Hi, Carolin-
Just to zoom out for a second:
A) Are you wanting to characterize the spikiness in an ROI at the start of processing, before anything has been done to the time series?
B) Also, do you want your spikiness measure to reflect temporal associations of spikes? That is, say there are 10 voxels in the left-hippocampus and you have 100 time points. Do you care if 3dDespike finds 10 spikes scattered around at different voxels at different time points, or do you want to know if those 10 spikes occur just within a small number of voxels (so, spatially associated), or do you care if those 10 spikes occurred all in a single time point (temporal association)?
- my guess is that from a QC point of view, having temporal synchronization is the main feature, but that is just a guess---that might be the sign of a motion event or dropout or something, which would suggest that time point's data might be compromised.
C) Note that very noisy/problematic time series could have very low spikiness because the variability is so high---so just using spikiness as a measure might mask the problem. So actually, a low variability time series might be susceptible to finding some high spikiness, not because of badness but just because baseline variability is so low. That is why we often use cross-space and within-time checks for badness with outlier fraction count at a given time point---that seems to me to be the most useful role for spike-checking.
D) How do you plan to use an omnibus spikiness measure to QC your hippocampal data? Do you want to possibly censor out bad voxels, or do you want to possibly censor out bad time points of data, or do you want to know whether a subject can be used at all?
--pt