Since its inception, functional neuroimaging has focused on identifying sources of neural activity. Recently, interest has turned to the analysis of connectivity between neural sources in dynamic brain networks. This new interest calls for the development of appropriate investigative techniques. A problem occurs in connectivity studies when the differing networks of individually analyzed subjects must be reconciled. One solution, the estimation of group models, has become common in fMRI, but is largely untried with electromagnetic data. Additionally, the assumption of stationarity has crept into the field, precluding the analysis of dynamic systems. Group extensions are applied to the sparse irMxNE localizer of MNE-Python. Spectral estimation requires individual source trials, and a multivariate multiple regression procedure is established to accomplish this based on the irMxNE output. A program based on the Fieldtrip software is created to estimate conditional Granger causality spectra in the time-frequency domain based on these trials. End-to-end simulations support the correctness of the pipeline with single and multiple subjects. Group-irMxNE makes no attempt to generalize a solution between subjects with clearly distinct patterns of source connectivity, but shows signs of doing so when subjects patterns of activity are similar. The pipeline is applied to MEG data from the facial emotion protocol in an attempt to validate the Adolphs model. Both irMxNE and Group-irMxNE place numerous sources during post-stimulus periods of high evoked power but neglect those of low power. This identifies a conflict between power-based localizations and information-centric processing models. It is also noted that neural processing is more diffuse than the neatly specified Adolphs model indicates. Individual and group results generally support early processing in the occipital, parietal, and temporal regions, but later stage frontal localizations are missing. The morphing of individual subjects' brain topology to a common source-space is currently inoperable in MNE. MEG data is therefore co-registered directly onto an average brain, resulting in loss of accuracy. For this as well as reasons related to uneven power and computational limitations, the early stages of the Adolphs model are only generally validated. Encouraging results indicate that actual non-stationary group connectivity estimates are produced however.
Level of Degree
First Committee Member (Chair)
Second Committee Member
Third Committee Member
Magnetoencephalography Functional Connectivity Non-parameteric Granger Causality Group Models
Lysne, Per. "Development of a Group Dynamic Functional Connectivity Pipeline for Magnetoencephalography Data and its Application to the Human Face Processing Network." (2016). https://digitalrepository.unm.edu/psy_etds/85