An Efficient MR Image Reconstruction Method for Arbitrary K-space Trajectories Without Density Compensation

Non-Cartesian sampling is widely used for fast magnetic resonance imaging (MRI). The well known gridding method usually requires density compensation to adjust the non-uniform sampling density, which is a major source of reconstruction error. Minimum-norm least square (MNLS) reconstruction, on the other hand, does not need density compensation, but requires intensive computations. In this paper, a new version of MNLS reconstruction method is developed using maximum likelihood and is speeded up by incorporating novel non-uniform fast Fourier transform (NUFFT) and bi-conjugate gradient fast Fourier transform (BCG-FFT) techniques. Studies on computer-simulated phantoms and a physically scanned phantom show improved reconstruction accuracy and signal-to-noise ratio compared to gridding method. The method is shown applicable to arbitrary k-space trajectory. Furthermore, we find that the method in fact performs un-blurring in the image space as an equivalent of density compensation in the k-space. Equalizing MNLS solution with gridding algorithm leads to new approaches of finding optimal density compensation functions (DCF). The method has been applied to radially encoded cardiac imaging on small animals. Reconstructed dynamic images of an in vivo mouse heart are shown