I came across the concept of Granger causality recently and was fascinated by the concept of it, it is capable of finding the correlation between two different time series signal. Now i think that maybe this could be used in EEG processing, like there's this famous concept of alpha asymmetry between the 2 hemispheres of the brain, so maybe could this method be a good measure of that. And also, can this help in making new features for 2 or more individual correlated data in things like training the dataset.on some machine learning model.
So what are your thoughts on this concept, and am i understanding the concept of Granger causality properly?
Granger causality is indeed an intersting concept, but this is also a problem, it is a concept and not a special methodology or algorithm. You have to know what you want to show and how to apply it.
I am not sure if it is applicable on classical resting state alpha-asymmetry research. If you are thinking about the classic reasearch like approach vs avoidance or BIS/BAS (Davidson, Coan, Allen, Harmon-Jones, Hagemann, Hewig and other authors), this wont work. The EEG data is reduced to the frequency domain and no time-series is left. In my opinion you need a stimulus and the whole TF-information between different electrodes to analyse the "information flow".
Granger causality is a mathematical abstract for linking two events together. Basically, it is appropriate for stationary data; however EEG is a non-stationary signal. We begin with the hypothesis that for a smaller event it could be stationary.
Thus, the biggest limitation is that it could not be applied for continuous EEG data but for the subset of the event. There are a lot of algorithms used for detecting fatigue, depression, etc using a similar concept.
So, my suggestion would be that if you begin with frontal asymmetry or cerebral asymmetry then you should narrow down the approaches of granger causality such as PDC or DTF.
- Badges
- Science topic
- Similar topics
- Cognitive Science
- Thinking