Regularization Parameter Control Strategy Of CS-MRI Image Reconstruction Algorithm

Magnetic resonance imaging(MRI),as an integral part of modern medical imaging technology.The main drawback is that the signal acquisition time is too long.To this end,many scholars are committed to the research of MRI rapid imaging.Compressed sensing(CS)theory is a new signal acquisition and processing theory born in the field of signal processing in recent years.The theory of compressed sensing is applied to magnetic resonance imaging technology.A priori information and partial K-space data can be effectively utilized,and the magnetic resonance image can be accurately reconstructed by a nonlinear reconstruction algorithm.In the reconstruction algorithm,the selection of regularization model and the fixed step size are important factors affecting the accuracy of the algorithm.In view of the above deficiencies,this thesis improves the existing regularization model based on the application of compressed sensing theory and proposes A regularized parameter control strategy that corrects regularization parameters during algorithm iteration to accurately reconstruct high-quality MRI images.The alternating direction method of multipliers(ADMM)algorithm is essentially developed from the Augmented Lagrange Method.It combines the advantages of the multiplier method and the alternating minimization algorithm.Effectively solve the convex optimization problem with separable structure.In this thesis,based on the ADMM with Total Variation as a regular term,a Nonlocal total variation(NLTV)model is developed.Compared with the traditional method,NLTV uses the distance between image blocks instead of neighboring pixels to measure the structural similarity of the image,so it has a better protection for the details of the image.In the traditional nonlinear reconstruction algorithm,the value of the regularization coefficient is a fixed value,which plays the role of balancing the regular term.The energy difference of the reconstructed signals in the adjacent two stages during the accurate reconstruction of the image is gradually small.It shows that the energy difference of the reconstructed images of two adjacent stages drops rapidly at the beginning stage,then decreases slowly and finally stabilizes within a certain range.In this thesis,based on the variation law of reconstructed images energy difference in adjacent iteration stages,a regularized parameter control strategy for correcting regularized parameters(MRP)is proposed.In the iterative process of the compressed perceptual magnetic resonance reconstruction algorithm,a regularized parameter with a larger value is used to smooth the noise,and a regularized parameter with a smaller value is used to recover the edge detail information.This balances the regular and data items.The experimental results show that under the same measurement data and experimental conditions,The reconstruction algorithm uses the control strategy proposed in this thesis to reconstruct the quality of the image,which is improved,regardless of the subjective and objective evaluation indicators.