Information-Theoretic Registration with Explicit Reorientation of Diffusion-Weighted Images

We present an information-theoretic approach to the registration of images with directional information, especially for diffusion-weighted (DWIs), explicit optimization over scale. call it locally orderless directions (LORDs). focus on normalized mutual information as a robust similarity measure DWI. The framework is extension LOR-DWI density-based hierarchical scale-space model that varies and optimizes integration, spatial, intensity scales. As affine transformations are insufficient inter-subject registration, we extend nonrigid deformations. illustrate proposed deforms orientation distribution functions (ODFs) correctly capable handling classic complex challenges in DWI registrations, such fiber crossings along kissing, fanning, interleaving fibers. Our experimental results clearly novel promising regularizing effect, which comes from nonlinear orientation-based cost function. show properties different image scales, including orientational our makes better at retrieving deformations contrast standard scalar-based registration.