I'm working with raw timeseries waveform data from a resting state dataset. There's one run per participant, so no between-run normalization is required. With Doug's help, I have mastered extracting the timeseries information from individual ROIs in surface space. The timeseries data is the product of the preproc-sess script, to which I additionally added the -inorm parameter to extract the mean timeseries for the entire volume. One concern I had was that correlations between timeseries from different ROIs might be artificially inflated by scanner drift. For example, if the overall signal increases or decreases as a function of time over the course of the run or fluctuates periodically, then for any two regions, their signals will go up or down together as a result. This should in turn make the timeseries from these two regions more correlated. I have no strong evidence that scanner drift is a particular problem in my dataset, but it seems likely enough in a 6 minute run that I want to mitigate the problem.
I can't really put these data through selxavg, as I do not have a model to fit for resting state data, and in any case would like to work with the timeseries itself, rather than residuals or any other by-product of an HRF model. I was wondering if simply subtracting the global mean signal waveform from each ROI waveform would be a reasonable strategy for removing drift, or if there are any standalone commands that are called by some of the super-scripts that I can use without carrying out a first level analysis.
Thanks, Chris