As stated in the introduction, voxel similarity measures have an intuitive basis in intramodality registration. The assumption is that the images are very similar, i.e., that after calculation of the geometric transformation required to align the structures in the patient being studied, the vast majority of voxels in the registered images will differ only by noise. The simplest voxel similarity measure is the sum of squared intensity differences between images, SSD, which is minimized during registration. For N overlapping voxels i in images A and B:
Was this article helpful?