Room: Exhibit Hall | Forum 8
Purpose: Magnetic resonance spectroscopic imaging (MRSI) can noninvasively interrogate metabolism in vivo. However, excessively long scan times have prevented its incorporation into routine clinical practiceÂ¹. Generalized autocalibrating partially parallel acquisitionsÂ² (GRAPPA) is a parallel imaging technique that allows one to significantly reduce scan time by selectively not acquiring every phase-encoding line and reconstructing the missing lines using a fully-sampled central autocalibration region. A GRAPPA algorithm is characterized by four parameters: the acceleration factor (R), the number of acquired autocalibration lines (NACS), the kernel dimension in the fully-sampled direction (kâ‚“), and the distance (S) between consecutive kernel positions as it traverses the autocalibration region. GRAPPA can be extended to MRSI, but how varying these parameters affects the accuracy of the reconstruction was poorly understood. This study quantitatively characterizes how each parameter influences reconstruction accuracy.
Methods: A fully-sampled, non-water suppressed 2D proton MRSI dataset (64x64 matrix, TR=280 ms, TE=2.2 ms, 24x24 cmÂ² FOV) was acquired from a healthy volunteer using a 24-channel head coil at 7 T. Artificial undersampling was retrospectively performed in MATLAB by deleting phase-encoding lines in a GRAPPA-like pattern. The missing lines were reconstructed, and reconstruction accuracy was quantified by calculating the mean-squared-error between the fully-sampled and reconstructed MRSI datasets. The full analysis was repeated for both possible undersampling directions since coupling between coil elements exhibits a directional dependence.
Results: As R increases and NACS decreases, scan time is reduced and reconstruction accuracy degrades, but keeping Râ‰¤3 and NACS â‰¥10 maintains less than a 10% error. Additionally, kâ‚“ should be set to 1 or 3, and S should be set to 1.
Conclusion: The behavior of the four requisite GRAPPA reconstruction parameters has been characterized for use in parallel MRSI with this coil. This analysis demonstrates that a 2.3-fold reduction in overall scan time is achievable within 10% accuracy.
Not Applicable / None Entered.