A high precision method to estimate effective connectivity networks at the group and individual levels

认知脑成像的前沿正在从对孤立的单个大脑区域的检查转向对分布式大脑网络中多个区域之间相互作用的检查。经验证据一直表明，识别大脑网络结构和动力学是理解认知信息处理作为发育、衰老和临床条件等因素的功能的关键。在美国国家科学基金会的资助下，宾夕法尼亚州立大学园区的Peter C.Molenaar博士和他的同事Frank Hillary博士、Ping Li博士和Michael Rovine博士正在开发一种新的方法，即“有效连接图”，以确定不同的大脑区域如何因果地影响彼此的活动。目前可用的方法有几个已知的缺点，拟议的项目正在通过开发新的方法来解决这些缺点，以明确地对大脑区域活动之间的同时和时滞关系进行建模。这一新方法将对脑功能成像研究具有非常重要的意义。它旨在有效地适应连接图中的个体差异，并执行自动数据驱动搜索以获得最佳网络解决方案。研究人员特别强调将该方法与降维技术相结合，以识别感兴趣的大脑区域，并应用新的估计技术，允许大脑区域之间有效连接的任意随时间变化的强度。新的方法基于有或没有外部输入的双线性向量自回归模型，并将该模型扩展到随机状态空间模型。脑成像的应用越来越广泛，越来越注重了解大脑区域的活动是如何整合到相互连接的网络中的。每个人的大脑结构和功能有很大的不同，由于多种因素，他们大脑区域之间的连接可能会随着时间的推移而变化。因此，需要稳健的统计方法来可靠地识别感兴趣的大脑区域网络。该项目旨在通过应用创新的统计方法来实现这一目标，这种方法得到了模拟和经验数据的广泛验证，并在软件工具中得以实施，这些工具既可以是理论驱动的，也可以是数据驱动的。研究人员正在通过大规模的模拟研究和对现有数据集的应用来验证该方法。新的分析工具将在常用的计算平台上实施。为了更广泛的科学界的利益，还将通过开发用户友好的界面、课件和咨询来公开这些方法。

The frontiers of cognitive brain imaging are shifting from examinations of isolated individual brain regions to examinations of interactions between multiple regions in distributed brain networks. Empirical evidence consistently shows that identification of the structure and dynamics of brain networks is the key to understanding cognitive information processing as functions of development, aging, and clinical conditions, among other factors. With funding from the National Science Foundation, Dr. Peter C. Molenaar and his colleagues, Drs. Frank Hillary, Ping Li, and Michael Rovine of the Pennsylvania State University University Park are developing new methods, known as "effective connectivity mapping," to identify how different brain regions causally influence each other's activity. Currently available methods have several known shortcomings, and the proposed project is addressing these shortcomings by developing new methods to explicitly model both contemporaneous and time-lagged relationships among the activities of brain regions. The new method will be highly important for functional brain imaging research. It is designed to efficiently accommodate individual differences in connectivity maps and to carry out automatic data-driven search for the optimal network solutions. The researchers are giving special emphasis to integrating the methodology with dimension-reduction techniques in order to identify brain regions of interest and to application of new estimation techniques allowing for arbitrarily time-varying strengths of effective connections among brain regions. The new methodology is based on bilinear vector-autoregressive models, with or without external input, as well as extension of this model into stochastic state-space models. Application of brain imaging is widespread and increasingly focused on understanding the ways in which activities of brain regions are integrated into interconnected networks. Individual humans differ substantially in brain structure and function, and connections among their brain regions can vary across time due to multiple factors. Therefore, robust statistical methods are required to reliably identify the networks of brain regions of interest. This project is aimed at accomplishing this goal by means of the application of innovative statistical methodology that is extensively validated with simulated and empirical data and implemented in software tools that can be applied in both theory-driven and data-driven ways. The investigators are validating the methodology with large-scale simulation studies and with applications to existing data sets. The new analysis tools will be implemented in commonly used computational platforms. For the benefit of the greater scientific community, the methods will also be made publicly available through the development of user-friendly interfaces, courseware, and consultation.