Spatiotemporal maps for dynamic MRI reconstruction: a proof-of-principle demonstration on single-coil animal gastrointestinal data

Motivation: Partially separable functions (PSF) are commonly used in dynamic MRI to model the signal in the (x,t)-space; however, the number of components in the PSF expansion can be restrictive in applications where voxels present different temporal/spectral characteristics (e.g., gastrointestinal MRI reconstruction). Goal(s): To propose a parsimonious representation for the dynamic MRI signal, called spatiotemporal maps (STM). Approach: STM are derived by proving that shift-invariant linear predictability relationships can exist across the k-spaces of multiple time-frames. Results: STM provide a parsimonious representation for the spatiotemporal MRI signal; they can be efficiently calculated from autocalibration data; and they can be synergistically combined with modern regularizers. Impact: Spatiotemporal maps provide a parsimonious voxel-dependent expansion for the dynamic MRI signal, even when voxels present various temporal/spectral characteristics. They can be efficiently calculated from autocalibration data, and can be synergistically combined with modern regularizers to reconstruct highly accelerated data.