3dNetCorr warning message: FATCAT

Hi everyone,

I was running a command 3dNetCorr,
And I receive this warning message:
*+ WARNING: THD_Tcorr1D: 16 voxels skipped because were constant in time

I have warning messages for as much as 35 voxels skipped. When I tried to add a mask as an argument to the command 3dNetCorr it actually resulted in more of these warning errors.

For me to resolve the issue, I have to first fully understand what it means.
Now my question is: I’m not fully understanding this warning error.

Is it that the ROI voxels are overlayed on top of functional data but there are some ROI voxels are that don’t have corresponding functional values?
Is it that the two masks: The ROI and the functional data are not being overlayed properly?

And if so, what do you suggest I do?

Also how many voxels skipped because it was constant in time is it considered to be okay? and how many voxels to be constant in time would you say is problematic?

Thanks for your time.

This is the code if it helps:

for f in "${allfiles[@]}";   3dNetCorr -inset $f -in_rois /Users/network6_thresh7.nii.gz -fish_z  -ts_wb_corr -prefix $f ;  done

Hi, Sondos-

The issue with “constant” time series when calculating correlation is: the correlation value includes the variance of each time series as a term in the denominator. The variance of a constant time series is zero-- dividing by zero is bad mathemetically. So, having a constant time series is skipped.

Typically, a constant time series occurs in rbain imaging if you are applying a mask to a time series dset and there are voxels that were previously zeroed out inside the mask. Or, if you have a non-fullbrain coverage for the EPI dset, but you are using a mask derived in another way and again “all zero” voxels are included in the mask. Other examples can occur as well. So, you should check and see whether you have whole ROIs that only contain voxels of allzeros, or if you are calculating whole brain correlation maps from each ROI in your inputs (which may be the case, because you have the “-ts_wb_corr” flag on), then just having zero-time series won’t matter too much.