Workflow of the proposed strategy. Complex respiratory motion is modeled using voxel-based linear translations, which are estimated using 3D image registration after data binning using a respiratory surrogate signal. The voxel-based motion vector is represented by fewer approximations (N = 50) after k-means clustering for reduced computation. Linear k-space phase correction is performed for each motion vector and combined with reference bin for image reconstruction. The final output image is generated by pixel-based image fusion where the voxel value is selected from the motion corrected images (1..N) with the corresponding motion vector.