Causal Inference for Neuroimaging

翻译后摘要：新技术，如功能性磁共振成像（fMRI）已经帮助我们的能力， 了解人脑的功能。区分因果关系和虚假关系是非常有趣的， 重要的基础神经科学和应用在医学，药物开发和临床实践。 神经科学家想知道实验刺激如何影响大脑功能， 大脑区域之间存在因果联系，以及神经活动如何调节刺激和刺激之间的关系。 结果。研究人员应用各种统计程序（例如，格兰杰因果关系，动态因果模型， 结构方程模型和直接图形模型）的功能磁共振成像数据，解释由此产生的关联 作为影响，往往是不恰当的。这是一个大问题。上一个赠款供资周期奠定了 为神经科学研究中偶然推理的原则性方法奠定了基础， 统计文献中关于因果推断的结果符号。在这里，我们试图扩展框架- 在那里开展的工作是为了更好地研究中介作用，即一个或多个大脑区域的神经活动进行中介作用 治疗-结果关系和有效连接，其中不同区域的激活是因果关系 有联系开始，我们定义了新的因果效应，并构建了一个新的全脑模型， 在更现实的假设下，研究区域层面的本地化、中介和有效连通性 而不是那些在功能性磁共振成像数据建模中产生的数据。此外，我们还创建了一种新的行为统计方法- 使用高维中介分析，识别可能调解网络之间的关系， 治疗和结果。最后，建立在文献上的工具变量和结构方程模型- els，我们开发了新的方法来研究调解和有效的连接后，应用外部 神经刺激使用最近开发的技术，如经颅磁刺激（TMS）， 经颅直流电刺激（tDCS），开始看到广泛使用的理论神经， 科学和应用背景。我们将这些方法应用于创伤后应激障碍研究的fMRI数据， 热痛和社会评价。

Abstract: New technologies such as functional magnetic resonance imaging (fMRI) have aided our ability to understand human brain function. Distinguishing causal from spurious relationships is of great interest and importance for both basic neuroscience and applications in medicine, drug development, and clinical practice. Neuroscientists want to know how experimental stimuli affect brain function, how neural activity in different brain regions are causally linked, and how neural activity mediates the relationship between a stimulus and an outcome. Researchers apply various statistical procedures (e.g., Granger causality, dynamic causal models, structural equation models, and directed graphical models) to fMRI data, interpreting the resulting associations as effects, often inappropriately. This is a major problem. The previous funding cycle of this grant laid the groundwork for a principled approach to casual inference in neuroscience research, employing the potential outcomes notation used in the statistical literature on causal inference. Here we seek to extend the frame- work developed there to better study mediation, where neural activity in one or more brain regions mediates a treatment-outcome relationship, and effective connectivity, where activations in different regions are causally linked. To begin, we define new causal effects and construct a new whole brain model that will facilitate the study of localization, mediation, and effective connectivity at the regional level under assumptions more realistic than those typically made in modeling fMRI data. Further, we create a new statistical method for conduct- ing high-dimensional mediation analysis that identifies networks that may mediate the relationship between a treatment and outcome. Finally, building on the literature on instrumental variables and structural equation mod- els, we develop new methods for studying mediation and effective connectivity after the application of external neurostimulation using recently developed technologies such as transcranial magnetic stimulation (TMS) and transcranial direct current stimulation (tDCS) that are starting to see widespread use in both theoretical neuro- science and applied contexts. We apply the methods to fMRI data from studies of post-traumatic stress disorder, thermal pain and social evaluation.