Functional Magnetic Resonance Imaging of Brain States and Their Transitions

建立准确的人脑区域交流模型对于理解负责认知过程和产生行为的潜在功能至关重要。我的研究计划致力于开发信号处理和数据分析策略，用于在网络层面上进行区域间脑通信的静息状态功能磁共振成像(FMRI)研究。FMRI通过其对血液磁性的敏感性来告诉我们大脑的活动，而血液的磁性取决于血液的氧合水平。静息状态fMRI信号的内在波动在网络的大脑区域之间是高度同步的；同步程度被称为“连通性”。然而，随着大脑区域在网络之间相互作用，连接性会发生变化，导致与认知过程相关的临时大脑状态之间的转换。因此，了解连接的动态性质很重要。现有的连通性数据分析方法缺乏在任何时间点对功能连接的存在或改变做出统计推断的能力。这极大地阻碍了我们准确描述人类大脑活动的复杂性的能力，并浪费了最近开发的亚秒功能磁共振成像的潜在好处。在这个拟议的项目中，我的实验室将开发信号处理和数据分析方法，以准确地建模人类大脑功能网络的空间和时间动态，并具有统计置信度。在目标1中，我们将开发用于静息状态fMRI的数据处理和分析技术，该技术使用两级分层观察模型和贝叶斯(条件)方法来估计模型参数。使用这种方法，我们可以计算条件均值和协方差，这反过来又允许在任何时间点计算t统计量。在目标2中，我们将开发生成代理数据的技术，以准确估计零分布数据，这是准确计算统计显著性所必需的。这不是微不足道的，当前的动态连通性分析方法缺乏这一点。然后，我们将使用模拟数据集和来自在线功能连接组数据库的人类数据来验证我们的方法并测试其重复性。在目标3中，我们将使用我们在集群和机器学习算法中的方法，定义不同但临时的静止大脑状态以及它们之间的转换。这些目标将产生新的技术进步，显著提高我们对人脑功能连接的知识，并将为未来研究更复杂的大脑系统提供能力。我们的研究计划还将提供一个良好的环境，在其中培训高素质的成像技术开发人员，这是加拿大研究和工业加速增长的一个领域。

Establishing accurate models of human brain region communication is critical for understanding the underlying functions responsible for cognitive processes and generating behaviour. My research program is dedicated to the development of signal processing and data analysis strategies for resting-state functional magnetic resonance imaging (fMRI) studies of inter-regional brain communication at the network level. fMRI tells us about brain activity through its sensitivity to the magnetic properties of blood, which are dependent on blood oxygenation levels. Intrinsic fluctuations of resting-state fMRI signals are highly synchronous between brain regions of a network; the degree of synchrony is termed "connectivity". Connectivity, however, varies as brain regions interact between networks, leading to transitions between temporary brain states associated with cognitive processes. Thus, is it important to understand the dynamic nature of connectivity. Existing connectivity data analysis approaches lack the ability to make statistical inferences regarding the presence of or change in a functional connection at any point in time. This greatly hinders our ability to accurately characterize the complexity of human brain activity and wastes the potential benefits of recently-developed sub-second fMRI. During this proposed project, my lab will develop signal processing and data analysis approaches to accurately model the spatial and temporal dynamics of the human brain's functional networks, with statistical confidence. In Aim 1, we will develop data processing and analysis techniques for resting-state fMRI that use a two-level hierarchical observation model and a Bayesian (conditional) approach to estimate model parameters. Using this approach we can compute the conditional mean and covariance, which in turn permits the computation of a t-statistic at any point in time. In Aim 2, we will develop techniques to generate surrogate data to accurately estimate null distribution data, which is necessary for accurately computing statistical significance. This is non-trivial and is lacking from current dynamic connectivity analysis approaches. We will then validate our approaches and test their reproducibility using simulated datasets and human data from online functional connectome databases. In Aim 3, we will use our approaches in clustering and machine learning algorithms, to define distinct but temporary stationary brain states and the transitions between them. These aims will generate novel technical advances to significantly advance our knowledge of the functional connections of the human brain, and they will provide the future ability to investigate more complex brain systems. Our research program will also provide an excellent environment in which to train high quality personnel in imaging technology development, an area of accelerating growth in Canadian research and industry.