The Improvement Of Compressed Sensing Magnetic Resonance Imaging

Magnetic resonance imaging system(MRI)is a noninvasive detector.It is widely used in the diagnosis of vital organs such as the brain,heart and mammary gland.However,imaging time is the main bottleneck of this technique.The proposed of compressed sensing provides a new idea for rapid acquisition of MRI,and has become one of the hotspots in the field of medical image processing.In this paper,we first study the basic theory of compressed sensing and magnetic resonance imaging,then improve the CS-MRI model.Specific work includes the following three aspects:(1)Aiming at the shortcomings of the conventional wavelet bases,this paper proposes a new CS-MRI reconstruction model based on Contourlet transform.Firstly,the Contourlet transform is presented to show the superiority of curve and contour information.Then the total variational terms are added to enhance the smoothness of the reconstruction image.At the same time,we increase the redundancy of Contourlet transform to further suppress the artifacts and reduce the reconstruction error.The simulation results show that Contourlet transform works well on reconstructing the image edge.(2)Aiming at the problem of slow reconstruction of traditional compressed sensing model,this paper proposes a new CS-MRI reconstruction model based on wavelet tree sparsity.The research shows that the wavelet coefficients of MR images have non-complete quadtree structure.Firstly,we group the wavelet coefficients using the structural features,and an efficient algorithm is used to reconstruct the new model.Simulation results show that the wavelet tree sparsity structure can accelerate the speed of MR imaging and improve the quality of the reconstructed image.(3)Aiming at the problem that the poor quality of traditional compressed sensing model,this paper proposes a new CS-MRI reconstruction model based on PBDW.Firstly,we train the geometric direction of each image block.And then use it as a priori information to the CS-MRI reconstruction model based on PBDW.The simulation results show that the proposed method can improve the reconstruct quality significantly,but the reconstruction time is too long.We can improve it by making the smooth patch zero directly in the process of training geometric direction.