Research On Multi-modal Medical Image Fusion Based On Edge-preserve Filter And Saliency

Multi-modal medical image fusion represents a process of merging multiple medical images containing different feature information into one image by some means.Compared with single modal images,fused images can more fully reflect the information description of tissues and organs,and provide more abundant auxiliary medical information for clinical practice.The existing multi-modal medical fusion research includes two modes: fusion of gray images and gray images(such as MRI-CT)and fusion of gray images and pseudo color images(such as MRI-PET,MRI-SPECT).MRI-CT fusion can obtain a fused image with clear organ contour information and soft tissue information which suitable for diagnosis and treatment of tumors.The fusion of MRI-PET and MRI-SPECT can provide a fused image with clear soft tissue information and rich color information,which are suitable for cancer diagnosis and treatment.In this thesis,MRI and CT,PET,and SPECT fusion are the research objects,and two effective fusion methods are proposed for the problems of principal component loss(structure,color)and fused image blur in existing fusion methods.Aiming at the problem of main component(structure and color)loss in existing MRI-PET and MRI-CT fusion,this thesis proposes a multi-modal image fusion method based on guided filtering and graph-based visual saliency(GBVS).First,the source images are decomposed into multi-scale rough and detailed sub-band images by using guided filtering algorithm.Then,the information entropy and the graph-based saliency detection algorithm are used to construct the fusion rules of the rough and detailed sub-band images respectively,and the fused sub-band images are obtained.Finally,the fused sub-band images are reconstructed by summation algorithm to generate the final fused image.Experimental comparison analysis shows that the proposed method has the advantages of more complete structural information retention and less color loss.Aiming at the problem of fused image blur existing in MRI-PET and MRI-SPECT fusion,this thesis proposes a multi-modal medical image fusion method based on gradient domain guided filtering and multi-saliency.First,the source images are decomposed using a gradient-directed filtering algorithm to obtain sub-band images of multiple scales.Then,the spatial residual spectrum saliency detection algorithm and the graph theory saliency detection algorithm are used to construct the fusion rules fordifferent types of sub-band images,and the fused coefficients are combined by the generalized intensity-hue-saturation method.Finally,the fused sub-band images are reconstructed by summation algorithm to generate the final fused image.Experiment shows that compared with many classical and effective fusion methods,the fusion results of the proposed method are clearer in the display of texture and edge information.