基于高维数据压缩感知的磁共振快速成像关键技术研究

Magnetic resonance imaging (MRI) is an advanced imaging technology that plays a vital role in the diagnosis and treatment of human diseases such as tumor and cardiac disease. However, the current MR imaging speed can’t meet the recent clinic needs such as multidimensional dynamic imaging. Parallel imaging and compressed sensing are two major promising techniques for the development of fast imaging in MRI. Most of the research in this direction focuses on the low-dimension domains. In this research, we focus on the development of fast imaging methods for processing high-dimensional MRI signals. In our proposed method, the sub-sampled MR data are arranged into 3D/4D tensors, and then a mathematical operation- Walsh transformation is used to implement its sparse representation used in compressed sensing imaging. Accelerated with GPU computing, an efficient reconstruction can then be realized for high-dimensional imaging. In this method, the image quality has been improved through a full exploration of the data redundancy of MRI data, and the reconstruction speed is enhanced by optimization the data flow and memory access mechanisms during GPU computing. Through acceleration in two aspects, that is, the MRI data sampling and image reconstruction, high-quality cardiac MR images can be obtained in a real-time manner. With the success of this project, the new multidimensional MRI technology will meet the increasing clinical demands, impacting positively on the early detection and diagnosis of diseases. On the other hand, this project has significant potential in promoting our domestic MRI industry through technological innovations embedded in this proposal.

磁共振成像是诊断和研究肿瘤、心脏病等人类重大高发疾病最为有效的影像学方法，但其成像速度远不能满足多维动态成像等最新临床需求。由于理论和技术原因，当前以并行成像和压缩感知为主的快速成像研究主要集中在低维图像重建上。本项目以高维磁共振图像重建为研究重点，把并行成像多线圈高度欠采样的动态磁共振图像数据视为四维张量，提出高维数据稀疏变换方法，改进重建算法，优化图像处理器并行计算，经过磁共振体模和真实人体成像测试，实现基于压缩感知的快速多维动态重建。项目充分挖掘不同维度上磁共振数据的结构相关性，优化高维数据在图像处理器内存的访问机制，在数据采集和图像重建两个环节上创新加速策略，实现多维动态目标的磁共振高质量快速成像，使得磁共振成像进一步满足临床诊治需求，为促进相关产业的发展提供理论和技术支持。

磁共振成像是肿瘤、心脏病等人类重大高发疾病诊断和研究的有效影像学方法，但成像速度不够，难以满足较高要求的动态医学成像需求是其最大短板。特别是在心脏动态成像、脑功能研究相关的高分辨率成像。多线圈并行采集成像，通过压缩感知方法减少数据采集量是确保图像质量情况下快速成像的主要方法。.项目通过对图像数据稀高效疏变换表达和高分辨率图像重建的研究，将多线圈采集的多幅，系列时序一般分辨率的图像视为一个三维张量，通过约束图像内部和之间的变换稀疏性，进行三位压缩感知重建，从而实现降低采样率情况下的高质量图像快速成像。项目组研究了一维到三位不同维数沃尔什变换对磁共振图像的稀疏变换特性，并详细研究了沃尔什变换从高维数到低维数的降维实现过程，三维沃尔什变换可以通过一系列的一维沃尔什变换实现，首先在每个二维平面内逐行做一维沃尔什变换，然后在每个二维平面内逐列做一维沃尔什变换，最终通过系列一维沃尔什变换实现三维数据的稀疏变换。对高维MRI信号，沃尔什变换具有良好的稀疏变换能力和高效的代码执行效率。在三维动态成像中，项目对多线圈图像做三维沃尔什稀疏变换，同时挖掘其在三个维度上的信息相关性，然后使用L1范数约束其稀疏性。L1约束的ISPIR方法被重定义为三维压缩感知重建方法与ISPIR并行成像方法的结合。结合快速组合分裂法（FCSA）重建算法，形成改进的重建方法，借助FCSA算法中值分裂和运算符分裂的思路，加入对标定一致性的约束。该方法提高了磁共振成像速度，被用于心脏磁共振动态成像等高动态性成像场合，以及脑淋巴成像等高分辨率大数据量采样场景。