UCLA

This dissertation is concerned with solving two problems in multi-dimensional magnetic resonance spectroscopic imaging (MRSI): the use of non-uniform, under-sampling (NUS) together with non-linear, iterative reconstruction to accelerate the acquisition of 4D (2 spatial+2 spectral) MRSI scans in vivo to clinically acceptable times, and the characterization of spatio-temporal phase errors during 4D MRSI acquisition so they can be decoupled and removed from the appropriate domain in post-processing. In support of these goals, the Cambridge reconstruction algorithm was implemented from partially documented sources in order to solve the Maximum Entropy (MaxEnt) reconstruction problem for 4D MRSI. It was compared to Compressed Sensing (CS) reconstruction down to NUS rates as low as 4X in vivo in the human breast and was found to produce superior results. Additionally, the Group Sparse (GS) iterative reconstruction problem of 4D MRSI was defined and the solution within the Split-Bregman iterative framework was derived. It was compared to MaxEnt, CS, and Total Variation (TV) reconstructions and demonstrated the best metabolite peak reproduction, lowest mean metabolite peak RMSEs, and best denoising characteristics down to NUS rates as low as 10X in the human brain in vivo. Lastly, the Interleaved Navigator Scan corrected J Resolved Echo Planar Spectroscopic Imaging (INSEP-JRESI) pulse sequence was implemented for this dissertation, which acquires a reference navigator scan at each TR in order to remove phase errors caused by B0 field drift in 4D EP-JRESI data. A spatio-temporal framework of 4D MRSI acquisition phase errors was derived using the INSEP-JRESI sequence. With this framework in place, the various sources of phase errors that are coupled in space and time could be decoupled and removed from the data in post-processing. Gray matter phantom results using the new post-processing technique were compared to Klose's method-based post-processing and resulted in higher metabolite peak amplitudes, improved spectral line-shapes, and a more even distribution of metabolite peak energy throughout the homogeneous phantom. These projects are all original work completed in support of this dissertation by Brian Burns working under Dr Albert Thomas.