Image registration is widely used in different areas. It plays an important role in medical image analysis, group analysis and statistical parametric mapping. For the medical image analysis, image registration is useful for diagnosis, treatment planning, treatment evaluation, and so on. All these applications are relied on a correct registration result to provide higher treatment quality, increase the precision of diagnosis, and reduce the workload of doctors. Thus, it is essential to improve the robustness and accuracy of image registration. According to the nature of the transformation, image registration can be categorized into two main classes: Rigid Registration and Non-rigid Registration. The objective of this thesis is to develop a novel learning-based dissimilarity meas...[ Read more ]

Image registration is widely used in different areas. It plays an important role in medical image analysis, group analysis and statistical parametric mapping. For the medical image analysis, image registration is useful for diagnosis, treatment planning, treatment evaluation, and so on. All these applications are relied on a correct registration result to provide higher treatment quality, increase the precision of diagnosis, and reduce the workload of doctors. Thus, it is essential to improve the robustness and accuracy of image registration. According to the nature of the transformation, image registration can be categorized into two main classes: Rigid Registration and Non-rigid Registration. The objective of this thesis is to develop a novel learning-based dissimilarity measure for both rigid and non-rigid medical image registrations. This novel measure utilizes Bhattacharyya distances to measure the dissimilarity of the testing image pairs by incorporating the expected intensity distributions (priori knowledge) which learned from the registered training image pairs. The proposed dissimilarity measure can be easily adopted to the existing framework of rigid image registration whereas it is not trivial to apply it into the existing framework of non-rigid image registration. Therefore, an approximation of the proposed dissimilarity measure is also derived in this thesis such that the proposed measure can be applied to the Markov Random Field (MRF) modeled non-rigid image registration approach. By the help of Bhattacharyya distances, the priori knowledge and the MRF modeled registration framework, the experimental results demonstrated that our novel learning-based dissimilarity measure can achieve higher robustness and accuracy, as compared with state-of-the-art approaches, in both rigid and non-rigid image registrations.