Panu Tapani Vesanen1,2, Fa-Hsuan Lin2, Risto Juhani Ilmoniemi1
1Department of Biomedical Engineering and Computational Science, Helsinki University of Technology, Espoo, Finland; 2Institute of Biomedical Engineering, National Taiwan University, Taipei, Taiwan
Compressed sensing (CS) is a method to measure and reconstruct signals from highly incomplete information. CS can be applied to parallel MRI to achieve great reductions in the scanning time. The performance of CS reconstruction depends on the randomness of k-space sampling. Here we consider a true random 2D under-sampling of k-space. By simulations based on real brain MRI data using 2-12 fold imaging time reduction, we show an 8-12% reduction in the reconstruction error compared to the constricted sampling patterns suggested earlier. In practice, the method can be implemented with 3D sequences by randomizing the phase encoding gradients.