Multimodal Medical Image Fusion Based On Non-negative Matrix Factorization

With the development of medical imaging,CT,MRI and PET have become indispensable tools for clinical diagnosis.Different medical images provide convenience for doctors’ diagnosis and facilitate the periodic review of patients.However,imaging is different due to the different imaging mechanisms.If CT focuses on bone imaging and MRI focuses on soft tissue,medical multimodal image fusion has attracted more and more attention.Similar to image fusion,medical multimodal image fusion is divided into three steps.First,the source image is decomposed in multi-scale and multi-direction;secondly,fusion rules are adopted for fusion;finally,the final fusion image is obtained by inverse operation.Combining different modality medical images into one image is of great significance for clinical diagnosis.There are many fusion algorithms at the moment and the article focuses on the following aspects:(1)For the problem of high complexity in non-negative matrix factorization,an improved image fusion algorithm based on weighted non-negative matrix factorization and dual-channel pulse-coupled neural network is proposed.For image low-frequency sub-bands,an improved weighted non-negative matrix factorization algorithm is used to dynamically update the weight matrix to better extract image feature information.For high-frequency sub-bands,an improved dual-channel pulse coupled neural network algorithm is used to improve link strength.The value is the gradient value of the block proposed in the article to better preserve the tiny details of the image.Experiments show that combining the weighted non-negative matrix factorization with the dual-channel pulse-coupled neural network can not only extract the image feature information well,but also retain more detailed information.At the same time,the dual-channel pulse coupled neural network method can improve the algorithm operation.effectiveness.(2)Aiming at the problem of many parameters and uncertain numerical values of the current pulse coupled neural network(PCNN)model,a simplified PCNN model is proposed.For image low-frequency subbands,fusion rules of edge energy and gradient energy sum are used to improve the edge direction sensitivity of the image and extract image feature information.For the high-frequency subbands of the image,the simplified PCNN model presented in this paper has less parameters of the model.At the same time,the simulated annealing algorithm is used to find the optimal value of the parameter,and the link input value adopts the visibility value.It has been confirmed through experiments that the simplified PCNN model can improve the running speed of the algorithm.(3)A multimodal medical image fusion system is implemented.The system consists of three modules: image preprocessing,image fusion,and objective evaluation.The image fusion uses the improved weighted non-negative matrix decomposition algorithm and the simplified PCNN fusion algorithm.The evaluation of the image fusion results proves the effectiveness and significance of the proposed algorithm.