Research On CT/MRI Medical Images Fusion Algorithm Based On Deconvolution Neural Network Adaptive Transformation

Image fusion combines images from multiple sensors to finally obtain a more complete image.Image fusion technology is mainly used in clinical diagnosis in the medical field.By combining multi-modal images of lesions,it is helpful for doctors to locate lesions more accurately and improve diagnosis accuracy.The CT/MRI images fusion algorithm based on deconvolutional neural network adaptive transformation is proposed by this thesis.The specific research content is as follows:An image adaptive transformation method based on deconvolution neural network is proposed.In order to obtain the best transformation of the image to be fused,a Gaussian directional filter is used to design filter combinations with different cut-off frequencies and different directions as the initial filter of the deconvolutional neural network.In order to avoid the phenomenon of insufficient image feature extraction due to missing angles,when designing the directional filter combination,it is necessary to fully consider the combination of multiple directional filters at different angles to achieve 360 omni-directional coverage.According to the network learning mechanism that minimizes the error between the reconstructed image output by the deconvolutional neural network and the source image,the filter is continuously adjusted and optimized,so that the network learning performance is continuously improved,and finally,the optimal directional filter combination that minimizes the error between the reconstructed image and the source image is obtained,and the best decomposed high and low frequency feature maps are obtained.The above-mentioned image adaptive transformation method is used in the fusion algorithm research of the subsequent chapters.The experimental results show that the image adaptive transformation method based on Gaussian directional filter design proposed in this thesis can obtain better fusion image quality than wavelet transform.A CT/MRI image fusion algorithm based on image adaptive transformation and edge excitation DCPCNN is proposed.First,the image adaptive transformation method based on deconvolution neural network is carried out.By combining a Gaussian low-pass filter and 4Gaussian high-pass filters in different directions as the filter of the initial state of the deconvolution neural network,a single set of CT and MRI images are input into the network for adaptive transformation,and the high and low frequency feature maps of the corresponding number and corresponding direction are obtained.Based on the characteristics of the low-frequency feature map,a fusion rule with a large absolute value is designed for fusion processing.The high-frequency feature maps are fused by a dual-channel pulse-coupled neural network based on edge detection operators.The addition of edge detection operators can further improve the preservation of contour details.Comparing the image fusion algorithm based on image adaptive transformation and edge excitation DCPCNN proposed in this thesis with the existing image fusion algorithm based on neural network and guided filtering,the results show that the fusion image obtained by the algorithm in this thesis has higher definition,more detailed information,and better average gradient and edge intensity values.A CT/MRI image fusion algorithm based on image adaptive transformation and multifeature combination is proposed.First,the image adaptive transformation method based on deconvolution neural network is carried out.By combining a Gaussian low-pass filter and 8Gaussian high-pass filters in different directions as the filter of the initial state of the deconvolution neural network,a single set of CT and MRI images are input to the network for adaptive transformation,and the high and low frequency feature maps of the corresponding number and corresponding direction are obtained.The low-frequency feature maps are fused by using guided filter fusion rules,and the particle swarm optimization algorithm is used to find the optimal guide filter window radius to reduce the time complexity of the algorithm.In order to effectively solve the problem of the loss of detailed information of the fused image due to insufficient consideration of the image area contrast during the fusion process,a multi-feature combination fusion rule based on improved Laplacian energy combined with regional deviation coefficient is proposed in the design of the fusion rule.Comparing the fusion algorithm based on image adaptive transformation and multi-feature combination proposed in this thesis with the existing image fusion algorithm based on neural network and guided filtering,the results show that the fusion image quality obtained by the algorithm in this thesis is better,and the average gradient and edge intensity value are also better than comparison algorithm.The research results of this thesis have important application value in disease diagnosis.