multi-echo Dixon sequence with joint deep learning (DL)-based reconstruction of multiple contrasts at 0.55T with the established approach at 1.5T. Results: High SNR images reflected in the PDFF precision reaching a reproducibility coefficient of 0.23% and 1.52%, compared to 3.14% and 4.48% achieved using conventional parallel imaging. Impact: Multi-echo Dixon acquisitions with short echo times combined with iterative DL-based reconstruction using locally low-rank regularization yields high SNR images. This approach enables precise PDFF quantification and provides high-SNR R2* maps at 0.55T.