Functional magnetic resonance imaging (fMRI) studies are instrumental in characterizing local properties of behavior-related neural activity and for investigating regional associations in brain activity. fMRI studies produce massive data sets that pose challenges for the development and application of appropriate statistical methods. Data from fMRI studies consist of 3-D movies for each subject, contain a large number of spatial locations (voxels) within each scan, and exhibit complex patterns of spatial and temporal correlations. In this talk, we develop Bayesian modeling procedures that capture aspects of the spatial correlations between brain locations and temporal correlations between repeated measures on each subject. Incorporating these correlations borrows strength across distinct, but functionally related, brain locations and yields increased estimation precision. Our methods provide a unified framework for the distinct objectives of detecting localized alterations in task-related brain activity and determining associations between different brain regions, and they can also be used for prediction. We demonstrate the applicability of our modeling approaches using data from studies of psychiatric and neurological disorders, such as Parkinson's disease, Alzheimer's disease, and schizophrenia.
Dept of Biostatistics
biostats [at] columbia [dot] edu