Compressed Sensing in Parallel Imaging: Towards Optimal Sampling

Compressed sensing (CS) is a novel method to measure and reconstruct N-dimensional compressible signals from M << N linear-combination (e.g. Fourier-component) samples [1,2]. CS has been applied to brain MRI to achieve acceleration factors R = N/M of 2–3 with only modest degradation in image quality [3]. A further reduction in the scan time can be achieved by multi-coil parallel MRI (pMRI) [4]. The two techniques have been recently combined to achieve high acceleration factors [5]. However, the performance of CS reconstruction relies on incoherent (e.g. random) sampling of the k-space. Here we show that a truly random k-space sampling in the context of pMRI allows an 8-12% reduction in the reconstruction error compared to the simple sampling patterns proposed earlier.