Research On Medical Image Registration Methods Based On Mutual Information

Medical image registration has been applied to the diagnosis of cardiac studies and a variety neurological disorders including brain tumors. Image registration is the process of aligning two images so that corresponding features of the inages can be easily related. Registration using different modalities, or geometric alignment of two-dimensional and three-dimensional image data, is becoming increasingly important in diagnosis, treatment planning, functional studies, computer-guided therapies, and in biomedical research. Registration based on intensity values usually requires optimization of some similarity function between the images.Maximization of mutual information(MI) of intensities is one of the most popular registration methods for multimodal medical image registration.This method measures the statistical dependence between the image intensities of corresponding voxels in two inages,this statiatical dependence is maximal when the images are totally aligned. Unfortunately, Local optimization techniques frequently fail because these functions with respect to transformation parameters are generally no convex and irregular and, therefore, global methods are often required.In this thesis we study firstly the image registration of 2D to 2D, we used respectively Particle Swarm Optimization algorithm (PSO),Powell method, Quantum-Behaved Particle Swarm Optimization (QPSO) and a hybrid algorithm combined by QPSO algorithm and Powell's method to solve the imgae registration of 2D to 2D and compare their results. Secondly, we study the multimodal medical image registration of 3D to 3D, we used QPSO and a hybrid algorithm combined by QPSO algorithm and Powell's method to solve the multimodal medical image registration of 3D to 3D and compare their results with the website results of the gold standard. We find that the performances (namely the result of search, speed of convergence, stability, and so on) of QPSO are more efficiently in the image registration of 2D to 2D and the multimodal medical image registration of 3D to 3D than Powell and PSO.This paper proposes a registration method based on wavelet representation. In this method the mutual information is used as the similarity measure and a hybrid algorithm combined by QPSO algorithm and Powell's method as the search technique. This method is applied to the 2D and 3D image registration. Experiments results shows that this image registration method could efficiently restrain local maxima of mutual information function and improve accuracy and speed. And it can achieve the subvoxel accuracy.This solution provides a realistic method in medical image registration to use in the diagnosis, treatment planning, functional studies, and so on.