Accelerated radially encoded tissue phase mapping

Abstract

Methods In 3 healthy volunteers, a mid-level short axis slice was acquired at 3 T (Achieva, Philips) using a 32-channel cardiac coil. Two types of segmented, triggered, and gated velocity encoding sequences were used: a Cartesian acquisition (fully sampled; R = 1) and radial acquisitions with undersampling factors of R = 1,3, and 6. Acquisition parameters were: FOV = 340 mm, resolution = 2x8 mm, acq. matrix = 172, a = 15°, 3 k-lines/ segment, gating window = 6 mm (except for the radial acquisitions in one volunteer: 8 mm), phase interval≈31 ms, 4-point balanced velocity encoding (Hadamard) with VENC = 30 cm/s. TR/TE = 6.3/4.6 ms (Cartesian) and TR/TE = 6.1/3.3 ms (radial). Reconstructed velocities were exported from the scanner. Higher undersampling was analyzed from a similar radial acquisition (R = 1) of a beating heart phantom [2] by leaving out profiles to simulate undersampling factors of R = 2 to 10. Each velocity encoding direction was reconstructed iteratively in Matlab by an intermediately regularized sparse SENSE algorithm with temporal total variation regularization [3]. Reconstructed in-plane and through-plane velocities were further processed in Matlab. Background phase errors were corrected according to a linear model. Median radial [vr(t)], circumferential [vc(t)] and longitudinal [vl(t)] velocities were obtained for each cardiac phase t and corrected by their mean velocity over time. Velocity curves were compared relative to fully sampled data (in vivo: Cartesian acquisition) by correlation coefficient CC. SNR of velocities within the myocardium was calculated as local mean to standard deviation ratio and averaged over all cardiac phases and volunteers.