Font Size: a A A

Medical Image Registration And Fusion Method Based On Multi-Scale Analysis

Posted on:2007-05-26Degree:DoctorType:Dissertation
Country:ChinaCandidate:S C PengFull Text:PDF
GTID:1118360242961495Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
With the development of imaging technology, medical image processing, as an indispensable part of the modern medical treatment, has been used widely in clinic medicine. In order to provide more comprehensive information with multi-modality medical image, the valid information needs to be integrated, because of the complementarities in multi-modality image information. The characteristics of image depend on the scale space where they inspect in multi-scale provides an effective description manner for image or the characteristics of image. It becomes an important research task how to integrate and analysis the different resolution image data. With the analysis of multi-scale, the dissertation studied relevant key technology in image registration and fusion, including three aspects of medical image processing: preprocessing (enhancement, interpolation), registration, fusion and segmentation. The following work has been completed:1. The dissertation had conducted in a comprehensive review of the significance, tasks, methods, applications and current difficulties in the medical image processing, and carried the necessity and importance of image fusion. A new imaging decompose method was carried out based on filter, and the improved methods ware discussed. A new image decomposition method based on multi-scale filter had been proposed, which has been used in the image enhancement. The experiment result validated this method.2. A new algorithm for image interpolation based on the wavelet comparability analysis was carried out. The high frequency weights of the interpolation result ware fetched through the analysis the comparability of the border upon two layers. This method used different homomorphic filter parameter for different decomposed image layer, and fused the enhancement result of each image layer to get the final result. The experiment validated the advantage of this method.3. A new image processing method, Empirical mode decomposition method, was introduced and the realization process was discussed. Empirical mode decomposition is a relatively new tool of signal processing. Compared with the Fourier transform and wavelet transform, it has better filtering capacity and more accurate data analysis capacity. It can deal with short-term nonlinear and non-stationary signals adaptively. Here we will study empirical mode decomposition process of image, and explore issues such as border pollution and extreme point search from the perspective of application, and then apply EMD to medical image registration. The registration idea here is to resist each layer with maximum mutual information method in accordance with the sequence of coarse-to-fine components of empirical mode decomposition, which reduces the number of the same name and to some extent improves the accuracy of registration. A new Image interpolation was carried out based on EMD; meanwhile the Image registration was carried out based on the EMD image layers. The result of experiment using the image registration of MRI and PET validated the method.4. A medical image fusion was carried out with wavelet analysis based on wavelet transforms and direction gradient. Taking into account the wavelet decomposition in different directions with different characteristics, the fusion result can be obtained by integrating the anisotropy characteristics of different components in different directions. This method fully embodies the directional characteristics of the image.5. The medical image segmentation methods was introduced, and the conception of multi-scale local character was fetched out. The method of image fusion based on segmentation was to segment image and to extract interest regions, and then use corresponding fusion strategy to obtain the fusion result by integrating the internal and the external of interest regions. The key issue of the method was how this method gives accurate segmentation results. Here from the perspective of multi-scale, local features of medical images was constructed, furthermore, a regional medical image segmentation method was given which was based on multi-scale feature, ultimately this image segmentation method was used in medical image fusion in interest regions.
Keywords/Search Tags:Multi-Modality Medical Images, Image Registration, Image Fusion, Multi-Scale, Similarity Analysis, Empirical Mode Decomposition, Orientation Gradient, Local Feature Segmentation
PDF Full Text Request
Related items