Analytical Tools for Complex Brain Networks: Fusing Novel Statistical Methods and Network Science to Understand Brain Function

点击翻译按钮获取中文摘要

The emerging area of brain network analysis considers the brain as a system, providing profound clinical insight into links between system-level properties and health outcomes. Network science has facilitated these analyses and our understanding of how the brain is structurally and functionally organized. Nonetheless, methods for statistically modeling and comparing groups of networks have lagged behind. The development of such methods would constitute a significant innovation and have a significant impact on scientific progress for researchers seeking to better understand brain function and how it changes across different mental states and disease conditions. Current approaches for group comparisons rely on either (a) a specific extracted summary metric, which has limited clinical value due to low sensitivity/specificity and a lack of clinical interpretability, or (b) mass-univariate nodal or edge-based contrasts that ignore the network’s inherent topological properties and yield poor statistical power. While some univariate approaches have proven useful, gleaning deeper insights into changes in functional organization demands methods that leverage the data from an entire brain network. We are currently ill-equipped to answer many fundamental and pressing questions, including the relationship between cognition and (a) rest-to-task brain network changes, (b) within-task dynamic brain network changes, and (c) brain network topology. In this proposal, we will address these needs by fusing novel statistical methods with network-based functional neuroimage analysis to advance our understanding of normal and abnormal brain function. More specifically we will: Develop a mixed modeling framework that allows integrating multitask brain network data to assess state changes (Aim 1a) and that allows assessing within-task network dynamics (Aim 1b), develop a permutation testing framework for brain network comparisons that allows assessing continuous predictors and controlling for confounding covariates (Aim 2), and develop and deploy a Matlab package implementing the new methods (Aim 3). Our novel methods have transformative potential: they will allow use of validated statistical methods to compare brain networks and thereby illuminate neurobiological correlates of abnormal brain changes. This innovation will enable researchers to investigate how phenotypic traits are related to brain network organization, and are critical for further progress in this field. The insights gained from this project will be important for the study of numerous brain diseases and chronic health conditions; they will also have clinical utility in the realm of precision medicine strategies.