Research On CT/MRI Medical Images Fusion Algorithm Based On Deconvolution Neural Network And Intuitionistic Fuzzy Reasoning

In the field of medical images,image fusion is an important medical auxiliary technology,which can process the image information comprehensively provided by different medical imaging devices and combine the information effectively,obtaining a more informative image.The CT/MRI images fusion algorithm based on deconvolutional neural network and intuitionistic fuzzy reasoning is discussed in the thesis.The specific contents are as follows:A CT/MRI image fusion algorithm based on multi-directional gradient intuitionistic fuzzy reasoning is discussed.It is adopted to obtain feature maps that can fully characterize the source images for a deconvolutional neural network based on Gaussian direction filters.According to the data to be fused,an adaptive transformation method of the image is designed,and the high-frequency and low-frequency feature maps are obtained with the smallest error between the reconstructed image and the source image.Based on the directionality characteristics of the high-frequency feature maps,multi-directional gradient is used for deeper intuitionistic fuzzy reasoning.The fusion rule based on multi-directional gradient intuitionistic fuzzy reasoning is designed,where the direction gradient can reflect a specific directional detail contrast,so that the details of the fusion image are clearer.The rule of intuitionistic fuzzy reasoning based on the average gradient is used to fuse low-frequency feature maps,where the average gradient is used for deeper intuitionistic fuzzy reasoning.The algorithm is compared with the fusion algorithm based on intuitionistic fuzzy index and NSCT and the fusion algorithm based on Contourlet and intuitionistic fuzzy reasoning.Experiment results show that the fusion image of the algorithm is clearer,and the average gradient and edge strength are higher.A CT/MRI image fusion algorithm based on multi-propositional intuitionistic fuzzy reasoning is discussed.A deconvolutional neural network based on Gaussian direction filters is employed.Based on the images to be fused,an adaptive transformation method of the image is designed.A Gaussian low-pass filter and four Gaussian high-pass filters in different directions are preset into this deconvolutional neural network,and after iterative training,the optimal high-frequency and low-frequency feature maps of the source image are obtained.The multi-fuzzy set is defined as multi-proposition in the thesis.The fusion rule based on multi-propositional intuitionistic fuzzy reasoning is designed for high-frequency and lowfrequency feature maps.The influence of the number of propositions on the fusion results is studied.The output of the fusion results corresponding to the best number of propositions is adopted,and the CT and MRI fusion images are obtained.The algorithm is compared with the image fusion algorithm based on wavelet transform and multi-proposition intuitionistic fuzzy reasoning and the fusion algorithm based on the intuitionistic fuzzy reasoning of Contourlet transform that introduces the average gradient.Experiment results show that the fusion image of the algorithm is clearer,the detail information is richer,and the average gradient and edge strength are higher.A CT/MRI image fusion algorithm based on adaptive membership intuitionistic fuzzy reasoning is discussed.A deconvolutional neural network based on Gaussian direction filters is employed.Based on the images to be fused,an adaptive transformation method of the image is designed.A Gaussian low-pass filter and four Gaussian high-pass filters in different directions are used instead of random filters preset into the deconvolutional neural network,and after iterative training,the high-frequency and low-frequency feature maps with the smallest error between the reconstructed image and the source image are obtained.The algorithm regards the construction of the membership function as an optimization problem,and the fusion rule based on adaptive membership intuitionistic fuzzy reasoning is designed.Membership,non-membership,hesitation and intuitionistic fuzzy entropy are defined by gradient values and SML values of high-frequency feature maps in different directions.The maximized intuitionistic fuzzy entropy is used to determine the parameters adaptively,and the intuitionistic fuzzy sets are constructed for fusing high-frequency feature maps.Based on the characteristics of the low-frequency feature map,a fusion rule with a large absolute value is applied to fusion processing.The algorithm is compared with the infrared and visible image fusion algorithm based on intuitionistic fuzzy set and the fusion algorithm based on image adaptive transformation and intuitionistic fuzzy set.Experiment results show that the detail information of the fusion image of the algorithm is richer,and the average gradient and edge strength are higher.The image fusion algorithms discussed in the thesis can obtain better quality medical fusion images,which is of great significance in medical diagnosis.