Parallel MRI reconstruction using B-spline approximation

                         Jan Petr

                  petrj5@cmp.felk.cvut.cz

Abstract:

Parallel MRI is a technique that exploits the intrinsic spatial
encoding of multiple receiver coils with distinct spatial
sensitivities to speed up MRI data acquisition. The increase in data
acquisition speed is achieved by sampling the k-space more sparsely in
the phase encoding direction. Thus, not all image information is
retrieved and the images are affected by aliasing. The missing k-space
information is synthesized using the sensitivity information from the
receiver coils. We present a new reconstruction method for parallel
MRI that works in the image domain. Unlike other image-domain methods
that estimates the reconstruction independently in each pixel using
smoothed sensitivity maps, we represent the reconstruction
transformation in B-spline basis and estimate the B-spline
coefficients using the raw calibration images by minimizing the
expected reconstruction error. This regularizes the estimation and
makes the method more robust to noise in the reference images.  We
present the results achieved on simulated and in-vivo data and compare
it with two widespread parallel MRI methods - GRAPPA and SENSE.