Multilevel mediation techniques for fMRI

近年来，在对人脑内部运作进行成像的技术能力方面取得了巨大的进步。当人类参与心理过程时，功能性磁共振成像（fMRI）测量整个人类大脑的活动-数十万个数据点-每秒都可以获得。收集如此庞大而复杂的数据集的能力的发展已经超过了分析它们的方法。一个关键的差距是，虽然大脑是按照连接多个大脑区域的功能通路组织起来的，但还没有足够的、广泛可用的方法来识别和测试人类功能通路的完整性。在美国国家科学基金会的支持下，哥伦比亚大学的Tor Wager博士及其同事将开展研究，旨在开发用于模拟人类大脑功能连接的分析技术，并为此目的开发高质量的软件。他们打算开发的多级调解/调节（M3）框架将扩展多方程系统多级建模的最新进展，并扩展其功能以检查许多大脑区域之间的连接性。M3框架将是第一种可以提供关于跨越两个以上区域的大脑连接的人口推断的方法。神经对疼痛的反应将被用作模型系统来开发和测试框架，因为疼痛通路的特征很好。M3框架将用于识别丘脑皮质上行通路-现有方法无法提供的创新-并检查它们的活动如何受到额叶皮层中“控制”回路的调节。总之，该方法和应用程序有望阐明如何在人脑中研究功能通路和内部反馈。这项研究将导致结构方程建模的进步，包括协方差作为随机变量在多层次框架中的建模，稳健估计，以及结构模型和降维技术的融合。这项工作的另一个创新方面将是开发递归算法，允许在结构模型中选择变量，这将允许混合验证/模型构建方法应用于大型数据集。用户友好的软件和文档将使不同领域的研究人员能够使用新工具，包括神经科学，经济学，生物工程和心理学，并将鼓励更广泛地参与基于大脑的人类认知和情感过程分析。这些方法还将通过美国多学科研究小组之间的科学互动，在课程和研讨会上传播。S.以及通过在哥伦比亚网站上发布的供公众下载的软件。最后，本科生和研究生在统计学和认知神经科学交叉培训将使学生在这两个领域受益于接触到新的概念和technology.This奖是支持作为2006财政年度数学科学优先领域的数学社会和行为科学（MSBS）特别竞争的一部分。

Recent years have seen dramatic advances in the technical capacity to image the internal workings of the human brain. As humans engage in mental processes, functional magnetic resonance imaging (fMRI) measurements of the activity of the entire human brain -- hundreds of thousands of data points -- can be acquired each second. Developments in the ability to collect such enormous and complex datasets have outstripped methods for analyzing them. One critical gap is that whereas the brain is organized in terms of functional pathways connecting multiple brain regions, no adequate, widely-available methods exist for identifying and testing the integrity of functional pathways in humans. With support from the National Science Foundation, Dr. Tor Wager and colleagues at Columbia University will conduct research aimed at developing analysis techniques for modeling functional connectivity in the human brain and produce high-quality software for this purpose that will be made publicly available. The multilevel mediation/moderation (M3) framework that they intend to develop will extend recent advances in multilevel modeling of multi-equation systems and expands their functionality to examine connectivity among many brain regions. The M3 framework will be the first method that can provide population inference about brain connectivity spanning more than two regions. Neural responses to pain will be used as a model system to develop and test the framework because pain pathways are well-characterized. The M3 framework will be used identify ascending thalamocortical pathways - an innovation not afforded by existing methods - and to examine how their activity is regulated by "control" circuits in the frontal cortex. Together, the method and application are expected to shed light on how functional pathways and internal feedback may be investigated in the human brain. This research will lead to advances in structural equation modeling, including the modeling of covariances as random variables in a multi-level framework, robust estimation, and the fusion of structural models and dimension reduction techniques. Another innovative aspect of this work will be the development of recursive algorithms that permit variable selection in structural models, which will allow for hybrid confirmatory/model-building approaches necessary for application to large datasets. User-friendly software and documentation will make the new tools accessible to researchers in diverse fields, including neuroscience, economics, bioengineering, and psychology, and will encourage broader participation in brain-based analysis of human cognitive and affective processes. The methods will also be disseminated in courses and workshops, through scientific interactions between multi-disciplinary research groups in the U. S. and internationally, and through software posted for public download on Columbia's website. Finally, undergraduate and graduate cross-training in statistics and cognitive neuroscience will allow students in both areas to benefit from exposure to new concepts and techniques.This award was supported as part of the fiscal year 2006 Mathematical Sciences priority area special competition on Mathematical Social and Behavioral Sciences (MSBS).