Water signal suppression is critical for magnetic resonance spectroscopic imaging (MRSI). At ultra-high field (7T and above), the increased SNR can be used to significantly improve MRSI spatial resolution, but acquisition matrixes of 32x32 or higher only have acceptable measurement times when the TR is short (<300ms). The most time-efficient pulse sequence-based water signal suppression methods are based on pre-saturation using 90 degree excitation pulses followed by spoiling gradients (CHESS)^1,2,3. However, at ultra-high field, B1⁺ and B0 inhomogeneities degrade the performance of the pulses used by these methods. To address this, we propose to replace conventional spectrally-selective pulses with subject-tailored spiral in-out spectral-spatial (SPSP) saturation pulses⁴ that are designed using subject-specific B1⁺ and B0 maps.

Design Specifications The pulses comprise a train of identical tailored spiral in-out spatial subpulses that are weighted by a spectral envelope. The design targeted: Maximum total duration 26ms; Spectral FOV = 5.2 ppm (1550Hz) (so that passband replicas fell on either side of the metabolites of interest); 1 ppm wide passband (water-band) and transition band widths (300Hz); Stopband (metabolite-band) width 4.2 ppm (1250 Hz). The spectral FOV dictated a maximum spatial subpulse duration of 0.645 ms.

Spatial Pulse Design The 0.645 ms spatial subpulses were first designed to produce a uniform flip angle pattern over the brain slice of interest. The 0.645 ms spiral in-out trajectories had FOV 10 cm and resolution 3.2 cm. Prior to pulse design the waveforms were preemphasized⁵ and the final measured trajectory was used for design. Pulses were designed on a 32x32 spatial grid using a power-regularized matrix pseudoinverse⁶ that was alternated with a target phase pattern update to implement a magnitude least-squares⁷

Spectral Pulse Design A minimum-phase spectral envelope with a time-bandwidth product of 5.72 was then designed using SLR⁸. Root-flipping based on an exhaustive search of passband root-flip combinations was applied to reduce the peak amplitude of the pulses while maintaining their excitation amplitude profile⁹. The final pulse was constructed by replicating the spatial subpulse 41 times and weighting each replica by its position in the spectral envelope.

Experiments The pulses were implemented on a Philips 7T Achieva scanner (Philips Healthcare, Cleveland, Ohio, USA) and used to scan healthy subjects. To image the pulses’ saturation patterns they were played immediately before a gradient spoiler and the excitation pulse of a 2D spin echo sequence. The frequency of the pulses was varied during imaging. The pattern of a spectral pulse with the same duration and spectral characteristics was also imaged, as well as a zero pulse for dividing out receive sensitivities and other signal variations. A pulse-acquire MRSI experiment was performed using a crusher coil for lipid suppression at the skull¹⁰, and a CHESS sequence in which three spectral pulses were replaced by the tailored RF pulses was used for water suppression. That sequence had parameters: TR 340ms, TE 2.5ms, flip angle 30 degrees, matrix 18 x 18, 3 averages, acquisition time 6:21 min.

The total pulse design time (including spatial subpulse and spectral envelope designs) was 0.4 seconds on a Core i7 laptop. Root-flipping the spectral envelope reduced peak B1+ demand by 33% compared to the initial spectral envelope (Fig.1). Figure 2 shows a simulated comparison of spiral in and spiral in-out subpulses. The spiral in trajectory achieves ~2x finer resolution than the spiral in-out trajectory (Fig. 2A), but the spiral in-out pulse has 47% lower peak amplitude and 40% lower SAR since it visits the center of k-space twice (Fig. 2B). Figure 2C shows that the spiral in-out subpulse is also more robust to off-resonance, since small off-resonance effects cancel out in redundant spiral in-out readouts¹¹. Figure 3 shows that gradient preemphasis converged to triangular-shaped waveforms which produced measurements that closely matched the desired waveforms. The minimum TR of the in vivo experiment with the saturation sequence using a single pulse was 53 ms. In the passband, the mean normalized residual signal was 0.53 for the conventional spectral RF, and 0.16 for the tailored pulse (3.3x lower) (Fig.4). Figure 5 shows that MR spectra with reasonable quality were obtained using tailored RF pulses for water suppression. In that data, residual water signal was removed by fitting unsuppressed data to the water-suppressed data.We have proposed and validated spiral in-out spectral-spatial tailored RF pulses for water suppression in MRSI measurements. Combining this fast water suppression technique with a crusher coil for lipid suppression will enable artifact-free ultra-high resolution MRSI in the brain within 10 min acquisition times.