Large-scale Network Modeling for Brain Dynamics: Statistical Learning and Optimization

脑动力学大规模网络建模：统计学习和优化

• 批准号: 9360104
• 负责人: Xi Luo
• 金额: $ 39.38万
• 依托单位: BROWN UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-09-30 至 2019-06-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/9360104
• 关键词: Address; Affective; Algorithms; American; Anatomy; Behavior; Behavioral; Brain; Brain region; Cognitive; Collaborations; Communities; Complex; Computational algorithm; Custom; Data; Data Analyses; Data Science; Data Set; Dependence; Dimensions; Ensure; Equation; Event; Experimental Designs; Foundations; Functional Magnetic Resonance Imaging; Future; Goals; Graph; Human; Investigation; Journals; Machine Learning; Mental disorders; Methods; Modeling; Modernization; Motor; Neurobiology; Neurosciences; Neurosciences Research; Non-linear Models; Nonlinear Dynamics; Pathway interactions; Pharmacology; Process; Publishing; Scientist; Seeds; Space Models; Statistical Models; Stimulus; System; Techniques; Testing; Therapeutic; Time; Validation; Variant; analytical tool; base; behavioral outcome; brain pathway; cognitive control; cohort; cost; data modeling; design; experimental study; flexibility; high dimensionality; improved; information model; information processing; insight; large scale simulation; model development; nervous system disorder; network models; neural circuit; neuroimaging; neuromechanism; novel; open source; relating to nervous system; simulation; statistics; tool

Summary The human brain is a large, well-connected, and dynamic network. Using functional MRI data, modeling how this network processes the stimulus information has yielded insight on some of the mechanisms of the brain. However, the past efforts, including ours, on using small-scale models yielded limited understanding of how the complete and dynamic neural system functions in task-related experiments. Such understanding cannot be recovered from the data without substantial and collaborative efforts on model development. Towards this goal, we formed a collaborative team from modelers to end-users, and we will develop large-scale methods for task related fMRI (tfMRI), including event-related fMRI, to model whole-brain network dynamics responding to task challenges. Using modern statistical learning principals and large-scale optimization algorithms, we will develop novel methods to model nonlinear, spatial-temporal dependence in high dimensional data of fMRI, stimuli, and behavior outcomes. We will primarily base our methods in the regularized, constrained graphical model (GM) framework, a promising multivariate framework for inferring brain connectivity that has been validated by simulation and anatomical studies. Using this framework, we will develop novel methods to investigate, at a large scale, how changes in connectivity and activation are driven by task challenges and how multiple brain pathways process stimulus information. We will perform comprehensive validation and assessment of the newly developed methods, using both simulated and multiple tfMRI data from large cohorts. Using the scale of modeling that previous approaches cannot readily address without substantial time penalties and maybe also inaccuracies, our collaborative team will also use these methods to investigate various novel questions and hypotheses concerning the neural basis for cognitive control as one of the use cases. We will also develop publicly available, open source implementations for a broad range of use in the neuroimaging community. These modeling efforts will lead to new insights on the networks of large-scale neural circuits, and provide pharmacological targets that may be overlooked using small-scale models.

总结 人脑是一个庞大的、连接良好的、动态的网络。使用功能性MRI数据， 这个网络处理刺激信息，使我们对大脑的某些机制有了深入的了解。 然而，过去的努力，包括我们的努力，对使用小规模模型的理解有限， 在任务相关的实验中完整和动态的神经系统功能。这样的理解不能 从数据中恢复，而无需对模型开发进行大量的协作努力。为了实现这一目标， 我们组建了一个从建模人员到最终用户的协作团队，我们将开发大规模的任务方法 相关功能磁共振成像（tfMRI），包括事件相关功能磁共振成像，以模拟全脑网络动态响应任务 挑战使用现代统计学习原理和大规模优化算法，我们将 开发新的方法来模拟非线性，时空依赖的高维数据的功能磁共振成像， 刺激和行为结果。我们将主要基于我们的方法在正则化，约束图形 模型（GM）框架，一个有前途的多变量框架，用于推断大脑连接， 通过模拟和解剖学研究验证。使用这个框架，我们将开发新的方法， 大规模调查任务挑战如何驱动连接和激活的变化，以及 多条大脑通路处理刺激信息。我们将进行全面的验证， 评估新开发的方法，使用模拟和多个tfMRI数据从大型队列。 使用以前的方法在没有大量时间损失的情况下无法轻易解决的建模规模 也可能是不准确的，我们的合作团队也将使用这些方法来调查各种新的 关于认知控制的神经基础的问题和假设作为用例之一。我们将 我还开发了公开可用的开源实现，用于神经成像领域的广泛使用。 社区这些建模工作将导致对大规模神经电路网络的新见解， 提供了使用小规模模型可能被忽略的药理学靶点。