GPU accelerated iterative SENSE reconstruction of radial phase encoded whole-heart MRI

Isotropic whole-heart imaging has become an important protocol in simplifying cardiac MRI. The acquisition time can however be a limiting factor when acquiring large field-of-views at high spatial resolutions. To reduce acquisition times, a three-dimensional (3D) acquisition scheme combining Cartesian sampling in the readout direction with a radial sampling scheme in the phase encoding plane, was recently suggested [1]. This encoding scheme allows high undersampling factors in the phase encoding plane when obtaining data with a 32-channel coil array and employing non-Cartesian iterative SENSE [2] for reconstruction. To reconstruct the 3D volume, initially the fast Fourier Transform (FFT) is applied in the fully sampled readout direction, converting the remaining reconstruction problem into a series of two-dimensional radial reconstructions. Unfortunately this reconstruction is a time consuming process; using Matlab a reconstruction time of 3 minutes/slice when using 32 coils for a 256 slice acquisition was reported (the total reconstruction time was over 12 hours) [1]. The purpose of the present abstract is to demonstrate that the reconstruction time can be brought to a clinically acceptable level using commodity graphics hardware (GPUs) and a dedicated parallel implementation of the non-Cartesian SENSE reconstruction. Such an approach was recently demonstrated for real-time iterative SENSE reconstruction of twodimensional (2D) radial imaging [3].