大呼吸运动下高时空分辨率腹部动态磁共振灌注图像的压缩感知重建

发布时间：2018-05-06 07:45

本文选题：动态MRI + 压缩感知重建　；参考：《浙江大学》2016年硕士论文

【摘要】：磁共振成像(MRI)因其具有无辐射,多平面成像,扫描参数多,提供病理信息丰富,软组织对比分辨率高等优点,在临床医学和科研中得到了广泛的应用。而动态MR2可用于捕捉成像目标的动态变化过程,如心脏的收缩舒张、腹部对比剂灌注成像等,所以普遍应用于人体运动器官和功能成像等领域。但是受限于扫描时间,现有的动态MRI技术很难获得同时满足高时空分辨率和高信噪比的重建图像序列。而且扫描期间受检者的呼吸会引起内脏器官的非刚性运动,破坏K空间扫描数据的一致性,引入运动伪影。多次屏气是动态MRI扫描中提高图像质量的一种常用呼吸控制协议,但是在长时间的屏气之后,受检者通常会快速地大幅度换气,导致图像质量严重下降。本文针对大呼吸运动,研究了基于高度降采样数据的动态MRI重建技术,主要成果包括：(1)针对降采样动态MR成像问题,提出了一种基于时空块的双稀疏字典学习算法,通过结合双稀疏模型和时空域的高相关性来训练3D字典。相比于传统的单层字典方案和k-tFOCUSS算法,在重建图像的时空域局部细节特征和去除混叠伪影上获得了更好的效果。(2)针对人体腹部大幅度呼吸运动问题,提出了一种基于马尔可夫随机场的离散优化配准算法,解决了传统的基于B样条的自由形变配准算法存在的问题。实验表明此算法不但能够很好地建模slice方向大幅度和不连续运动场,而且性能指标要优于同类deedsMST和Elastix配准算法。(3)结合上述字典学习和呼吸运动配准算法,实现了大呼吸运动下降采样腹部灌注动态MR重建系统。利用多线圈螺旋采样方式采集人体腹部数据进行测试,对实验结果进行多方面分析,并与TRACER和PROUD算法进行比较,本文算法能够在高时空分辨率下有效改善MR体三维图像的重建质量。本文实现的腹部灌注动态MR重建系统,能够在呼吸运动和对比剂双重动态因素影响下,从约50倍降采样的K空间数据中重建出高帧率高信噪比的MR体三维图像。本系统在数据采集中无需呼吸门控等其他技术的配合,采样方式较简便。后期重建过程中对关键算法模块进行多方面改进,经过充分的实验验证,并与现有的先进算法进行比较,重建效果基本满足应用需求。
[Abstract]:Magnetic resonance imaging (MRI) has been widely used in clinical medicine and scientific research because of its advantages of non-radiation, multi-plane imaging, multiple scanning parameters, abundant pathological information and high contrast resolution of soft tissue. Dynamic MR2 can be used to capture the dynamic changes of imaging targets, such as systolic and diastolic heart, abdominal contrast media perfusion imaging, so it is widely used in the field of human motor organs and functional imaging. However, limited by the scanning time, the existing dynamic MRI technology is difficult to obtain the reconstruction image sequence which can satisfy both high spatio-temporal resolution and high signal-to-noise ratio (SNR). Moreover, the respiration of the subject during scanning will cause non-rigid movement of visceral organs, destroy the consistency of K space scanning data, and introduce motion artifacts. Multiple breath-holding is a common respiratory control protocol for improving image quality in dynamic MRI scanning. However, after a long period of breath-holding, the subjects usually breathe quickly and greatly, which results in a serious deterioration of image quality. In this paper, a dynamic MRI reconstruction technique based on height downsampling data is studied for large respiratory motion. The main achievements include: 1) to solve the problem of dynamic Mr imaging, a novel algorithm based on spatio-temporal block is proposed for learning double sparse dictionaries. A 3D dictionary is trained by combining the double sparse model with the high correlation in time and space domain. Compared with the traditional monolayer dictionary scheme and k-tFOCUSS algorithm, a better result is obtained on local detail features and aliasing artifacts of reconstructed images in space-time domain. A discrete optimal registration algorithm based on Markov random field is proposed, which solves the problem of traditional free deformation registration algorithm based on B-spline. Experiments show that the proposed algorithm not only can model large and discontinuous motion fields in slice direction, but also has better performance than deedsMST and Elastix registration algorithms. A large respiratory motion descending sampling abdominal perfusion dynamic Mr reconstruction system is realized. The multi-coil spiral sampling method is used to collect human abdominal data for testing, and the experimental results are analyzed in many aspects, and compared with TRACER and PROUD algorithms. This algorithm can effectively improve the reconstruction quality of Mr volume 3D images at high spatial and temporal resolution. The abdominal perfusion dynamic Mr reconstruction system can reconstruct 3D Mr volume images with high frame rate and high signal-to-noise ratio from K-space data of about 50 times lower sampling under the influence of both respiratory movement and contrast agent. This system does not need the cooperation of other technologies such as respiratory gating in data acquisition, and the sampling method is relatively simple. In the later stage of reconstruction, the key algorithm modules are improved in many aspects. After full experimental verification, and compared with the existing advanced algorithms, the reconstruction results basically meet the needs of application.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：R445.2

【参考文献】