Consistent projection reconstruction (CPR) techniques for MRI.

Motion of the subject causes inconsistencies in MR scan data which in turn produce streaks in projection reconstruction (PR) scans or ghosts in 2DFT imaging. It is shown that a data manifold consistent with a static object may be extracted from input scan data by applying consistency criteria based on moment or Fourier expansions of the object's projections. The criteria limit the spectrum of azimuthal variation relative to the radial moments or radial spectral components, and their application diminishes k-space inconsistencies from motion or flow by discarding unallowed expansion terms. This consistent projection reconstruction (CPR) algorithm was applied to the reconstruction of projection data from a moving phantom and from human abdominal sections, and in each case provided reduced motion streaks relative to conventional reconstructions. In addition, CPR was found to be effective when used with 2DFT data by resampling the Cartesian data as radial projections, but to a lesser extent.