Nady M. Obeid1, Ian C. Atkinson2, Keith R. Thulborn2, Wen-Mei W. Hwu1
1Electrical & Computer Engineering, University of Illinois at Urbana-Champaign, Champaign, IL, United States; 2Center for Magnetic Resonance Research, University of Illinois at Chicago, Chicago, IL, United States
In this work, we propose and elaborate on a method for accelerating gridding for the reconstruction of Non-Cartesian MRI using graphics processors (GPUs). Gridding interpolates the non-Cartesian input data onto a Cartesian grid as a preprocessing step to the FFT. We will explain the transformations needed to port the original CPU algorithm to the GPU, and how those transformation help improve performance. Our GPU-accelerated algorithm achieves up to 29X in runtime speedup for the gridding step, making fast reconstruction of non-Cartesian MR imaging data possible.