Abstract
This thesis focuses on new ways to more efficiently acquire the signal for fat/water separated MRI, also known as Dixon methods. In paper I, the concept of dual bandwidths was introduced to improve SNR efficiency by removing dead times in a spin echo PROPELLER sequence. By correcting for the displacement of fat, we were able to improve the motion correction. This required additional considerations during reconstruction in order to avoid noise amplification, which was solved with a noise-whitening Tikhonov regularization.
Paper II explores the combination of fat/water separation in k-space with partially acquired data, i.e. partial Fourier sampling. With reduced sampling coverage comes the ability of increased spatial resolution, which is often limited in fat/water imaging, particularly in gradient echo sequences. A modified POCS routine was also developed with real-valued estimates, exploiting Hermitian symmetry to improve the inverse problem conditioning in the fully sampled region. A single-TR dual-bandwidth RARE (fast/turbo spin echo) sequence without dead times was developed in Paper III, which uses partial Fourier sampling with late and early echoes to improve the chemical shift encoding. The proposed sequence can acquire images with 0.5 mm in-plane resolution without dead times, with image quality exceeding current state-of-the-art techniques. An automated selection of gradient waveforms based on Cramér-Rao bounds was implemented on the scanner.
In Paper IV, the dual-bandwidth concept was generalized to continuous bandwidths. Instead of the conventional shift of a trapezoidal readout gradient, we describe a new method of encoding chemical shift by using asymmetric readout waveforms. Asymmetric readouts were implemented in a RARE sequence to completely remove dead times from multi-TR acquisitions, with typical scan time reductions of 25 %. The developed methods enable fat/water imaging with reduced scan times and increased spatial resolution, which has previously limited their use.
