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 (MRI) are limited in temporal resolution to time scales on the order of, at best, one second by physiological and signal-to-noise considerations. However, the importance of EEG and MEG as functional imaging modalities has been diminished by the absence of definitive studies demonstrating their ability to provide accurate spatial localization of complex sources either in vivo or in a realistic head model. The absence of such studies is due to the highly ill-posed nature of the inverse problem, and to a concentration in the EEG/MEG community on human and animal experiments where true validation is often difficult. Another factor limiting progress towards the use of EEG/MEG in studying brain activation has been the lack of inverse algorithms capable of fully utilizing the spatial and temporal information collected during sensory, motor or cognitive activation. Furthermore, recent advances in the other anatomical and functional imaging modalities offer the potential for bringing a wealth of additional information to the problem. There is therefore a clear need both for the development of new algorithms which exploit the most recent advances in sensor design, signal processing theory, and other functional and anatomical imaging modalities, and a detailed study of the limitations of these and existing inverse procedures. It is the goal of this project to develop new algorithms to exploit the full potential of EEG and MEG based source estimation. Models based on multiple dipoles and distributed current source will be developed. In both cases we will consider parametric and nonparametric models for the associated temporal activity. Spatial constraints on the source locations, based on volume MRI and functional studies (fMRI or PET), will be incorporated in these models where appropriate. The forward model will allow incorporation of realistic head geometries as well as locally fitted spheres. Starting from a general Bayesian framework incorporating all of the above, we will develop inverse procedures for a number of specific spatio-temporal source configurations. All procedures will be studied using (i) theoretical tools such as the Cramer-Rao lower bound; (ii) Monte-Carlo studies of bias, variance and robustness to modeling errors; and (iii) experimental evaluation using special phantoms including one constructed from a human skull containing multiple dipole and distributed current sources. Platform independent software and documentation for these algorithms will be distributed to interested researchers in the brain imaging community. In addition to providing a suite of thoroughly tested inverse procedures, we anticipate that this work will provide important insights into the fundamental limitations of EEG and MEG based source estimation.
Showing the most recent 10 out of 16 publications
