基于低秩约束的磁共振图像重构方法研究

发布时间：2017-12-26 16:43

本文关键词：基于低秩约束的磁共振图像重构方法研究　出处：《浙江理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：动态磁共振成像数据是一组变化的图像序列,由于运动会导致磁共振图像出现伪影,所以动态磁共振成像通常要求减少采样K空间的数据,以提高磁共振扫描速度,便于寻求更高的磁共振图像重构质量。基于低秩约束的磁共振(MR)图像重构,需处理矩阵的秩越低,重构图像的精度会越高。因此,如何更好的运用磁共振成像的低秩性,对提高重构图像的质量具有重要的研究意义。论文的主要工作与成果如下:(1)低秩+稀疏(L+S,Low-Rank plus Sparse)矩阵分解模型是一种基于低秩组件和稀疏组件的模型,该方法可以提高动态MRI数据的压缩性,用于处理欠采样磁共振成像的背景与动态组件的分离,是比较成功的图像重建模型之一。另外,基于分块低秩(patch based low rank)方法是一种基于搜索相似块进而形成低秩组件的模型,分析了低秩与块之间的关系以及如何处理低秩最小化问题。通过与L+S方法,direct IFFT方法进行对比,实验结果表明基于分块低秩的方法能够更加准确地重建心脏电影图像重要的组织结构和腹部数据的局部细节结构,改善了图像质量并使得噪声和混叠伪影更少。(2)基于多尺度低秩模型(MSL,Multi-Scale Low Rank)的磁共振成像方法将矩阵分解成多尺度的块低秩矩阵之和,并将多尺度块低秩矩阵之和的最小化作为约束条件用于磁共振成像。采用交替方向乘子方法(ADMM,Alternating Direction Method of Multiplier)实现基于多尺度低秩模型的磁共振图像重构凸优化问题的求解。利用不同的采样方式及加速因子,对不同类型的磁共振图像数据进行重构。实验结果表明,相比于k-t SLR(k-t Sparsity Low-Rank)和L+S(Low-Rank plus Sparse)方法,我们所提出的MSL方法具有更好的重建效果(图像结果纹理清晰、边缘光滑),减少了重构误差,获得更高的重构信差比(SER,Signal to Error Ratio),具有更好的结构相似性。
[Abstract]:Dynamic magnetic resonance imaging data is a sequence of image changes, because the movement will lead to artifacts in magnetic resonance images, so the dynamic magnetic resonance imaging is usually required to reduce the sampling K spatial data, in order to improve the scanning speed of magnetic resonance, magnetic resonance to seek higher image reconstruction quality. Based on low rank constraint (MR) image reconstruction, the lower the rank of the matrix, the higher the accuracy of the reconstructed image. Therefore, how to better use the low rank of MRI is of great significance to improve the quality of the reconstructed image. The main work and achievements of the thesis are as follows: (1) low rank + (L+S, Low-Rank plus Sparse sparse matrix decomposition) model is a kind of low rank component and a sparse component based model, this method can improve the compression of dynamic MRI data processing, for separation of background and dynamic component sampling under magnetic resonance imaging. Is one of the more successful model of image reconstruction. In addition, the block based low rank (patch based low rank) method is a model based on searching similar blocks to form low rank components. It analyzes the relationship between low rank and block and how to deal with low rank minimization. By comparing with the L+S method, direct IFFT method, the experimental results show that the method of block based on low rank can more accurately reconstruct local detail structure of heart structure and film image of abdominal data, improve the quality of image and makes the noise and aliasing artifacts less. (2) based on the multi-scale low rank model (MSL, Multi-Scale Low Rank), the magnetic resonance imaging method decomposes the matrix into the sum of multi-scale block low rank matrix, and the minimum of the sum of multi-scale block low rank matrices is used as a constraint condition for MRI. The alternating direction multiplier method (ADMM, Alternating Direction Method of Multiplier) is applied to solve the convex optimization problem of MRI reconstruction based on multi-scale and low rank model. Different types of magnetic resonance image data are reconstructed by different sampling methods and acceleration factors. The experimental results show that, compared with K-T SLR (K-T Sparsity Low-Rank) and L+S (Low-Rank plus Sparse) MSL method, our method has better reconstruction effect (the image texture clear, smooth edges), reduce the reconstruction error, get higher reconstruction contrast (SER, Signal to channel Error Ratio). The structural similarity is better.
【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R445.2;TP391.41

【参考文献】