This study presents a finite difference regularized magnitude-least-squares algorithm that ensures robust RF shimming and small-tip-angle multi-spoke pulse design against excitation nulls and sub-optimal pulse solutions. It also calculates a monotonic trade-off between flip angle error and RF power. It was validated in simulations and experiments, and was effective for brain and knee imaging. During
an EPI-based fMRI at 7T with dynamic RF shimming, the algorithm ensured high
SNR throughout the human brain, compared to a near-complete local signal
loss by the conventional magnitude-least-squares algorithm. Overall, the algorithm streamlines the workflow for patient-tailored 2D multislice imaging at UHF.