文摘
Registration algorithms are usually based on measures that must be optimized, whose choice has a great influence in the final result. For diffusion tensor imaging, scalar measures cannot be directly applied, and therefore new cost functions must be defined, regarding the special features of these data. We present a new pointwise similarity measure, named diffusion type based (DTB), that considers the physical meaning of the diffusion tensor. Theoretically, it is proved that DTB corrects the drawbacks of previous analogous measures, as well as it provides a more realistic result than generic measures. These conclusions are corroborated by experiments, where registration algorithms are driven by DTB and other broadly used measures. It is shown both quantitatively and visually that DTB leads to better results, and is more robust in presence of noise.