Efficient NUFFT algorithm for non-Cartesian MRI reconstruction

We recently introduced a family of optimized interpolators to approximate the non-uniform Fourier transform of a finitely supported function. Theoretical comparisons indicated a significant improvement in performance over conventional approximations. In this paper, we study the utility of this approximation in the inversion of non-Cartesian MRI data. Using numerical simulations, we show that the new interpolators can provide iterative non-Cartesian inversion algorithms with considerably reduced memory demands. This will enable us to implement the algorithms on memory limited systems such as GPU, leading to a further acceleration in the iterative inversion of large MRI datasets.