Gang:
Thanks for the quick response.
For clarification I have a few questions:
1. Are you saying that after normalizing the time series data with (x/Mean ) * 100 and using this as input into 3dDeconvolve, that the regressor coefficients (i.e., previously fit coefficient) will now express signal magnitude in units of percent signal change? The t-statistic associated with each % signal change will still serve as a statistic that we can use as a threshold.
2. 3dClipLevel only works with high-resolution anatomical images, and not with functional data. Correct? However . . .
3. 3dAutomask documentation indicates that the input dataset is EPI 3D+time and that it runs 3dClipLevel. Our data was acquired using a spiral pulse sequence. And this seems to suggest the 3dClipLevel can operate on a functional time series?
4. One of our goals in converting the functional time series data into % signal change is to compare the signal magnitude from run2 to run 1 at selective groupings of TRs down the time domain. Thus, do we not need to go further than normalization ( (x / Mean) * 100) and actually mask the brain and convert raw MR signal into % signal change with (x – Mean / Mean ) * 100)???
5. We did find that using (x – Mean / Mean ) * 100) at every voxel with 3dcalc followed by 3dDetrend –expr 1 –expr “t” to remove the mean and linear trend in the % signal change and this seems to work well.
philippe