On density compensation in Bayesian k-space trajectory optimization


In the field of Compressive Sensing (CS) MRI reconstruction, a lot of research has gone into improving reconstruction techniques. Even though many authors initially used (uniformly) randomized sampling patterns, it was experimentally found, and proven in [1], that variable density trajectories have vastly better performance. From a Bayesian viewpoint of CS MRI, an inventive technique was recently proposed in [2] to optimally construct a k-space sampling pattern from a set of different k-space segments, using variational Bayesian approximation. In this work, we will first demonstrate how different similarly subsampled k-space trajectories give rise to vastly different reconstruction results. Then, we show that using a trajectory optimization technique, such as the one in [2], to optimize variable density trajectories requires special care. More specifically, we will show that correctly adjusting for the non-uniform sampling density, using a proper density compensation function (DCF) is important and easily overlooked.

20th Annual Scientific Meeting & Exhibition, Abstracts