Should we use/apply this to a particular volumetric dataset while prototyping different methods, to ensure accurate comparisons? Should anisotropic, regularly sampled voxels be supported?