Study On Deeply Sparse Magnetic Resonance Imaging

Magnetic Resonance Imaging(MRI)have been widely used in medical clinical research because of its has many advantages like ionizing radiation free,multi-parameters imaging and multi-directional imaging and so on.However,imaging speed has always been one of the important reasons for restricting the rapid development of MRI and sometimes it will bring artifacts in the process of imaging,which will affected the reconstructed result.Therefore,how to ensure the image quality at the same time,rapid imaging is the hotspot of MRI technology research.Compressed Sensing(CS)theorem can break the Nyquist sampling theorem,which can use less sampling data through the reconstructed algorithm to restore the original signal.Because MR images are highly compressible,and MR images can be encoded to meet the characteristics of CS coherence.Therefore,MRI technology will bring more room for development with CS in the practical application.For CS-MRI technology,imaging sparseness is very important to restoring image,and single sparse transform which commonly used do not meet people's needs.Therefore,this paper focuses on how to deep-sparse MRI images to complete the rapid imaging problem,mainly completed the following work:(1)In order to deeply sparse images,we substitute the sharp frequency localized Contourlet transform(SFLCT)for wavelet transform,and a new imaging model is proposed which based on SFLCT and nonlocal total variation MRI imaging method.In this method,SFLCT can efficiently capture the curve characteristic of the image,and in the curve singular expression is better than wavelet transform.Besides,Nonlocal total variation(NLTV)is not only can suppresses the noise but also can overcomes the blocky effects were enabled by TV and the problem of losing of edges and details.Based on this,combined with the advantages of these two types of regular terms,a new imaging model is proposed,then solved the model by Algorithm1.The experiments on MR images demonstrate that the proposed method can preserve image details and edges effectively.It produces more accurate reconstruction compared with conventional CS-MRI methods with the same undersampled measurements which can get verified from PSNR,SNR and RENL evaluation standard.(2)By utilizing the structure information besides sparsity of the MR images,a novel CS-MRI method is proposed.The method contains wavelet sparse transform,tree structure,overlapping group sparse and total variation regular term.This method can makes up for the image structure information absence or damage when reconstruction images without considering the influence of the image structure prior information.On the basic of sparse image,the method makes full use of the structure prior information and under this guidance,it can sparse images deeply and accelerates the imaging speed.The experiments on MR images proved that the proposed composite sparse method can increases the image signal to noise ratio effectively.It produces more accurate reconstruction compared with other conventional CS-MRI reconstruction algorithm with the same undersampled measurements which can get verified from SNR,PSNR and SSIM criteria.The methods described in this thesis mainly uses the image structure information to promote the image deep sparse,which is more conducive to the reconstruction of MR images.Experiment simulation also proves that the reconstruction images is more complete in the edge detail than other algorithms.