Registration Method For Medical Images Based On Wavelet Transform And Mutual Information

Image registration is a hot topic in image processing area, which is widely used in remote sensing image processing, computer vision, motion estimating and medical image analysis. It is the premise and key of image fusion. Medical image registration developed from 1990s is one of important research area of medical image processing, which is widely applied in clinic diagnosis and treatment.Mutual information based medical image registration comes from information theory, which is the development trend of medical image registration due to its virtues of no segmenting, high precision and solidity. In this paper, application of the method is discussed in medical image registration in detail.Wavelet transform developed from 1990s has been an important research direction in signal processing area with its unique advantages in non-stationary signal processing aspects, which plays more and more important role in actual application. According to wavelet transform and multi-resolution analysis theory, the paper discusses medical image registration method based on wavelet transform and mutual information. Due to the excellent characteristics of dual-tree complex wavelet transform, it is introduced to medical image registration and proved to be effective in practice. The main work and contents of the dissertation include:(1) The basic concepts,application prospects,development status and registration process of medical image registration are introduced, and main registration methods,classification,registration assessment methods and research status are systematically induced and concluded.(2) The basic theory,registration principle and basic steps are deeply described, involving several interpolation algorithms and optimization algorithms. Calculating mutual information or normalized mutual information by the methods of non-use optimization algorithms and Powell or PSO optimization algorithms is discussed and the effectiveness of two optimization algorithms is analyzed in this paper, which stresses the local searching characteristics of Powell. The two algorithms are succeeded in the brain CT/MRI image registration and demonstrate good results in experiment.(3) Wavelet transform and multi-resolution analysis theory are discussed in detail. It puts emphasis on discussing medical image registration method based on wavelet transform and mutual information. In order to improve registration precision and robustness, using Powell algorithm two experiments have been done in small scale and large scale image registration. Compared to registration method based on mutual information and wavelet transform and non-use wavelet transform in each experiment, the results of experimental data analysis prove this method can effectively overcome local extremum and have the advantage of high precision. Meanwhile, the experiment shows that the method can be used for general CT, MRI image registration, also suitable for ultrasonic image registration.(4) The principle, characteristics of dual-tree complex wavelet transform, and the features of dual-tree complex wavelet decomposition image are described in detail. In view of the approximate translation invariance, the smaller computation, and other fine features, we discuss the medical image registration based on dual-tree complex wavelet transform and mutual information. By using Powell optimization registration of the two groups of medical images, that is, small scale images and large scale images, the method and wavelet transform registration method are compared in the experiment. The results show that this method has kept good precision and can effectively accomplish image registration, is a stable and effective method.