An Investigation of Raw Data Corrections for Radial Fat-Water Magnetic Resonance Imaging

Abstract

In addition to the inherent chemical phase shift between different proton species, fat-water MRI k-space raw data are corrupted by several sources such as magnetic field inhomogeneity; chemical shift phase accumulated during the data readout window and trajectory shifts due to non-ideal gradient performance. NMR signal can be modeled within a single voxel as the mixture of different type protons with all corrupting factors clearly defined. If multiecho data are acquired, the evolving fat-water signal can be described as a linear system which can be unmixed. A reversed readout-based method is investigated in this work to correct the field inhomogeneity for radial fat-water MRI data. In the Cartesian case, the field map can be estimated using an iterative approach when other corrupting factors are precisely modeled. In addition, accurate fat-water signal modeling includes the use of a multipeak fat spectrum, and precise sampling time information. Multipeak fat spectrum information is obtained from nuclear magnetic spectroscopy, and the precise sampling time information is based on the employed pulse sequence. On the other hand, fat-water image reconstruction from radial trajectory data requires non-uniform Fourier transformation including regridding, density correction and interpolation. All these procedures are inserted as a part of radial fat-water separation. The artifact caused by imperfect gradients for radial MRI is also discussed and corrected in this work. Reconstruction results for both Cartesian and radial data with all corrections applied are displayed and compared.