ABSORB: Atlas Building by Self-Organized Registration and Bundling

Abstract

A novel groupwise registration framework, called Atlas Building by Self-Organized Registration and Bundling (ABSORB), is proposed in this paper. In this framework, the global structure of relative subject image distribution is preserved during the registration by constraining each subject to deform locally within the learned manifold. A self-organized registration is employed to deform each subject towards a subset of its neighbors that are closer to the global center. Some subjects close enough in the manifold will be bundled into a subgroup during the registration, and then deformed together in the subsequent registration process. This framework performs groupwise registration in a hierarchical way. Specifically, in the higher level, it will perform on a much smaller dataset formed by the representative subjects of all subgroups generated in the previous levels of registration. The atlas image can be eventually built once the registration arrives at the upmost level. Experimental results on both synthetic and real datasets show that the proposed framework can achieve substantial improvements, compared to the other two widely used groupwise methods, in terms of both registration accuracy and robustness.