UCLA

Magnetic resonance spectroscopy (MRS) and spectroscopic imaging (MRSI) are powerful, non-invasive tools that are capable of assessing the concentrations and distributions of various metabolic compounds in vivo. Single-voxel MRS methods such as STEAM and PRESS measure the temporal signal from a specific, localized volume of interest. As such, single-voxel MRS does not require any type of spatial encoding, such as frequency and phase encoding which are used routinely in magnetic resonance imaging (MRI). Although simpler to implement for clinical applications, MRS methods are nonetheless limited in their ability to efficiently acquire spectra across large anatomical regions, since only a relatively small volume can be probed per measurement. On the other hand, multi-voxel acquisitions can be done with MRSI, which incorporates additional two-dimensional (2D) or three-dimensional (3D) spatial encoding dimensions (i.e., k-space) to resolve multiple spectra from a large volume or slice within a single scan session. However, conventional MRSI techniques currently in clinical use depend on sequential phase encoding of each spatial dimension, which often results in long scan durations. Therefore, the focus of much research in MRSI has been to accelerate the acquisition through various means such as by undersampling or by using, often also in combination with undersampling, advanced sampling methods such as simultaneous spatiotemporal sampling of one spatial dimension and the spectral (time) dimension. The latter approach is accomplished by implementing so-called echo-planar k-t trajectories, which interleave the acquisition of one frequency-encoded spatial dimension (k) with the temporal samples (t) necessary for resolving the spectrum. The other spatial dimensions are often resolved with conventional phase encoding. Thus, echo-planar spectroscopic imaging (EPSI) is able to accelerate an MRSI scan session by at least an order of magnitude. When first proposed in the mid 1980’s, EPSI was done with Cartesian trajectories and, since the late 1990’s, non-Cartesian trajectories such as spirals, concentric circular, rosette, and radial trajectories have been implemented for fast MRSI. These non-Cartesian trajectories provide advantageous trade-offs in imaging speed, signal-to-noise ratio, and motion robustness compared with Cartesian EPSI. More recently, as late as 2019, radial echo planar spectroscopic imaging (REPSI) has been described as a nascent subfield in proton (1H) MRSI. Although radial projections were the first to be demonstrated for MRI, the adoption of radial sampling for MRSI had only found limited applications for non-proton MRSI, such as for phosphorus (31P) and carbon (13C), and had not yet been demonstrated for in vivo 1H MRSI. This work presents a study of 1H MRSI in the human brain in vivo using radial echo-planar trajectories, as well as applications for diffusion-weighted MRSI. The capability of REPSI for further acceleration compared to Cartesian EPSI are shown within a compressed sensing framework, in which the undersampled REPSI data can be reconstructed with good fidelity by exploiting the sparsity of the data within a transform domain. In addition to its higher tolerance for accelerations, the motion robustness of REPSI is shown in free-breathing healthy liver and prostate acquisitions. Both MRS and MRSI methods are compatible with diffusion-weighted (DW) techniques. DW-MRS and DW-MRSI are able to explore the microstructural characteristics of tissues in vivo due to the predominantly intracellular compartmentalization of metabolites. Unlike water, which permeates both the intra- and extra-cellular spaces, most metabolites are confined within the intracellular space, so that their diffusion reflects the structure and function of tissues at the microscopic scale. This compartment-specific assessment of tissue structure enables a clearer understanding of the cellular-level conditions and alterations that underlie various pathologies. This work also presents the first demonstration of a diffusion-weighted technique, first proposed in the mid 1990’s and early 2000’s, for in vivo single voxel DW-MRS and DW-MRSI in the human brain. This so-called “single-shot diffusion trace-weighted” scheme had been untestable in humans, until recently, due to earlier hardware limitations of clinical scanners. The acquisition and processing of the single voxel DW-MRS data was optimized as a precursor for the spectroscopic imaging version of the sequence. It is shown that radial echo planar trajectories are particularly advantageous for DW-MRSI, due to their self-navigation capability that enables post-processing-based corrections of the diffusion-weighted data, which is susceptible to shot-to-shot phase and frequency inconsistencies. In the Appendix, further work in acceleration in the context of parallel imaging using low-rank approximations is also demonstrated for MRI acquisitions.