I am measuring field-field coherence between two electrode arrays (48 channels on each array). My questions regards to finding time-frequency ROIs where coherence is significantly modulated by task parameters. Due to high number of channels and trials it practically is not feasible to run randomization tests. I was wondering if there is any other common way to mark ROIs on the coherogram when you do not have any presumptions about potential time-frequencies that might be modulated by task conditions?
Hi Jafar. Is your question about the dimensionality explosion? You’ll need to reduce the dimensionality of the data somehow, or use more qualitative analyses.
You could use a “mechanical” data-reduction method by picking the one electrode from each region that shows some effect, e.g., maximum power at a frequency of interest. Then compute synchronization just between those channels.
Or you could use statistical data-reduction methods by multivariate decompositions, like ICA or GED. The idea here would be to combine the data from all channels (per region) to have one component that maximizes some criteria (the criteria depends on the method, the hypotheses, etc.).
Another way would be to pick a time-frequency region based on hypothesis or task design, and then compute the all-to-all synchronization in that window.
So, lots of options here I guess the main point is to figure out the most appropriate way to reduce the dimensionality, using some combination of statistics and hypotheses.
Hope that helps, happy to chat further about it.
Thanks for your response, It was a great help and new approach to the the question.