American Statistical Association
Functional connectivity seeks to understand how remote brain regions are related during the course of a functional MRI study. In this talk, we outline statistical innovations in the two major approaches to studying functional connectivity: (1) pattern recognition via dimension reduction and (2) network analysis. First, methods such as PCA (principal components analysis) and especially ICA (independent components analysis) are commonly used to find connected brain regions or patterns of co-activation. These techniques, however, implicitly assume that the noise in fMRI data is independent. By assuming that the noise follows a spatio-temporal process, we introduce a generalization of PCA that better decomposes the variation in fMRI data, yielding more scientifically interpretable results. Second, when connections between brain regions are denoted via a network, one may ask how these functional connectivity networks differ between two sets of subjects. Modeling connectivity via Markov Networks, we introduce a novel inferential framework for determining statistically significant differences across a population of brain networks. We will demonstrate these methods with preliminary results on an fMRI study of synesthesia and the ABIDE autism study.
|Date:||Tuesday, December 10, 2013|
“Central Park” meeting area at the Child Study Center
One Park Avenue (between 32nd and 33rd Streets)
New York, New York