ABSTRACT
As reviewed by Grevera et al. (see (Grevera and Udupa 1998)) and Lehmann et al. (see (Lehmann, Göonner, and Spitzer 1999)), the scene-based method is simple, intuitionistic and easy to use, but the accuracy is usually poor. The object-based method has better accuracy generally, but the computational complexity is higher. In some object-basedmethods, flexible alignment is used to align adjacent slices, and then image interpolation is carried out between the corresponding positions in each slice (see (Goshtasby, Turner, and Ackerman 1992; Williams and Barrett 1993)). These methods, including using the optical flow model, elastic model, fluid model, finite element model and B-spline based free-form deformation, become popular in the 1990s (see (Penney, Schnabel, Rueckert, Hawkes, and Niessen 2004; Penney, Schnabel, Rueckert, Viergever, and Niessen 2004; Rueckert, Sonoda, Hayes, Hill, Leach, and Hawkes 1999; Williams and Barrett 1993)). The success of the alignment-based image interpolation depends on two prerequisites (see (Penney, Schnabel, Rueckert, Hawkes, and Niessen 2004)): (1) the adjacent slices contain similar anatomical features, and (2) the alignment algorithm is capable of finding the transformation whichmaps these similar features correctly. If the
first prerequisite is violated, and an anatomical feature disappears from one slice to the next, then the advantages of the alignment-based approachwill be lost.The second prerequisite is concerned with the transformation types that the alignment algorithm is capable of. If the transformation between features in adjacent slices is beyond the capabilities of the alignment algorithm, then the results will be sub-optimum.
