UCLA

Magnetic resonance spectroscopy (MRS) is used to obtain localized biochemical information noninvasively based on the principles of nuclear magnetic resonance. 1H in vivo spectra consist of a large number of metabolites in a relatively small spectral range, making identification difficult. Multidimensional MRS incorporates a variable evolution period to enhance the information content and increase spectral dispersion. Recently, multidimensional MRS has been combined with echo planar gradient readout techniques to produce multidimensional magnetic resonance spectroscopic imaging (MRSI). Despite the fast imaging acquisitions, these scans are long for in vivo studies, so more efficiently sampling strategies were investigated.

The first strategy consisted of nonuniform undersampling (NUS) of the volume spanned by the phase-encoded spatial dimensions and the indirect spectral dimension in 5 dimensional (3 spatial + 2 spectral) MRSI. Nonlinear reconstruction was performed according to the theory of compressed sensing (CS) using the split Bregman framework. Formulations that promoted sparsity of the data and its spatial finite differences in 5D J-resolved brain studies were applied, and results were compared favorably to a time-equivalent single slice J-resolved scan. In 5D correlated MRSI calf studies, reconstruction minimized the group sparsity of nearby points, which produced much better results than reconstruction that minimized the overall sparsity of the data.

The second strategy used concentrically circular k-space trajectories instead of the conventional rectilinear ones. Concentric circles have the advantages of reduced hardware demands, higher achievable spectral bandwidth, less sensitivity to motion, and faster k-space coverage. Single slice 4D (2 spatial + 2 spectral) correlated MRSI using concentrically circular trajectories was compared to a rectilinear counterpart and showed similar data quality. An improved single slice J-resolved MRSI sequence was presented. The new sequence used adiabatic refocusing pulses that are less sensitive to RF field inhomogeneity and result in reduced chemical shift displacement error compared to conventional pulses. Comparison was made to the nonadiabatic sequence with the same echo time as well as with its minimum echo time.