Research On MR Image Reconstruction Algorithm Based On Compressed Sensing

Magnetic Resonance Imaging (MRI) has been widely applied in clinical diagnosis because of high contrast resolution, many imaging parameters, no harm to the human body,multi-directional imaging and so on. However, long magnetic resonance imaging time has been restricting the further development of MRI technology. Compressive Sensing (CS) theory breaks through the limitation of the traditional sampling theorem, and realizes the reconstruction with a small number of acquisition signals. The application of this theory in MRI technology, can reduce the sampling time, and accelerate the magnetic resonance imaging. In recent years, the MRI sparse reconstruction technology based on CS has become a hot study for many scholars at home and abroad.A variety of MRI sparse reconstruction methods based on CS are studied. This thesis focuses on the improvement of the existing Dictionary Learning Magnetic Resonance Image(DLMRI) reconstruction method to increase the quality of MRI down-sampling reconstruction.This thesis discusses mainly the existing DLMRI reconstruction method, and the main work is to improve dictionary learning and sparse coding. On the one hand, the initial image is reconstructed by Shift-Invariant Discrete Wavelet Transformation (SIDWT) in the dictionary learning stage in this thesis. This method analysises the image signal omnidirectionally, and gives a better description on the edge detail information of the image, which makes up for the deficiency of the zero-filling initial image obtained by Fourier reconstruction. Comparison experiment shows that the improved algorithm improves the quality of the initial training data of the dictionary learning, reduces the number of iterations of the program and enhances the quality of reconstructed MRI. On the other hand, in the sparse coding stage, this thesis obtains the sparse representation coefficient by the Stage-wise Orthogonal Matching Pursuit (StOMP)algorithm, which weakens the effect of sparse threshold on Orthogonal Matching Pursuit(OMP) algorithm. Selecting multiple atoms every time, the method enhances the matching of dictionary atomic selection. Additionally, an error parameter is adopted in the StOMP algorithm to reduce the running time of the program on the premise of a relatively high quality of image reconstruction.All in all, this thesis improves the DLMRI reconstruction algorithm to realize the down-sampling reconstruction of MRI. The experimental results demonstrate the algorithm proposed in this thesis is adaptable to different types of magnetic resonance images and the error of the reconstructed MRI is smaller compared with the existing MRI sparse reconstruction algorithms, which effectively reconstructs the image under down-sampling.