The simplest conceptual method for aligning two images is to move the source image interactively with respect to the target image. Combined with a means of displaying the two images so as to allow the quality of the resulting alignment to be judged, this can provide an effective alignment method. Pietrzyk et al.  have described a method for the alignment of PET and MR images of the head by interactively aligning contours derived from PET [18F]-FDG image with slices of MR images. While reasonably quick with the appropriate user interface and fast reslicing of PET contours, the method is prone to user error and has largely been superseded by more automated methods.
Interactive alignment remains very useful when aligning images that show gross abnormality, when attempting alignment in applications for which a specific algorithm is not designed, or when registering images where registration has failed for whatever reason. The skilled observer will bring into play expert knowledge of anatomy and radiological appearances that are not captured by even the most sophisticated computer algorithm. Care must, of course, be taken to verify that the accuracy of registration is sufficient for the application, as the purpose of registration is to extract useful and perhaps surprising information from aligned images.
Even when a fully automated algorithm is used, a certain amount of user interaction is desirable. Careful visual inspection of the registration results is always strongly recommended as no such algorithm will produce an accurate result one hundred percent of the time, and, similarly, methods for detecting failure are never one hundred percent reliable.
Was this article helpful?