Research On Key Techniques Of Medical Image Fusion Based On Multiresolution Analysis

Medical image fusion (MIF) is a process of combining information from different medical images to generate a resultant image which can contain vital information presented in all source images. This technology can make full use of phenomena that lesion information from different imaging modalities are both redundancy and complementarity so as to obtain a more complete?accurate description of the same object. With the development and improvement of multiresolution analysis theory, multiscale transform based MIF is generally regarded as the more ideal method. However, the image quality composited by such method depends largely on the functionality of multiscale transform tool which is used for image decomposition. Recently, two multiscale transformation tools with two shift-invariant versions, which are respectively called the nonsubsampled contourlet transform (NSCT) and the nonsubsampled shearlet transform (NSST), have more perfect representation for images and have been verified to be more suitable for image fusion. In this dissertation, we focus on multimodal MIF based on NSCT and NSST. The main research contents and contributions of this dissertation are summarized as follows:1. Detail-enhanced MIF method in NSCTdomainAiming at the problem that many traditional fusion methods cannot well extract and preserve details of source images, a novel MIF method based on NSCT is proposed in this dissertation, which extracts?preserves and enhances details of fused image by using following ways:? According to measurement based on visibility to calculate weight map of source images, the aim of which is to accomplish extraction and transfer of information from different source images;? Using this algorithm (content adaptive image detail enhancement) to enhance details of the decomposition subbands, the aim of which is to highlight the detail information of source images;? To adjust the brightness and clarity of the resultant image, the blended subbands is further enhanced by gain control. Experimental results demonstrate that the proposed method can effectively enhance the details of the fused image.2. MIF based on statistical dependencies of the NSCT coefficientsFor the problem that many existing fusion methods based on NSCT neglect the dependencies between subband coefficients, we propose a novel MIF scheme based on the statistical dependencies between coefficients in NSCT domain, in which the probability density function of the NSCT coefficients is concisely fitted using generalized Gaussian density (GGD), as well as the similarity of two subbands is accurately computed by Jensen Shannon divergence (JSD) of two GGDs and then the dependencies between coefficients are embedded to the fusion rule for high frequency coefficients. Experimental results demonstrate that the proposed method significantly outperforms the conventional NSCT based MIF approaches in both visual perception and have excellent performance for evaluation indices based on mutual information and information entropy.3. MIF based on dual-channel unit-linking PCNN model in NSST domainTo make up for these defects that there are too many uncertain parameter lack adaptability and high complexity in standard PCNN model, we present a novel MIF algorithm based on the adaptive dual-channel unit-linking pulse coupled neural network (PCNN) in NSST domain. The proposed method uses a simplified adaptive dual-channel PCNN model and combines the characteristics of flexible multiscale and directional expansion for images in NSST domain with global coupling and pulse synchronization characteristic of dual-channel PCNN. Compared with other typical PCNN models, the proposed model has fewer parameters and better adaptability. For our algorithm, the constrast of image clarity in rogion is used as the linking strength (3 automatically and the time matrix T is utilized to determine the number of iteration. Thus, it effectively improves the adaptability of the algorithm and reduces computational complexity. Experimental results show that the proposed algorithm has excellent image fusion performance, and exceeds other MIF methods based on PCNN in both visual effect and multiple objective evaluation indexes.4.3D MIF based on 3D Shearlet TransformTo solve this problem that the third dimension information is lost when the traditional MIF algorithms are used for 3D medical volumetric data fusion, we present a novel 3D MIF method based on 3D shearlet transform. Condsering the internal structure characteristics of human organs and tissues, we define feature level fusion rule based on the physical characteristics of voxels. The implementation of the algorithm first uses structure tensor to extract structure information of the corresponding voxels for all high frequency coefficients, and then carries out analysis of the matrix rank. Lastly, the novel fusion rules are determined by the structural similarity of the corresponding voxels of source images. Experimental results indicate that the proposed algorithm can well merge 3D medical images, with a good clinical value.In summary, this dissertation takes efforts on hotspot and difficulty of the current MIF field. Four MIF algorithms are proposed and a series of experiment results validate that these methods can achieve good performance.