Research On MRI Image Reconstruction Based On Compressed Sensing And Sparse Dictionary

In recent years,MRI imaging technology,as an important medical diagnosis,has been widely used because of its lack of radiation damage to the human body,and has been highly accepted by the society.However,as the time of K spatial data acquisition is too long in MRI imaging,it is generally used only for the diagnosis of MRI imaging in the part of the patient's body.Aiming at this problem,this paper optimized the MRI imaging based on the compressed sensing theory,mainly studied the mathematical model and reconstruction algorithm of MRI image reconstruction.The main contents include:(1)To discuss the basic working principles and processes of compressed sensing,and focus on the in-depth analysis of dictionary learning K-SVD algorithm and reconstruction optimization algorithm OMP algorithm.(2)To compare the MRI image reconstruction under the framework of compressed sensing model-fixed sparse dictionary algorithm model and dictionary learning algorithm model for the performance analysis,and summarize the advantages and disadvantages of the two models.(3)Aiming at the characteristics of MRI image,the algorithm flow of DL-MRI was improved.Before dictionary learning,the image segmentation technique was introduced and the fuzzy c-means clustering algorithm was used to increase the image sparsity.The experimental results showed that the PSNR of MRI image reconstructed by MRI image compression combined with image segmentation algorithm was 10%~20% higher than that of traditional dictionary learning algorithm.(4)For the noisy signal acquisition conditions in reality,the mathematical model of the algorithm was improved by using the image segmentation technology.In order to keep the edge information of the image while reducing the noise,the total variational penalty term was added in the mathematical model,and the OMP algorithm was replaced by the alternating direction multiplier ADMM algorithm that can more easily deal with the optimization problem with the penalty term.Experimental results under different noise conditions showed that the quality of the reconstructed image was much better than that of the MRI image reconstruction algorithm based on dictionary learning.The peak signal-tonoise ratio of the reconstructed image could be improved by 5% to 20%.