Diffeomorphic Image Registration with Cross - Correlation : Evaluating Automated Labeling

Avants et al.’s goal in writing this paper is to propose a new deformable registration method and compare it to existing methods using brain MRI data. The method they propose is called symmetric image normalization (SyN). The method is meant to achieve better registration by maximizing cross correlation within the space of diffeomorphic maps, and the authors provide the Euler-Lagrange equations necessary to achieve this. SyN is advantageous in that it guarantees identical results each time the same two images are registered, and it takes advantage of exact inverse transformations guaranteed by diffeomorphisms. This method is most unique by the fact that cross correlation has not been investigated in diffeomorphic registrations. Such a combination allows for the possibility of symmetrizing cross correlation Euler-Lagrange equations. The authors test their method against the elastic method and the ITK implementation of Thirion’s Demons algorithm.