A measure of image alignment is computed directly from the voxel intensities and an optimization process used to search for the transformation that maximizes this measure. Although the number of computations required is high, modern computing power means that reasonably high-resolution image volumes can be registered sufficiently quickly to be useful. The successful methods can be fully automatic and recent validation studies have shown that for the particular case of registration of images of the head, voxel similarity-based registration methods can outperform feature-based methods in terms of accuracy and robustness .
As the methods operate directly on voxel intensities, different methods are required for the alignment images from the same modality and that of images from different modalities. In intramodality image registration we would expect the difference in the images at alignment to be dominated by image noise and little structure should be present. This suggests a number of possible measures of alignment that are outlined in more detail below. As it is only meaningful to compute a measure where there is overlap of data and as the volume of overlap will change with each trial alignment, care must be taken to normalize the registration measures.
Was this article helpful?