Simultaneous Tensor and Fiber Registration (STFR) for diffusion tensor images of the brain

Abstract

Accurate registration of diffusion tensor imaging (DTI) data of the brain among different subjects facilitates automatic normalization of structural and neural connectivity information and helps quantify white matter fiber tract differences between normal and disease. Traditional DTI registration methods use either tensor information or orientation invariant features extracted from the tensors. Because tensors need to be re-oriented after warping, fibers extracted from the deformed DTI often suffer from discontinuity, indicating lack of fiber information preservation after registration. To remedy this problem and to improve the accuracy of DTI registration, in this paper, we introduce a simultaneous tensor and fiber registration (STFR) algorithm by matching both tensor and fiber tracts at each voxel and considering re-orientation with deformation simultaneously. Because there are multiple fiber tracts passing through each voxel, which may have different orientations such as fiber crossing, incorporating fiber information can preserve fiber information better than only using the tensor information. Additionally, fiber tracts also reflect the spatial neighborhood of each voxel. After implementing STFR, we compared the registration performance with the current state-of-the art tensor-based registration algorithm (called DTITK) using both simulated images and real images. The results showed that the proposed STFR algorithm evidently outperforms DTITK in terms of registration accuracy. Finally, using statistical parametric mapping (SPM) package, we illustrate that after normalizing the fractional anisotropy (FA) maps of both traditional developing (TD) and Autism spectrum disorder (ASD) subjects to a randomly selected template space, regions with significantly different FA highlighted by STFR are with less noise or false positive regions as compared with DTITK. STFR methodology can also be extended to high-angular-resolution diffusion imaging and Q-ball vector analysis. © 2013 Springer-Verlag.