Radio frequency pulse designs for high resolution magnetic resonance imaging

Abstract

High resolution magnetic resonance imaging (MRI) is desirable for better in vivo tissue characterization. The overall aim of this dissertation is to develop new RF pulses that address problems hindering high-resolution MRI. MR spectroscopic imaging (MRSI), diffusion MRI (dMRI), and functional MRI (fMRI) are three applications of interest. For MRSI, we proposed to use patient-tailored spectral-spatial pulses to substitute the spectrally selective pulses used in the time-efficient chemical-shift based water suppression methods, whose performance is degraded by the subject-dependent B1 inhomogeneity that increases at ultra-high field. MR spectra of high spatial resolution with good quality were obtained in in-vivo experiments within clinically feasible durations using the proposed water suppression sequence. For dMRI, Generalized SLIce Dithered Enhanced Resolution (gSlider) is a newly developed RF encoding method to acquire high-resolution images. However, the peak power of the multi-banded refocusing pulse of this technique can become too high and require the use of the variable-rate selective excitation (VERSE) algorithm, which can mitigate this issue but brings in B0-related distortion in the slice profile. Here we show that the refocusing pulse can be root-flipped to minimize its peak amplitude and obviate the use of VERSE, while preserving gSlider encoding and linear-phase spin. High isotropic resolution in vivo whole-brain diffusion images were acquired with gSlider-SMS using proposed RF encoding pulses. For fMRI, MR Corticography is a developing imaging technique which aims for high resolution functional imaging. It will use inner volume suppression (IVS) to enable highly accelerated imaging, by reducing g-factor and suppressing physiological noise from ventricle cerebrospinal fluid. However, the subject-tailored 3D parallel-transmit RF pulse design for IVS has prohibitive memory and computational requirements if the conventional spatial domain formulation is used. We proposed a highly parallelizable k-space domain design method to substitute the conventional spatial domain deign method. The proposed k-space domain design largely decreased the computation time and provided equal IVS performance as the conventional spatial domain design in simulations. Its ability to accommodate excitation k-space undersampling and correct off-resonance were also shown.