Meeting Banner
Abstract #2556

Fast L1-SPIRiT Compressed Sensing Parallel Imaging MRI: Scalable Parallel Implementation and Clinically Feasible Runtime

Mark Murphy1, Miki Lustig

1EECS UC Berkeley, Berkeley, CA, United States

We describe the algorithmic and parallelization implementation decisions that lead to fast runtimes in l1-SPIRiT combined Parallel Imaging and Compressive Sensing reconstruction.

Keywords

abstract achieve achieves achieving additionally advantage algorithmic alternate arbitrary arrays available batch better caching calibration channel channels clinical clinically coarse coil coils collaborators combined complexity compressed computations consider consistent continuously core cores crucial density deployed describe detailed differently disc distribution dual efficient encodes exploitation extends factorization fast feasible fine flight frequent gathered grained graphics hierarchical implementation improvements incur independently iteration iterative length lengths mark matrix minute modern modified operations overhead parallel parallelism parallelization parallelizations performance prefer preliminary previous processor processors proportional pulse readout reconstruction reconstructions reduces require requires respond runtime scalable scale self sensing separately sequential simultaneously slice slices socket solves sources space speedup spirit suit synchronization system take torso transform transforms variable wavelet year