Abstract:
|
In studying mechanisms underlying a disorder or clinical phenomena of interest, we often face multiple potential mediators. For example, in the study of normal aging or pathology related cognitive decline, a hypothesis of interest in whether the effect of aging on cognition is mediated by neural substrates measured by brain imaging. In this case, a mediator is barely summarized as one variable, in fact, often summarized by either by a set of pre-identified ROIs. We proposed a sparse multiple mediation analysis for high-dimensional brain image mediators. Sparse structural equation models with L1-regularization can select a parsimonious model. For highly correlated image measures, functional principal component analysis can be used for dimension reduction and the proposed algorithm can find the most relevant information of mediation. Reference ability neural network data will be presented during the talk.
|