Chemical shift imaging (CSI) holds promise for studying brain metabolism and function in healthy and disease states by providing spatial localization of metabolites. However, widespread clinical use is impeded by local B₀ field variations arising from tissue susceptibility differences (esp. near sinus cavities) that are inadequately compensated by standard 2^nd-order spherical harmonic (SH) shim coils, resulting in diminished spectral quality and poor water and lipid suppression [1]. Multi-coil (MC) local shim arrays [2] provide an efficient alternative to higher-order SH shim coils for nulling high-order local B₀ variations without suffering from the low efficiency, high eddy currents and expensive current drivers associated with SH coils; however, the MC shim arrays compete with RF receive arrays for space near the head. The recent introduction of integrated RF-shim coils [3,4] provides the advantages of MC shim arrays and RF receive arrays without compromising the performance of either system. They have previously been used to reduce geometric distortion in EPI images (Fig. 1) [3] and here we demonstrate their use for improving spectral quality in CSI.

Experiments were performed using an anthropomorphic head phantom on a Siemens Skyra 3 Tesla scanner (Siemens Healthcare, Erlangen, Germany) to assess improvements in CSI data provided by a previously-described 32ch RF-shim array (Fig. 1) [3] over standard 2^nd-order SH shims. Phantom: A realistic head model [5] was 3-D printed in ABS plastic shells forming brain and muscle compartments (Fig. 2). The phantom includes a sinus air cavity that mimics ΔB₀ patterns in the frontal lobes in vivo. The brain compartment was filled with “Braino” metabolite solution [6] containing GABA, glutamate, choline, creatine, and NAA at 5X typical in vivo concentrations to reduce the need for signal averaging. Thus the phantom reflects realistic field maps (B₁⁺ and B₀) and spectral information encountered in vivo.

Acquisition: Constant density spiral-based k-space trajectories were appended to conventional PRESS-box excitation for single-slice, time-efficient spectral-spatial encoding [7]. Three TRs encoded a Cartesian matrix of (x,y,f) = (12,12,460) points (zero-padded to 16x16x512) over a 24cm FOV and spectral bandwidth of 1450Hz, for an overall voxel size of 16cc (2x2cm in-plane with 4cm slice). The PRESS-box excited VOI=[80x140x40]mm is inscribed wholly within the brain compartment. The spiral k-space data are 2X-gridded using a Kaiser-Bessel kernel. Ten averages yield NAA SNR ~ 25 for a total imaging time of 90s (TE=30ms and TR=2s). Within the shim adjust volume covering the PRESS-box excitation VOI, there were 24 target CSI voxels. Shimming: Scanner 2^nd-order shims were applied over the PRESS-box VOI and spiral CSI data were acquired, followed by gradient-echo B₀ field maps (2mm slice, 2.4mm in-plane). A previously mapped basis set of unit-current B₀ field maps for the shim coils were used to minimize a least squares objective on residual B₀ field subject to maximum current per loop (2.5A) and total current in the array (35A) constraints. After the application of optimal MC shim currents, the field mapping and CSI acquisitions were repeated. The high-resolution field maps were used to quantify shim performance using the standard deviation of ΔB₀ over the whole excited volume (σB₀^GLOBAL) as well as over each individual CSI voxel (σB₀^LOCAL) [1].

Multi-coil shims provide a 50% reduction in σB₀^GLOBAL (Fig. 3) and similar improvements in σB₀^LOCAL within the majority of CSI voxels (Fig. 4). Good agreement was obtained between calculated and acquired B₀ field maps with the MC shims applied (using 9A of total current for the prescribed shim). As compared with 2^nd-order shim field maps, the MC shims reduce peak ΔB₀ and produce a more even distribution of ΔB₀ over the whole CSI volume. Spectral linewidths were improved in 13 out of the 24 encoded voxels, remained comparable in 8 voxels, and modestly worsened in 3 voxels (Fig. 4). MC shimming substantially improved spectra in the posterior area of the PRESS-box where the 2^nd-order shim had compromised ΔB₀ while attempting to shim the anterior frontal lobe B₀ hotspot. Moreover, the large reduction in σB₀^GLOBAL enabled by MC shims is expected to benefit the performance of frequency selective RF pulses used for VOI excitation, water & lipid suppression, and spectral editing [1]. However, mismatch exists in parts of the frontal lobe (e.g. voxel A) between σB₀^LOCAL and the linewidth, possibly due to shifts in the acquired voxel location caused by ΔB₀. Since B₀ inhomogeneity in the “no-care” region is much worse in the MC shim case, the proposed methods would benefit from swapping the PRESS-box RF pulses to an excitation module that has sharper spatial localization and less sensitivity to chemical shift (e.g. LASER [8]).