3dDeconvolve fitts output


I work with rodent fMRI and when using 3dDeconvolve, I use a custom-made stimulus file as a regressor/model. Also, I normalize each rodent’s dataset to the baseline before inputting it into 3dDeconvolve. However, when I plot the -fitts output file, I notice that the fit’ed model doesn’t start at 0. I am comparing multiple conditions and was wondering if this baseline shift of the fitt’ed model can cause miscalculations when doing a 2 sample t-test?

I have attached 2 images. The first is the raw data (blue) with the -fitt output… showing good fit… however you can see the fit time series baseline is above 0. The second image is comparing the 3 conditions’ fit time series… all start above baseline and it appears that the blue SHOULD have a significantly higher amplitude… but due to the shift, it seems ANOVA isn’t finding significance during 2 sample comparisons…

Is there a way to fix this? I thought perhaps normalizing the fit time series to baseline and inputting these again back into 3dDeconvolve.

Thanks for any assistance



It is hard to say much about this without knowing more.

Would you please include the 3dDeconvolve command?
Also, can you verify that the means of your regressors is 0? For example, what is the output from:

3dTstat -mean -prefix - X.xmat.1D\'
  • rick


Thank you very much for replying and sorry for the late reply. I had thought no one could answer.

3dDeconvolve -input m1.nii -nfirst 0 -polort 0 -mask brain_mask.nii
-fitts fit_m1.nii -errts residual_m1.nii -num_stimts 1 -stim_file 1 $stim -stim_label 1 Visual -x1D X_m1.1D -xjpeg X_m1.jpg
-tout -bucket stats_m1.nii.gz;

Here is what I run with a representative mouse 1 dataset inputted. This is a spatially template registered image. I pre-detrend hence why I use polort 0. (I understand detrending here is better, but I found it to have negligible effect on mouse data).

The stim file is a 1D file of 0 baseline with the shape of the model represented from a range of 0-1. We’ve tried other model methods in 3dDeconvolve, but the rodent HRF shape is vastly different from humans that no model worked well from what I’ve tried.

The output from this mouse results in 1 and 0.135185. It seems the mean of the regressors is not 0?

Thank you for your help