Clinically Feasible Reconstruction Time for L 1-SPIRiT Parallel Imaging and Compressed Sensing MRI

Introduction: It has been shown that L1-SPIRiT [1], a method that combines parallel imaging and compressed sensing, has potential for reducing MRI scan times and improving image quality [2]. However, the clinical applicability of the technique is limited by the computational difficulty of the reconstruction. Reconstruction must be nearly instantaneous, but a straightforward implementation of L1SPIRiT’s non-linear minimization requires hours to complete. However, recent GPGPUs and multi-core CPUs have shown great potential for accelerating computations in many application areas, including in MRI reconstruction [3,4,5]. In this work, we demonstrate a fast implementation of L1-SPIRiT reconstruction that achieves clinical runtime; reconstruction of the 4-fold poisson disc