Analysis of Dynamic Brain Networks using VAR Models
| Type of publication: | Inproceedings |
| Citation: | moewes_analysis_2012 |
| Year: | 2012 |
| Pages: | 525--532 |
| Crossref: | kruse_synergies_2012: |
| URL: | http://link.springer.com/chapt... |
| DOI: | 10.1007/978-3-642-33042-1_56 |
| Abstract: | In neuroscience it became popular to represent neuroimaging data from the human brain as networks. The edges of these (weighted) graphs represent a spatio-temporal similarity between paired data channels. The temporal series of graphs is commonly averaged to a weighted graph of which edge weights are eventually thresholded. Graph measures are then applied to this network to correlate them, e.g. with clinical variables. This approach has some major drawbacks we will discuss in this paper. We identify three limitations of static graphs: selecting a similarity measure, averaging over time, choosing an (arbitrary) threshold value. The latter two procedures should not be performed due to the loss of brain activity dynamics. We propose to work on series of weighted graphs to obtain time series of graph measures. We use vector autoregressive (VAR) models to facilitate a statistical analysis of the resulting time series. Machine learning techniques are used to find dependencies between VAR parameters and clinical variables. We conclude with a discussion and possible ideas for future work. |
| Keywords: | dynamic networks, elctroencephalography, neuroimaging, Regression, vector autoregressive model |
| Authors | |
| Added by: | [CM] |
| Total mark: | 0 |
|
Attachments
|
|
|
|
|
Notes
|
|
|
|
|
|
Topics
|
|
