Abstract
Image-guided medical robot interventions require high quality medical image as well as high imaging speed. Parallel MRI reconstruction accelerates imaging speed with keeping quality of image content. An infinite impulse response (IIR) model has been proposed to improve the finite impulse response (FIR) model, which is applied in generalized auto-calibrating partially parallel acquisitions (GRAPPA) image reconstruction method. Recursive terms of IIR GRAPPA are able to improve conventional GRAPPA reconstruction quality. However it has the limitation that outliers and noise lead to poor estimation in the recursive coefficients. On the other hand, auto-regressive moving average (ARMA) is one of the most common models in time series analysis. Time series analysis is using the system time series data obtained by the curve fitting and parameter estimation to establish the mathematical model and theoretical methods. We propose a novel scheme using nonlinear ARMA (NLARMA) model to address the noise and outlier problems in IIR GRAPPA reconstruction. The proposed method extends the linear MA model which has been applied in conventional GRAPPA by incorporating both recursive and nonlinear terms. The results of experimental phantom and in vivo brain datasets illustrate the proposed method can decrease noise and aliasing artifacts comparing with conventional GRAPPA and IIR GRAPPA reconstruction.