Feasibility of a Grappa-Based Parallel Imaging Approach Exploiting Spatial–Spectral Correlations for Accelerated 31p MRSI
Abstract
Purpose
Magnetic resonance spectroscopic imaging (MRSI) provides spatially resolved metabolic information but remains constrained by long acquisition times, particularly for X-nuclei applications with low sensitivity. Parallel imaging methods such as GRAPPA have been applied to accelerate MRSI, though conventional implementations estimate coil weights using spatial autocalibration data and reconstruct each time point independently, without exploiting spectral correlations. This work evaluates a GRAPPA-based reconstruction that incorporates spatial–spectral information from k–t space to improve coil weight estimation and enable flexible undersampling strategies for ³¹P MRSI.
Methods
31P MRSI data were acquired in two healthy volunteers on a whole-body 7T MRI system using a 32-channel receive array and a three-dimensional chemical shift imaging sequence. Undersampling was performed along both in-plane spatial dimensions using CAIPIRINHA-based sampling patterns. In one subject, a fully sampled dataset was retrospectively undersampled (R=2×2) to assess reconstruction accuracy. In a second subject, data were acquired using prospective undersampling with the same acceleration. Reconstruction weights were estimated from an autocalibration region in k–t space, allowing spectral correlations during kernel fitting. Post-processing included phase correction, Gaussian filtering, and metabolite amplitude-quantification using AMARES. Reconstruction accuracy was evaluated using mean squared error of reconstructed spectral data and phosphocreatine amplitude maps.
Results
Reconstructed spectra showed close agreement with fully sampled data, with spectral mean squared error below 6%. Quantitative phosphocreatine maps demonstrated mean squared error below 7%, indicating minimal impact on metabolite estimation. Prospectively undersampled acquisitions produced metabolite maps of comparable quality, confirming feasibility under realistic acquisition conditions.
Conclusion
Incorporating spatial–spectral correlations during GRAPPA calibration enables undersampling along all spatial dimensions without requiring a fully sampled spatial axis. Applied to ³¹P MRSI at 7T, this approach supports accelerated acquisition while preserving spectral fidelity and quantitative accuracy, suggesting a practical strategy for improving the feasibility of spectroscopic imaging for low-sensitivity nuclei.