基于磁共振成像的大脑功能网络动态特性研究

发布时间：2017-12-28 07:17

  本文关键词：基于磁共振成像的大脑功能网络动态特性研究　出处：《西北工业大学》2015年博士论文　论文类型：学位论文

  更多相关文章： 动态功能连接 脑网络 静息态网络(RSN) 字典学习 多视图谱聚类 大脑亚稳态

【摘要】：脑科学研究是当今世界重点研究领域之一。在对大脑的复杂结构和功能进行研究的过程中,磁共振成像(magnetic resonance imaging,MRI)技术发挥了重要的作用,比如:通过弥散张量成像(diffusion tensor imaging,DTI)技术可以在活体无创研究大脑的白质神经纤维结构,利用功能磁共振成像(functional MRI,fMRI)技术可以对脑区功能进行分析。基于磁共振成像对大脑进行脑网络研究及功能连接分析是脑科学研究的一个重要方面。大脑在感知世界的过程中经历着信息感知、传递、协调、变换、存储以及新信息生成等一系列信息处理过程,大脑的功能活动状态随之发生相应的变化,然而,在这一动态变化过程中,大脑是否存在暂时稳定的亚稳态状态?如何准确地描述与表达大脑的亚稳态状态?在大脑完成认知任务的过程中,脑区之间的功能信息交互是如何动态变化的?这些问题的研究对于理解大脑的功能认知具有重要的意义。针对这些热点问题,本文利用大脑的DTI和fMRI图像,基于大脑的全脑功能连接网络,从以下三个方面对大脑功能状态动态性进行了研究:1)动态大脑通用状态模式表达;2)大脑静息态网络(resting state networks,RSN)动态性研究;3)大脑动态信息传递路径追踪。主要研究内容和创新点包括:(1)引入一套高精确度、个体一致对应的大规模全脑网络参考系统,即DICCCOL(Dense Individualized and Common Connectivity-based Cortical Landmarks)网络。利用大脑DTI图像得到白质神经纤维结构信息,通过数学模型表达和量化比较,在全局范围内寻求群体结构差异最小的神经纤维束所对应的一组节点作为不同个体上同一DICCCOL脑区标记,最后共得到358个群体一致的全脑脑区标记节点。这些节点在个体和群体上都具有很高的结构和功能的对应性、可重复性和可预测性,能有效地表示大脑的共有结构连接模式和主要的脑功能区,有助于我们对不同个体或群体进行大脑结构和功能的分析与比较。(2)基于DICCCOL全脑动态功能连接,发现并提出了一种动态大脑亚稳态状态表达形式。通过将fMRI图像映射到相应DTI空间,得到DICCCOL各节点的fMRI信号。采用滑动时间窗口分析方法,计算各节点间随时间变化的三维动态功能连接强度,进一步统计节点的连接强度得到二维节点动态功能连接强度矩阵。观察发现,该矩阵在一些短的时间段内呈现出一定的稳定性,即大脑的亚稳态状态。则大脑的动态状态可以用一系列基于DICCCOL网络的亚稳态全脑功能连接矩阵来表示,为后续研究大脑的动态信息处理过程及状态变化等打下基础。(3)基于大脑亚稳态状态样本的稀疏表达与分类,提出了一种动态大脑通用状态模式空间描述方法。将个体大脑的亚稳态状态,用全脑近似稳定连接模式(whole-brain quasi-stable connectome pattern,WQCP)样本表示后,将群体内所有WQCP样本组合在一起,采用一种基于Fisher判别准则和稀疏表达的字典学习算法,即FDDL(Fisher discriminative dictionary learning)方法对样本进行学习和分类,最后将大脑的动态状态变化用若干通用状态模式来表示,即动态大脑通用状态模式空间。通过对静息态和任务态大脑WQCP样本进行统一表达,发现二者具有不同的动态通用状态变化模式。任务态fMRI数据群组激活检测实验结果表明,任务扫描过程中出现静息态模式的个体没有很好的遵从实验设计的任务范式,这为大脑f MRI图像质量控制提供了一种有效的辅助手段。(4)基于静息态大脑的动态通用状态模式,对RSN网络的动态性进行了研究。将静息态大脑的每个动态通用状态模式看作是全脑功能连接状态的一种单独的视图,采用多视图谱聚类方法对DICCCOL节点进行聚类,得到一系列包含了丰富的动态信息的DICCCOL子集,即动态RSN网络,与同样由DICCCOL节点表达的静态RSN网络进行比较。结果发现,一些RSN网络在静息态下呈现出很好的稳定性,比如缺省模式网络(default mode network,DMN),视觉RSN网络等,另外一些RSN网络包括运动相关网络则呈现出强动态性和变化性,说明这些动态性强的网络对静息态大脑功能区域的动态交互起关键作用。该研究为RSN网络的研究以及研究静息态大脑功能信息处理机制提供了一种新的视角。(5)基于大脑亚稳态状态时间序列及动态通用状态模式表达形式,对各亚稳态下所隐含的大脑动态信息传递路径进行了追踪研究,即认为大脑亚稳态状态的形成是由于不同脑区之间的动态信息交互造成的。首先,根据亚稳态时间序列信号,将DICCCOL节点聚类到不同的空间子网络,对各子网络的平均信号进行拟合并检测其峰值激活时刻,根据激活时刻的先后对子网络进行排序;然后,建立信息传递概率模型,采用动态规划的方法求解最大概率路径,即最优信息传递路径,由每个子网络中的关键“路由”节点来表达。通过对视觉任务下正常青少年组和患有创伤后应激障碍疾病(post-traumatic stress disorder,PTSD)青少年组比较,发现两组群体大脑在该视觉任务中的高频率“路由”节点分布具有明显差异,正常组在视觉皮层区域呈大面积高频率分布,而PTSD组则相对弱许多,此外,PTSD组的功能活动涉及更多的大脑皮层区域。该研究对于了解PTSD疾病大脑的功能信息处理机制具有参考价值。
[Abstract]:The research of brain science is one of the most important research fields in the world. In the process of complex structure and function of the brain, magnetic resonance imaging (magnetic resonance imaging MRI) technology has played an important role, for example: by using diffusion tensor imaging (diffusion tensor, imaging, DTI) technology can be no white matter nerve fiber structure and study of the brain in vivo, using functional magnetic resonance imaging imaging (functional MRI fMRI) technology on brain function analysis. The brain network research and functional connection analysis based on magnetic resonance imaging (MRI) are an important aspect of brain science. The brain through perception, transmission, coordination, transformation, storage and generation of new information and a series of information processing in the process of perceiving the world, functional state of the brain change accordingly, however, in the process of dynamic change in the brain, the existence of metastable state temporarily stable metastable state? How to accurately describe and express the brain? The process of cognitive tasks in the brain, brain function and information interaction zone between how dynamic changes? The study of these problems for understanding the brain's cognitive function has important significance. Aiming at these issues, this paper use the brain DTI and fMRI images, the whole brain functional brain network connection based on studied the dynamic state of brain function from the following three aspects: 1) the expression of general dynamic brain state model; 2) brain resting state network (resting state networks, RSN) dynamic research; 3) tracking brain dynamic information transmission path. The main research contents and innovations include: (1) introduce a set of high-precision, individual correspondence corresponding large-scale large-scale brain network reference system, namely DICCCOL (Dense Individualized and Common Connectivity-based Cortical