Dixon‐based B0 self‐navigation in radial stack‐of‐stars multi‐echo gradient echo imaging

Purpose To develop a Dixon‐based B0$$ {\mathrm{B}}_0 $$ self‐navigation approach to estimate and correct temporal B0$$ {\mathrm{B}}_0 $$ variations in radial stack‐of‐stars gradient echo imaging for quantitative body MRI. Methods The proposed method estimates temporal B0$$ {\mathrm{B}}_0 $$ variations using a B0$$ {\mathrm{B}}_0 $$ self‐navigator estimated by a graph‐cut‐based water‐fat separation algorithm on the oversampled k‐space center. The B0$$ {\mathrm{B}}_0 $$ self‐navigator was employed to correct for phase differences between radial spokes (one‐dimensional [1D] correction) and to perform a motion‐resolved reconstruction to correct spatiotemporal pseudo‐periodic B0$$ {\mathrm{B}}_0 $$ variations (three‐dimensional [3D] correction). Numerical simulations, phantom experiments and in vivo neck scans were performed to evaluate the effects of temporal B0$$ {\mathrm{B}}_0 $$ variations on the field‐map, proton density fat fraction (PDFF) and T2∗$$ {\mathrm{T}}_2^{\ast } $$ map, and to validate the proposed method. Results Temporal B0$$ {\mathrm{B}}_0 $$ variations were found to cause signal loss and phase shifts on the multi‐echo images that lead to an underestimation of T2∗$$ {\mathrm{T}}_2^{\ast } $$, while PDFF mapping was less affected. The B0$$ {\mathrm{B}}_0 $$ self‐navigator captured slowly varying temporal B0$$ {\mathrm{B}}_0 $$ drifts and temporal variations caused by respiratory motion. While the 1D correction effectively corrected B0$$ {\mathrm{B}}_0 $$ drifts in phantom studies, it was insufficient in vivo due to 3D spatially varying temporal B0$$ {\mathrm{B}}_0 $$ variations with amplitudes of up to 25 Hz at 3 T near the lungs. The proposed 3D correction locally improved the correction of field‐map and T2∗$$ {\mathrm{T}}_2^{\ast } $$ and reduced image artifacts. Conclusion Temporal B0$$ {\mathrm{B}}_0 $$ variations particularly affect T2∗$$ {\mathrm{T}}_2^{\ast } $$ mapping in radial stack‐of‐stars imaging. The self‐navigation approach can be applied without modifying the MR acquisition to correct for B0$$ {\mathrm{B}}_0 $$ drift and physiological motion‐induced B0$$ {\mathrm{B}}_0 $$ variations, especially in the presence of fat.