Causality analysis is an approach to time series analysis that is being used increasingly to investigate neuroimaging data. The reason for its popularity is the useful perspective it provides in describing the ordered operations of various brain regions using indirectly and passively measured neurophysiological signals. Although there are numerous frameworks with which causality analysis can be performed, one concept in particular – termed Granger causality (GC) – is receiving much of the attention because of its ease of implementation and interpretability. GC makes use of the fact that a predictive relationship between the history of one signal and the future of another signal provides evidence for there being a causal relationship between the two signals, and as a result, the physical events underlying those signals. If such a relationship can be established across neural time series, causal dependencies between neural pathways can be inferred and their contribution to brain function can be studied. Several analysis frameworks exist for applying GC to neurophysiological questions but many of these frameworks have deficiencies that impede their application to large and highly multivariate neuroimaging datasets. To address some of these concerns, this study develops the theory and methods for a novel neural time series classification procedure – referred to as GC classification – based on concepts in GC analysis. Validation of this method in neuroimaging research is provided by showing that it can be applied to heterogeneous datasets, that it makes use of many parallel sources of information about causal relationships, and that it can be adapted to different types of preprocessing steps to uncover causal relationships in multivariate neural time series data. Application of this analysis method to human behavioural MEG data revealed that, during a cued button-pressing task, distinct causal relationships exist between sensory cortices and their downstream targets preceding the initiation of actions that differ by whether or not they were the result of a decision being made. These results provide evidence that the GC classification procedure is a useful and robust technique for studying causal relationships in neurophysiological time series.
Identifer | oai:union.ndltd.org:TORONTO/oai:tspace.library.utoronto.ca:1807/43619 |
Date | 10 January 2014 |
Creators | Kostelecki, Wojciech |
Contributors | Perez Velazquez, Jose Luis |
Source Sets | University of Toronto |
Language | en_ca |
Detected Language | English |
Type | Thesis |
Page generated in 0.0095 seconds