Hi,
This is not a question on AFNI but I thought that this was a good place to get some insight into.
I came across this question after looking at the source code for simuating artificial datasets from software such as neuRosim (in R). Here, datasets are simulated from a fMRI time series using a given design matrix and SNR. My question has to do with this SNR calculation. From what i see of the neuRosim code, the software calculates SNR by using the mean of the time series at a voxel and then dividing that by the standard deviation (not of the time series at the voxel) but by the marginal standard deviation of the error in the autoregressive time series. But in so doing, it ignores the fact that there is correlation in the time series and thus, the SNR is actually inflated. So, my question is on how the SNR in fMRI time series data is calculated in general.
I guess my point is that we need to have an accurate model for simulating fMRI simulation data which faithfully represents the difficulty (both CNR and SNR) of the fMRI problem in order to calibrate different methods and I think that accounting for the correlation in the calculation of the SNR for these simulated scenarios provides a more realistic assessment of fMRI data processing scenarios.
Any thoughts?
Thanks again!
Ranjan
