New similarity measures and deformation optimization comparisons for medical image registration

Image registration is widely used in different areas, including medical image analysis and image processing. In this thesis, we introduce a new similarity function for image registration and compare two optimization methods of Markov Random Field (MRF) based non-rigid image registration. The novel similarity function is based on a priori knowledge of the joint intensity distribution of a pre-aligned image pair. We have evaluated the proposed similarity measures with 3600 randomized rigid registration experiments on CT-T1 brain image pairs from the Retrospective Image Registration Evaluation (RIRE) project. The results show that the proposed similarity measures make significant improvements to the registration accuracies and success rates as compared with the mutual information (MI) based method. We have also tested the similarity measures and their derivatives on seven multi-modal non-rigid image registration experiments under Free-Form Deformation (FFD) registration framework and compared the results obtained by using the MI based FFD and the conventional KLD based FFD. The experimental results demonstrate that our method make remarkable improvements to the registration accuracy. In addition, we have compared two optimizations, graph cut and linear programming, of MRF based non-rigid registration. The experimental results show that graph cut is slower than linear programming but it can provide higher accuracy and use less memory