The Effect of Model Misspecification in Functional Connectivity

• Activity Number: 541
• Type: Invited
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Gina D'Angelo*+
• Companies: Washington University School of Medicine
• Keywords: fMRI ; brain regional correlations ; GEE ; time-series ; temporal dependence

Abstract

Resting-state fMRI studies are useful to evaluate regional connections of the brain and to distinguish functional organizational differences between groups for various diseases such as Alzheimer's disease, schizophrenia, and depression. When the objective is to compare group-wise regional correlations, many approaches for analyzing resting-state fMRI studies are subject to the following issues related to model misspecification: 1) the temporal correlation of fMRI data is ignored, and 2) the function of time and/or the model are misspecified. We discuss various model-based approaches such as generalized estimating equations (GEE), mixed models, and vector autoregressive (VAR) models to address the temporal correlation and demonstrate how robust they are to the misspecification of the time effects. These GEE, mixed models, and VAR models are compared to the standard pooling approach using simulation studies to demonstrate the effect of model misspecification. In addition, these methods and the effects of model misspecification are demonstrated using an Alzheimer's disease functional connectivity study.