Landmarks) network. Get the structure information of white matter nerve fibers by brain DTI images, through the mathematical model expression and quantitative comparison, in the global scope for different population structures a minimal set of nodes corresponding to the nerve fiber bundle as a DICCCOL marker on different individual brain regions, finally to mark the whole brain area of 358 nodes in group consensus the. These nodes have the structure and function of high in individuals and groups on the correspondence, repeatability and predictability, can effectively represent the brain's total structure connection mode and main functional areas, help us to different individuals or groups of brain structure and function and comparison. (2) based on the dynamic functional connection of DICCCOL whole brain, a dynamic state of brain metastable state was found and proposed. By mapping the fMRI image to the corresponding DTI space, the fMRI signal of each node of the DICCCOL is obtained. The sliding time window analysis method is used to calculate the three dimensional dynamic functional connection strength between nodes, and further calculate the connection strength of nodes to get two-dimensional node dynamic functional connectivity strength matrix. It is found that the matrix shows a certain stability in some short periods of time, that is, the metastable state of the brain. The dynamic state of the brain can be represented by a series of metastable whole brain functional connectivity matrices based on DICCCOL network, which lays the foundation for subsequent research on the dynamic information processing process and state changes of the brain. (3) based on the sparse representation and classification of brain metastable state samples, a dynamic general state pattern spatial description method is proposed. The individual brain metastable state, connection mode in whole brain (whole-brain quasi-stable connectome pattern approximate stability, WQCP) said after the group within the sample, all WQCP samples together, using a Fisher discriminant criterion and the expression of sparse dictionary learning algorithm based on FDDL (Fisher discriminative dictionary learning) learning and classification method of the sample, the dynamic state changes of the brain with a number of general state model to represent the dynamic state of brain, general pattern space. Through the unified expression of the resting state and the task state brain WQCP samples, it is found that the two models have different dynamic general state change patterns. Experimental results of task group fMRI data group activation test show that individuals with resting state patterns in task scanning process do not follow the task paradigm of experimental design well, which provides an effective auxiliary method for F MRI image quality control. (4) based on the dynamic general state pattern of the resting state of the brain, the dynamics of the RSN network is studied. The general state of each dynamic mode of resting state brain as a single view of the whole brain functional connectivity state, is used to cluster the DICCCOL node multi spectral clustering method, get a series of dynamic information contains rich DICCCOL subset, namely dynamic RSN network, compared with the static RSN network also expressed by DICCCOL node. The results showed that some RSN network shows a good stability in the resting state, such as the default mode network (default mode network, DMN, RSN) visual network, some RSN network including network sports related showed a strong dynamic and variable, the dynamic network strong state interaction plays a key role the dynamic of resting state brain function area. The research provides a new perspective for the research of RSN network and the study of the resting state brain function information processing mechanism. (5) the brain metastable state of time series and general dynamic state model expression based on brain dynamic information transmission path implied on the metastable studied, that the formation of metastable state of the brain is due to dynamic information between different regions of the brain caused by interaction. First, according to the metastable time series signal, the DICCCOL nodes are clustered into different space subnetworks, and the average signals of each subnetwork are combined to detect its peak.
【学位授予单位】：西北工业大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：O482.532;R338
，

本文编号：1345095