Spatio-temporal analysis in functional brain imaging
Author(s)
Ou, Wanmei
DownloadFull printable version (16.97Mb)
Other Contributors
Massachusetts Institute of Technology. Dept. of Electrical Engineering and Computer Science.
Advisor
Polina Golland and Matti S. Hämäläinen.
Terms of use
Metadata
Show full item recordAbstract
Localizing sources of activity from electroencephalography (EEG) and magnetoencephalography (MEG) measurements involves solving an ill-posed inverse problem, where infinitely many source distribution patterns can give rise to identical measurements. This thesis aims to improve the accuracy of source localization by incorporating spatio-temporal models into the reconstruction procedure. First, we introduce a novel method for current source estimation, which we call the l₁l₂-norm source estimator. The underlying model captures the sparseness of the active areas in space while encouraging smooth temporal dynamics. We compute the current source estimates efficiently by solving a second-order cone programming problem. By considering all time points simultaneously, we achieve accurate and stable results as confirmed by the experiments using simulated and human MEG data. Although the l₁l₂-norm estimator enables accurate source estimation, it still faces challenges when the current sources are close to each other in space. To alleviate problems caused by the limited spatial resolution of EEG/MEG measurements, we introduce a new method to incorporate information from functional magnetic resonance imaging (fMRI) into the estimation algorithm. (cont.) Whereas EEG/MEG record neural activity, fMRI reflects hemodynamic activity in the brain with high spatial resolution. We examine empirically the neurovascular coupling in simultaneously recorded MEG and diffuse optical imaging (DOI) data, which also reflects hemodynamic activity and is compatible with MEG recordings. Our results suggest that the neural activity and hemodynamic responses are aligned in space. However, the relationship between the temporal dynamics of the two types of signals is non-linear and varies from region to region. Based on these findings, we develop the fMRI-informed regional EEG/MEG source estimator (FIRE). This method is based on a generative model that encourages similar spatial patterns but allows for differences in time courses across imaging modalities. Our experiments with both Monte Carlo simulation and human fMRI-EEG/MEG data demonstrate that FIRE significantly reduces ambiguities in source localization and accurately captures the timing of activation in adjacent functional regions.
Description
Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2010. Cataloged from PDF version of thesis. Includes bibliographical references (p. 119-137).
Date issued
2010Department
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer SciencePublisher
Massachusetts Institute of Technology
Keywords
Electrical Engineering and Computer Science.