|
Abstract:
|
Data on the brain's structural or functional connections are frequently represented in the form of networks, with a different network for each subject in the study. These networks all share the same set of nodes and can thus be analyzed jointly. Current work tends to either reduce them to global summaries such as modularity, or vectorize the edge values and ignore network structure. Here we propose a method for modeling brain networks via linear mixed effects models which takes advantage of the community structure, or regions, known to be present in the brain. The model allows us to compare different populations (for example, healthy and mentally ill patients) both globally and at the edge level, and find significant areas of difference. Further, we can incorporate the correlation between edges inherent in brain data by allowing for a general variance structure in the mixed effects model. We illustrate the method by analyzing data from a study comparing schizophrenics to healthy controls.
|
