A framework for group analysis of fMRI data using dynamic Bayesian networks

Annu Int Conf IEEE Eng Med Biol Soc. 2007:2007:5992-5. doi: 10.1109/IEMBS.2007.4353713.

Abstract

FMRI experiments are usually performed to make inferences about groups of subjects, but current group analysis methods for dynamic Bayesian networks (DBNs) do not easily allow incorporation of covariates of interest. In this paper, we propose a group-analysis method which uses multivariate analysis of variance (MANOVA) to address this issue. The method is performed in two stages: first, deriving a DBN connectivity network among brain regions for each subject separately; second, regressing the connectivity coefficients of DBNs to the factors of interest and performing MANOVA. A case study involving fMRI data from Parkinson's disease (PD) subjects yields promising results. Ten out of the thirteen potential connections between Regions of Interest (ROIs) which are associated with disease state are functionally improved after medication (Table I), consistent with clinical observations. The results confirm that improvement in PD symptoms after medications is in part mediated by enhanced functional brain connectivity between brain regions.

Publication types

  • Evaluation Study

MeSH terms

  • Artificial Intelligence
  • Bayes Theorem
  • Brain / physiopathology*
  • Brain Mapping / methods*
  • Humans
  • Image Interpretation, Computer-Assisted / methods*
  • Magnetic Resonance Imaging / methods*
  • Nerve Net / physiopathology*
  • Parkinson Disease / diagnosis*
  • Parkinson Disease / physiopathology*
  • Pattern Recognition, Automated / methods
  • Reproducibility of Results
  • Sensitivity and Specificity