Empirical mode decomposition based GRAPPA reconstruction algorithm for parallel MRI

Abstract

This paper presents a zero-assisted bivariate empirical mode decomposition (bEMD) based generalized autocalibrating partially parallel acquisitions (GRAPPA) for improved reconstruction of magnetic resonance images. Each line in the readout direction of the zero filled sub-sampled k-space is decomposed into intrinsic mode functions (IMFs) using bEMD to obtain multiple EMD domain k-spaces constructed from the respective IMFs of each line. GRAPPA reconstruction is then performed on these EMD domain k-spaces to calculate the missing lines of the sub-sampled k-spaces. These fully reconstructed EMD domain k-spaces are combined together to obtain the final reconstructed k-space of the desired object. The results of experimental tests using phantom, human brain and spine data demonstrate that the proposed EMD domain GRAPPA reconstruction method can result in better image quality than that of the conventional GRAPPA. It is also shown that the proposed approach can be incorporated with other k-space based reconstruction algorithms, e.g. iterative and nonlinear GRAPPA, for further reduction of reconstruction artifacts, in particular, at higher acceleration factor.