Magnetic resonance imaging provides exceptional soft tissue contrast but it is limited by long scan times. Compressed sensing (CS) is a novel technique that leverages the underlying transform sparsity of medical images, randomized under-sampling of k-space, and nonlinear reconstruction to recover images from accelerated scans. However, in some cases image quality of prospectively implemented CS is not as good as predicted by retrospectively under-sampled data.
This dissertation investigates the source of decreased image quality in CS. Our findings demonstrate that random k-space trajectories are specifically susceptible to encoding errors between repetition periods. Sources of errors may be due to eddy currents or motion, and errors have implications for accelerated reconstructions. CS is sensitive to data inconsistencies from random trajectories as acceleration rates increase, but minimizing trajectory length in random under-sampled trajectories can reduce data inconsistencies. Solving these issues will improve image quality and help realize the potential of CS.