Carnegie Mellon University

Approaches for accommodating two problems inherent in causal discovery searches on functional MRI data.: Kathleen Gates

Abstract: Identifying network properties has become increasingly necessary for researchers seeking to understand brain processes. This need directly follows recently accepted notions that brain processes can best be thought of as coordinated activity of disparate regions across time as opposed to isolated brain regions driving cognitive functions. Such networks are often arrived at using functional MRI data. The data and questions present two problems in application: 1) given the biological nature of the data, both the lagged effects and contemporaneous effects often exist and must be accounted for in linear models; and 2) there exists heterogeneity in brain networks across individuals that must be precisely estimated without modeling noise inherent in individual-level data. The present talk demonstrates the utility of a graph search algorithm for estimating unified structural equation models (uSEM; also referred to as structural vector autoregression) in neuroimaging research at the individual level. Having arrived at network maps for individuals using uSEM, the present talk showcases recent applications of functional network mapping of the brain with a focus on the specific features of the networks selected by the scientists to test their respective hypotheses.