Research On MR Imaging Reconstruction Using Noise Removing Methods

Magnetic resonance imaging（MRI）is a commonly used medical imaging technology,because it has many advantages such as high tissue resolution,multi-sequence imaging,and no impact of ionizing radiation.However,magnetic resonance imaging has the disadvantages of long scanning time and slow imaging speed,which is particularly obvious in dynamic magnetic resonance imaging.Moreover,the reconstruction quality of MR images affects the doctor’s diagnosis of the patient’s lesion.Therefore,how to shorten the scanning time of MRI and improve the quality of MR image reconstruction is the focus of current scholars’research.Combining the theory of compressed sensing,this paper mainly studies the reconstruction algorithm of magnetic resonance image by using noise removal methods.The specific research contents are as follows:（1）An approximate message-passing magnetic resonance image reconstruction algorithm based on adaptive low-rank denoising method is proposed for MR image reconstruction.The algorithm uses the low-rank properties of the MR image to realize the denoising function in the approximate message passing algorithm,and adaptively designs the image block size and the number of similar blocks of the low rank denoising algorithm according to the estimated noise standard deviation.Specifically,using weighted Schatten-p norm minimization（WSNM）to achieve low rank denoising of MR images,which is used as a denoising algorithm for denoising-based approximate message passing algorithm,and finally achieve image’s reconstruction.Experimental results show that,under two different under-sampling masks,the proposed method can get higher peak signal-to-noise ratio and lower relative ₂L norm error,while being able to retain more image edge information.（2）A plug-and-play alternating direction method of multipliers（Plug-and-play ADMM）magnetic resonance image reconstruction framework is studied.The classic ADMM algorithm is designed as a magnetic resonance image reconstruction framework that can use the existing image denoising method as its reconstruction prior knowledge.On the one hand,this plug-and-play ADMM reconstruction framework has the advantages of fewer parameters and a more stable effect of the size of the parameter on reconstruction quality.On the other hand,the image denoising that is relatively mature today can be applied to the plug and play ADMM reconstruction,so this framework is universal.Since Blocking Matching of 3D filtering（BM3D）is a classic denoising method with superior denoising performance,a plug and play ADMM reconstruction algorithm（BM3D Plug-and-play ADMM,BPA）based on BM3D denoising method is proposed to implement MR image reconstruction.The experimental results show that compared with the state-of-the-art MRI image reconstruction algorithms,the BPA algorithm has higher peak signal-to-noise ratio and structural similarity,and can recover more edge information and details of MRI images.In summary,two kinds of noise removal-based MR image reconstruction algorithms are proposed in this paper.Furthermore,different sampling masks,factors and corresponding parameters are discussed about the reconstruction performances by using the proposed two reconstruction algorithms.And these experimental results verified that the effectiveness of the proposed method for reconstructing highly under sampled MR image.