The magnetoencephalogram (MEG) and electroencephalogram (EEG) provide unique insights into the dynamic behavior of the human brain as they are able to follow changes in neural activity on a millisecond timescale. In comparison, the other functional imaging modalities (positron tomography (PET) and functional magnetic resonance imaging (fMRI)) are limited in temporal resolution to time scales on the order of, at best, one second by physiological and signal-to-noise consideration. The goal of this project is to develop and evaluate computational techniques for estimating the location, extent and dynamic behavior of the current sources that produce the observed MEG and EEG. During the initial project period, we have developed a suite of methods and software for head modeling, source localization and imaging. We have also built and tested a skull based phantom and developed computational tools for comparing and quantifying performance of different models and inverse methods. We plan to build on this work in the proposed project period, by concentrating on using and extending the methods we have developed to date to address several fundamental questions of relevance to both EEG/MEG researchers and the brain imaging community as a whole: (i) How reliably can E/MEG find the locations of multiple current sources in the brain? (ii) To what extent can E/MEG determine the spatial extent of distributed current sources? (iii) How accurately can we find the time series or activation sequence of these sources? (iv) How do we best process data from cognitive studies involving the differences between conditions? (v) How is E/MEG data best combined with functional MR or PET activation data? Extensions of the methods that have been developed during the initial project period will include techniques for relating dipolar and multipolar estimates to neural activity in the cerebral cortex and methods for combining these estimates with fMRI data. We will also develop a source localization methodology for processing cognitive data which allows identification and removal of the signal components common to two different test conditions. The Bayesian imaging method developed during the initial project period will be extended to utilize fMRI data in the prior. For our multipolar and imaging methods, we will also examine the impact of different head models on computation cost and accuracy. Performance of all methods will be evaluated using a range of computational, phantom and human data. Computational tools include Cramer-Rao lower bounds, Monte-Carlo methods and the use of subspace correlations. Phantom data will be generated using the realistic 32-dipole human skull phantom that was constructed during the initial project period. Human data will be based on simple motor paradigms well documented in the literature, for which MEG, EEG and fMRI data will be collected. Software and data from this project will be made available to researchers via the Internet.
Showing the most recent 10 out of 16 publications
