Font Size: a A A

Research On The Multimodality Medical Image Registation And Fusion Technique

Posted on:2010-07-28Degree:DoctorType:Dissertation
Country:ChinaCandidate:W GeFull Text:PDF
GTID:1228330371450190Subject:Pattern Recognition and Intelligent Systems
Abstract/Summary:PDF Full Text Request
Since the image formation mechanisms and the applied environments of medical image instruments are different in the medical domain, different mode medical images reflect human body information from different aspects. Only one source image can not perform overall diagonose for a patient. Image registration and fusion technique can fuse the multimodality image information to a new image, make various image formation instruments attain the purpose for complement of their merits on the demonstration of image information. At present, the fusion technique of medical image is just on the start step. There are still a lot of problems and questions which are worth to explore and study. Therefore, a research on the multimodality medical image fusion method and registration method is performed, and a series of new algorithms is proposed.(1) For reasons that the registration method of the traditional mutual information spends too long time in the registration process and a certain degree mis-alignment is in existence, having combined both of the image registration method based on the mutual information and the image registration mehod based on the characteristics, a mutual information registration method based on edge characteristic points is proposed. This new registration method increases the registration speed greatly. Meanwhile, in order to give consideration to accuracy and reduce the mis-alignment phenomenon in the registration process, the normalized mutual information and improved Powell algorithm are adopted to perform the optimized search. Therefore, as compared with the traditional mutual information image registration method, this method is more stable and accurate.(2) According to the characteristic of multimodality medical image, a medical image fusion method based on separable wavelet transform and strengthened edge protection is proposed. This fusion method is possessed of stronger generalization ability without setting threshold based on the human’s experiences. Meantime, this fusion image obtained by the algnorithm can effectively reflect the profile detail information of the image, and be possessed of the stronger visual demonstrated ability.(3) For reasons that the partial edge loss and fuzzy texture information problems are in existence in the image fusion process of separable wavelet, a medical image fusion method based on non-separable wavelet transform is proposed. To compare with the separable wavelet, the non-separable wavelet is possessed of the merits for translational invariance, better frequency charesteristics and oriented property. Under the non-separable wavelet decomposition, using the HVS theory and considering the different characteristics of the medical images, two fusion strategys suitable to different conditions are proposed:a fusion rule based on regional information entropy and regional brightness and details priority for the clear disease focus region of the medical image; and another fusion rule based on local fuzzy entropy and regional brightness and details priority for the fuzzy diease focus region.(4) For reasons that the medical image fusion method based on wavelet transform has the limitation for different conditions adopting different fusion methods, a medical image fusion algorithm based on lifting scheme and contrast adaptive link strength PCNN is proposed, fully using the merits of PCNN and wavelet lifting respectively. Based on the wavelet lifting transform, this method adopts PCNN of which the link strength is local contrast, fuses organically the complement contents between the source images, increases the image clarity, simulates the process mode of human eye vision better, improves visual effect, and provides a strong support for clinical doctor to raise the diagnostic efficiency and reliability.(5) In the view of the limitation for pixel-level medical image fusion which requires the high registration accuracy, using the merit of BP nerve network, a fusion method of characteristic-level medical image based on BP nerve network is proposed. Experimental results show that this fusion method can also obtain better fusion effect even if the registration accuracy is not high.
Keywords/Search Tags:medical image, image registration, image fusion, non-separable wavelet transform, lifting scheme, mathematic morphology, grade of membership, information entropy, fuzzy entropy, pulse coupling nerve network, BP nerve network
PDF Full Text Request
Related items