3-D Deformable Registration of Medical Images Using a Statistical Atlas

Abstract

Registration between voxel images of human anatomy enables cross-patient diagnosis and post-treatment analysis. However, innate variations in the shape, size, and density of non-pathological anatomical struc-tures between individuals make accurate registration difficult. Characterization of such normal but inherent variations provides guidance for registration.We extracted the pattern of normal variations in the appear-ances of brain structures from the T1-weighted magnetic resonance imaging (MRI) volumes of 105 sub-jects. This knowledge serves as a domain-relevant constraint which increases the accuracy of deformable registration. We represent domain knowledge in the form of voxel statistics, and embed these statistics into a 3-D digital brain atlas which we use as the reference. The knowledge acquisition process involves registering a training set of MRI volumes with the atlas. The method employed is a previously developed 3-D hierarchical deformable registration algorithm. This associates each voxel in the reference atlas with distributions of nor-mal variations of intensity and 3-D positions of the training set. We evaluate statistical properties of these distributions for each atlas voxel to build a statistical atlas which contains the anatomical information of the population. When we register this atlas to a particular subject, the embodied statistics function as domain-relevant constraints. The deformation process tolerates non-pathological variations between subjects. When applied to 40 test cases, this knowledge-constrained registration method achieved a correct voxel classifica-tion rate above 95% for 36 cases; this is a 24% improvement over the performance of the algorithm without knowledge constraints. To overcome imprecisions in unconstrained registration that affect the rigorousness of the statistical atlas, we propose to build an initial statistical model of a small but accurately registered training set, then boot-strap it into a more reliable model. Besides guiding deformable registration, our knowledge representation also enables quantitative investigation of possible anatomical divergences between populations.