|
Abstract:
|
In fMRI studies, an important goal is to investigate functional brain networks that capture how different brain regions interact when participants are at rest, perform behavioral tasks or during brain stimulation. Thus, studies increasingly collect multiple fMRI sessions in the same subject under different conditions. Functional brain networks are commonly defined using marginal or partial correlations under the assumption that observations are independent. However, in multi-session studies, we expect the time-series both within and across sessions in the same individual to be dependent. Failure to account for this dependence can result in overoptimistic statistical inferences regarding brain networks. To flexibly model this temporal dependence and accordingly whiten multi-session data, I will discuss employing matrix-variate rather than multi-variate distributions with corresponding estimators to infer individual brain networks. Accounting for temporal dependence using whitening improves estimation and structure learning of brain networks. Additionally, this procedure also improves the reliability of subsequent topological metric analyses of networks.
|
