Optimization of Regularization Parameter for GRAPPA Reconstruction
Qu P, Yuan J, Wu B, Shen G
The University of Hong Kong
The effectiveness of regularization to improve SNR in parallel imaging techniques has been reported in previous works, but how to optimize the regularization parameter remains a problem. In this study, three regularization parameter choice strategies are compared in GRAPPA reconstruction: the L-curve method, the fixed singular value (SV) threshold method, and a novel discrepancy principle approach. In vivo experiment results show that the discrepancy-based parameter choice strategy significantly outperforms the others. It can automatically choose nearly optimal parameters for the reconstructions so as to achieve good compromise between SNR and artifacts.