Multiresolution MR image reconstruction from a pseudo-hex lattice using separable DID-DIS FFT.

摘要

Common MRI sampling patterns in k-space, such as spiral trajectories, have nonuniform density and do not lie on a cartesian grid. This paper illustrates that the sampled data on a spiral trajectory can be efficiently regridded onto a pseudo hexagonal lattice using a nearest neighbor regridding algorithm. Also, it is shown that separable FFT using cartesian indices can be used to compute the DFT of arbitrary lattices, and by Cooley-Tukey type decompositions, it is possible to efficiently reconstruct an image by taking advantage of the inherent multiresolution nature of the signal, with almost no further degradation from the originally sampled data. Although multiresolution experiments described in this paper require about 40% more computation compared to the standard method, the phase error resulting from perturbation in k-space due to regridding is four times less and amplitude distortion and aliasing is noticeably less compared to the standard method. Moreoever, when implementing multiresolution methods, a pseudo-hexagonal lattice requires 7% more computation compared to a square lattice while the resulting phase error is 6.2% less and the construction of FOV is 12.5% more efficient compared to a square-lattice.