Online dynamic magnetic resonance imaging based on radial sampling and GPU acceleration

Abstract

In many fields of MRI (Magnetic Resonance Imaging), especially in DMRI (Dynamic MRI) domain, long acquisition time often limits its clinical applications. The new online MRI technique can obviously accelerate the imaging time for DMRI based on under-sampling acquisition and sparsity performance between neighbor frames. The data acquisition of previous online MRI schemes was based on the Cartesian coordinate. Non-Cartesian sampling, such as radial sampling, can obtain higher data acquisition speed than conventional Cartesian sampling. In order to further accelerate the imaging speed for online MRI, radial under-sampling instead of Cartesian under-sampling, was proposed to shorten the data acquisition time for DMRI. However the reconstruction of non-Cartesian data is complicated and time consumed, so GPU (Graphics Processing Unit) was used to shorten the reconstruction time. In this primary study, the full sampled cardiac dataset was used as a reference dataset, radial under-sampling with 25% and 10% k-space coverage was simulated for the online imaging scheme. The results show that radial under-sampling scheme obtained higher image quality than that of Cartesian acquisition scheme, with its imaging time still satisfying online reconstruction after applying GPU acceleration.