Pairwise Intersubject Registration Methodologies for MR Brain Data

Abstract

Intersubject registration of the human brain is a difficult problem due to the high variability of intersubject anatomy. For this thesis, intersubject brain registration methods have been investigated through literature by critically analyzing and comparing 15 recent peer-reviewed journal papers. Additionally, a concise introduction to the concepts of intersubject brain registration methods is presented, as well as a short overview on each reviewed article. We have taken into account fully automated methods as well as landmark-based methods, using volume-based, surface-based and hybrid registration paradigms. We have seen that a wide variety of registration frameworks exist. While most methods are volume-based, we also see great recent interest in surface-based methods, which may align the intricate cortical surface more accurately; these are often used in functional studies. Furthermore, much focus in recent literature lies on topology preservation and invertibility. To this end, recent methods have been found to often constrain the transformation to be diffeomorphic, as this class of transformations have advantageous properties (e.g., topology preserving, non-zero Jacobian determinants). We found that due to both the nature of image registration and the large variety of methods that exist, the field of intersubject brain registration is in need of more thorough evaluation studies to gain a better understanding of the properties and relative performance of each method.