Boundaries or surfaces between organs and between the skin surface and the surrounding air can provide strong features for image registration. Automated image segmentation algorithms - or, failing that, manual or user-defined segmentation - can be used to define visible boundaries in the images to be registered. If these boundaries correspond to the same physical surface in the images to be registered then they can be used to derive the registration transformation. A number of algorithms have been based on registration between surfaces and lines or between points demarcated on surfaces. These algorithms are usually only used to determine rigid body transformations. Many surfaces in images have a high degree of symmetry (for example, the outline of the cranial vault has an almost circular outline in sagittal section) and nonrigid transformation solutions are likely to be highly susceptible to noise and hence error prone.
Was this article helpful?