Multimodality and Nonrigid Image Registration with Application to Diffusion Tensor Imaging

Multimodality and Nonrigid Image Registration with Application to Diffusion Tensor Imaging Mohammed Khader The great challenge in image registration is to devise computationally efficient algorithms for aligning images so that their details overlap accurately. The first problem addressed in this thesis is multimodality medical image registration, which we formulate as an optimization problem in the information-theoretic setting. We introduce a viable and practical image registration method by maximizing a generalized entropic dissimilarity measure using a modified simultaneous perturbation stochastic approximation algorithm. The feasibility of the proposed image registration approach is demonstrated through extensive experiments. The rest of the thesis is devoted to nonrigid medical image registration. We propose an informationtheoretic framework by optimizing a non-extensive entropic similarity measure using the quasi-Newton method as an optimization scheme and cubic B-splines for modeling the nonrigid deformation field between the fixed and moving 3D image pairs. To achieve a compromise between the nonrigid registration accuracy and the associated computational cost, we implement a three-level hierarchical multi-resolution approach in such a way that the image resolution is increased in a coarse to fine fashion. The feasibility and registration accuracy of the proposed method are demonstrated through experimental results on a 3Dmagnetic resonance data volume and also on clinically acquired 4D computed tomography image data sets. In the same vein, we extend our nonrigid registration approach to align diffusion tensor images for multiple components by enabling explicit optimization of tensor reorientation. Incorporating tensor reorientation in the registration algorithm is pivotal in wrapping diffusion tensor images. Experimental results on diffusion-tensor image registration indicate the feasibility of the proposed approach and a much better performance compared to the affine registration method based on mutual information, not only in terms of registration accuracy in the presence of geometric distortions but also in terms of robustness in the presence of Rician noise.