An image data processing method is for registering two four-dimensional image data sets, each representative of a time-series of three-dimensional image frames. The method comprises an initial pre-registration step in which 3D image frames of the two image data sets are rotated and translated (16) relative to one another so as to bring into alignment respective dominant motion vectors identified (14) for each, the dominant motion vector being a 3D motion vector representative of motion of an identified three-dimensional sub-region exhibiting maximal spatial displacement over course of the time series.