Image reconstruction for echo planar imaging with nonequidistant k-space sampling

Echo planar imaging is characterized by scanning the 2D k-space after a single excitation. Different sampling patterns have been proposed. A technically feasible method uses a sinusoidal readout gradient resulting is measured data that does not sample k-space in an equidistant manner. In order to employ a conventional 2D-FFT image reconstruction, the data have to be converted to a cartesian grid. This can be done either by interpolation or alternatively by a generalized transformation. Filtering methods are described to minimize ghosting artifact that is typical in echo planar imaging. Results both from computer simulation and from experiments will be presented. Experimental images were obtained using a 2-T whole-body research system.