Neural electromagnetic measurements (in particular MEG and EEG) provide a number of advantages for the noninvasive characterization of neural function. These techniques measure direct physical correlates of the currents associated with neural activation, and provide temporal resolution adequate for the study of neural population responses. However, the general computational problem of source localization from such surface measurements is ill-posed and ambiguous: multiple current configurations can account for any given set of surface measurements. By employing models of neuronal current configurations that limit the space of possible inverse solutions it is possible to obtain useful source reconstructions. The reliability and accuracy of such techniques can be significantly enhanced and by employing external information from alternative imaging modalities such as anatomical and functional MRI, however the use of rigid constraints based on such data may introduce errors under some circumstances.
We have developed a probabilistic Bayesian framework for neural source localization that effectively deals with the ambiguity of the neural electromagnetic inverse problem, while providing a powerful method for integrated analyses incorporating multiple disparate sources of data. This framework provides a useful and general method for describing extended neural sources that can exploit probabilistic maps of functional architecture. This might serve as a useful construct for the development of database schemes that incorporate probabilistic descriptions of neural temporal response.
Aims of future work include:
Continued development of the theoretical framework and computational methods for integrated analysis of multi-modality image data: Bayesian probabilistic analyses hold promise for useful solutions to the MEG /EEG inverse problem, and for integration of disparate forms of image data. Probabilistic constraints on source location and spatial extent will be developed, based on functional neuroimaging data (fMRI or PET) or drawn from databases. A second area of work will be development and evaluation of detailed finite difference forward models based on MRI techniques that produce volumetric estimates of tissue conductivity, allowing assessment of the effect of anisotropy and inhomogeneity of brain conductivity on magnetic field and electric potential distributions.
Enhanced experimental and analytical methods for functional neuroimaging: Event related fMRI can employ the same experimental designs routinely used for studies with MEG and EEG. Extensions to visual stimulus generation, timing capability and analysis tools will be implemented to conduct such experiments. Relevant physiological data (cardiac cycle, respiratory cycle, blood pressure and blood oxygenation) will be collected to allow assessment of the roles of these variables in modulating the fMRI BOLD response and to correct for such effects. Investigation and development of an MR technique for movement monitoring and correction will continue.
Computational tool development: A primary goal of proposed
work is to build a consolidated package for multimodality functional neuroimaging
based on computational modules developed in our laboratory, including MRIVIEW
(a tool for structure/function correlation and MRI volume processing) and
MEGAN (a tool for the analysis of neural
electromagnetic (NEM) data). The consolidated package include a series
of electromagnetic forward calculations of increasing anatomical realism,
and will provide an extensive (and extendible) suite of inverse procedures
with proven utility and merit. A public domain version of this package
will be produced, however, avenues for commercial development and
continuing support of the software will also be pursued.
Brain Imaging And Modeling
Human Brain Project
Bayesian
Inference
MEGAN
NetMEG File
MRIVIEW