A Nonlinear ARMA Model for GRAPPA Reconstruction

Yuchou Chang¹, Leslie Ying¹

IIR GRAPPA incorporates recursive terms to improve conventional GRAPPA, but has the limitation that outliers and noise lead to poor estimation in the recursive coefficients. A novel method using nonlinear ARMA (NLARMA) model is proposed to address the issue in IIR GRAPPA reconstruction. The proposed nonlinear AMRA model improves over the linear MA model used in conventional GRAPPA by incorporating both recursion and nonlinearity. The experimental results demonstrate that the proposed method is able to significantly improve the reconstruction quality of the conventional GRAPPA and IIR GRAPPA in suppressing noise and artifacts.