Understanding the GRAPPA Paradox
Beatty P, Brau A
GE Healthcare, Stanford University
Autocalibrating parallel imaging methods such as GRAPPA use low-frequency convolution kernels to calculate the weights needed to unfold the aliased images. Yet GRAPPA has been shown to accurately reconstruct images even if the image contains high-frequency components, such as when the reconstructed FOV is smaller than the object. In this work we examine the GRAPPA paradox: how can low-frequency convolution kernels accurately reconstruct images encoded with high-frequency coil sensitivities? To answer this question, we compare the unfold images generated by GRAPPA and self-calibrating SENSE reconstructions and show that, although functionally the same, they represent fundamentally different approaches to solving the reconstruction problem.