Brain Warping Via Landmark Points and Curves with a Level Set Representation
Andrew Y. Wang, Alex Leow, Hillary D. Protas, Arthur W. Toga, Paul M. Thompson · 2004
Abstract. This paper presents and validates a non-linear image registration method driven by points and curved landmarks using implicit representation. This approach produces smooth one-to-one mappings between topologically equivalent images by constraining the transformations to adhere to continuum mechanical laws. In this paper, the elastic operator is used for fast computation when only small deformation is needed. For large deformation, the same strategy is coupled with the method of infinite dimensional group actions to generate highly non-linear diffeomorphic maps. We applied this method to register brain magnetic resonance images in a flattened parameter space, and visualize sulcal variability by pulling back the mapping to 3D. Results show accurate registration of MRI images using delineated sulcal landmarks, while relaxing the registration field along the sulcal lines. I.