The goal of this proposal is to develop an integrated probabilistic approach to functional brain imaging, using MEG and fMRI. The mathematical basis for the approach is Bayesian Inference, which provides a formal robust framework for the integration of multiple imaging modalities into a common spatio-temporal of brain activity. The rationale for integrating electromagnetic and hemodynamic measures is that no single modality provides the needed spatial and temporal resolution; however, MEG and fMRI have complementary strengths ( and weaknesses) that can be exploited in an integrated analysis. The proposed work builds on our recently-developed Bayesian approach to the electromagnetic inverse problem [Schmidt et al., 1997]. In contrast to other approaches to the neural electromagnetic (EEG/MEG) inverse problem, the result of our approach in not a single """"""""best"""""""" estimate of brain activity according to some criterion. Instead, the approach yields estimates of the full probability distribution for major parameters of interest (e.g., the number, location, and size of active regions). Here we propose to extend this approach in two directions: (1) to develop a full spatio-temporal Bayesian analysis for MEG that yields probabilistic estimates of time-course in addition to number, location, and size of active regions; and (2) to develop a fully integrated analysis of fMRI and MEG data by extending our Bayesian probabilistic framework to include event-related fMRI. The result will be a very general hemodynamic response probability model, which connects fMRI and MEG data to neural currents in a single unified analysis. In contrast to other approaches to the integration of electromagnetic (EEG/MEG) and hemodynamic (PET/fMRI) measures,t he proposed approach does not simply constrain EEG/MEG inverse solutions using location estimates derived from PET/fMRI. Instead, we exploit the Bayesian approach to combine multiple sources of prior information (e.g., brain anatomy from MRI) with data from multiple imaging modalities to yield a composite estimate of brain activity. The capabilities of this approach will be evaluated with simulated data, empirical MEG data from a head phantom, and empirical fMRI and MEG data from evoked response experiments.
