The Research Of MR Imaging Reconstruction Method By Using Low Rank Constraints

Dynamic magnetic resonance imaging(d-MRI)data is a set of variational image sequence,because the motion leads to magnetic resonance image artifacts,so d-MRI usually decrease the number of sampling K space data to improve the MRI scanning speed,and to seek a better quality of MRI reconstruction.Based on the low rank constraints of MR image reconstruction,the matrix rank is lower,the higher the accuracy of the reconstructed image.It is important to make research on how to make full use of low rank properties of MRI,so as to improve the quality of the reconstructed image.The main researches can be included as follows:(1)Low rank plus sparse matrix decomposition model is implemented by decomposing the reconstructed image into low rank components and sparse components,which can improve the compressibility of dynamic MRI data.It is one of the successful image reconstruction models,which is used to deal with undersampling separation of the background and dynamic components of magnetic resonance imaging.Furthermore,the patch based low rank method is another kind model of low rank constraints based method,which consist of similar block.The relationships between Low rank and patches are analyzed,and how to handle low rank minimization problem is also discussed.By comparing with the other state-of-art methods,such as L+S method and direct IFFT method,the experimental results show that patch of low rank method can provide more accurate reconstruction of cardiac cine image organizational structure and abdomen data detail structure,improve the image quality with less noise and aliasing artifacts.(2)The multi-scale low rank based method is proposed to implement cardiac MR image reconstruction,which represented a data matrix as a sum of block-wise low rank matrices with increasing scales of block sizes.And the sum of block-wise low rank matrices is used as a constraint to approach the MR image reconstruction.To solve convex optimization ofmulti-scale low rank based method by adopting the method of Alternating direction method of multiplier.Use different ways of sampling and accelerating factor to reconstruct the different types of magnetic resonance image data.Compared with the state-of-art methods,such as the k-t SLR method and L+S method,the proposed MSL method can offer improved reconstruction solution in terms of texture clear and smooth edges,reduce the reconstruction error and get higher signal to error ratio and better structural similarity index.