Confidence Envelopes for Population-Level Activations in fMRI Studies

• Activity Number: 215
• Type: Invited
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): David Degras*+ and Martin Lindquist
• Companies: DePaul University and Johns Hopkins Bloomberg School of Public Health
• Keywords: fMRI ; Activation study ; Spatio-temporal model ; Confidence envelope

Abstract

In the analysis of fMRI data, activation maps (i.e. spatial maps of normalized contrasts, p-values, or posterior probabilities) provide important insights into group or population activation patterns. However, they do not show the variability of activations across individuals. In this paper we develop a hierarchical, spatio-temporal model for fMRI data and build confidence envelopes for population activations. The envelopes reveal the uncertainty about where population activations occur; they can also be used to predict individual activations. Our approach builds on factor models and describes the correlation in the BOLD signal in terms of brain function and of neuroanatomy. The estimation procedure borrows information across subjects and voxels in a computationally effective way. Confidence envelopes are obtained either by Gaussian Random Field simulation or by bootstrap methods. They are backed with theory and their coverage level is examined in simulations. The proposed methodology is also applied to real data.