Rapid non-Cartesian Parallel Imaging Reconstruction on Commodity Graphics Hardware

Introduction In this abstract we advocate the concept of real-time non-Cartesian magnetic resonance imaging with minimal reconstruction delay. We focus on cardiac imaging although the presented techniques are generally applicable. Parallel imaging techniques such as SENSE [1] and k-t SENSE [2] have been available for years for realtime acquisition. Until now however, available reconstruction implementations have been slower than the speed of the acquisition by at least an order of magnitude, adding a significant delay between the acquisition and the subsequent display of images. In this work we have overcome this problem. We present a parallel algorithm implementing iterative reconstruction and display of two-dimensional non-Cartesian SENSE and k-t SENSE in less than the acquisition time on inexpensive commodity graphics hardware (GPUs).