Parallel Magnetic Resonance Imaging Based On Dictionary Learning

Magnetic resonance imaging is a noninvasive and nonionizing imaging technique,which has a large number of imaging parameters and arbitrary azimuth imaging.It can not only display the morphological information such as anatomical structure,but also reflect the physiological and biochemical functions of human body,and it is an important tool in clinical application and scientific research.At present,there is a large demand for magnetic resonance market at home and abroad,and the domestic magnetic resonance application market is mainly occupied by foreign-funded enterprises,coupled with magnetic resonance imaging a major bottleneck problem: imaging time is long,which can't meet the requirements of rapid imaging such as human motion imaging and seriously affect its application in basic research and high-end demand.Therefore,to study and master the domestic high-end magnetic resonance imaging with independent intellectual property rights and improve the imaging speed and accuracy,has the important research value and practical significance.This paper focuses on how to improve the speed and accuracy of magnetic resonance imaging problems around the dictionary learning,compressed sensing theory and parallel imaging technology,and carries out the research of parallel magnetic resonance imaging method based on dictionary learning.It puts forward two kinds of reconstruction methods to ensure the accuracy of the reconstructed image and shorten the imaging time.The main work and contributions of this paper are as follows:(1)The overall architecture of parallel magnetic resonance imaging based on dictionary learning is constructed.Combing compressed sensing theory,two new parallel magnetic resonance imaging models based on dictionary learning are proposed,which include data preprocessing module,dictionary learning module and image update module,and finally obtain the reconstructed image.(2)The method of parallel magnetic resonance imaging method via adaptive sparse representation is proposed.In order to solve the problem that the sparse constraint of traditional magnetic resonance imaging method and the image structure information can't be captured adaptively.This method relies on the strong adaptive advantage of dictionary learn,which transforms the problem of parallel magnetic resonance imaging into L2-L2-L1 minimization problem.Finally,through the experimental test,the proposed parallel magnetic resonance imaging method based on adaptive sparse representation has the ability to preserve image details and to suppress strong noise,and it also can improve the precision of image reconstruction.(3)The method of learning joint-sparse codes for calibration-free parallel magnetic resonance imaging method is proposed.The model uses the joint sparse coefficient between channels to compensate for the lack of information sensitivity,and makes full use of the strong adaptive advantage of dictionary learning,which transforms the problem of parallel magnetic resonance imaging into L2-L-L21 minimization problem.Finally,to verify the feasibility and accuracy of the method,the experimental results show that the reconstructed image of the parallel magnetic resonance imaging method based on the calibration-free of sparse expression is good,and it can save the structure information and suppress the noise well at higher acceleration factor.