For an fMRI project, I am trying to calculate the sample size and the power via powermap on MATLAB. It requires me to input smoothness values for x, y, z in mm. I wonder what these values are in fmriprep preprocessing pipeline in default (without adding further analysis such as ICA-AROMA).
fMRIPrep does not apply any explicit smoothness into your data (with the only exception of the ICA-AROMA non-aggressively denoised outputs). That is clearly stated in the documentation at several instances. Please have a detailed look into https://fmriprep.org
For your power computation, I think you will need to use some smoothness estimation. In our Nat Meth. paper (preprint) we used AFNI’s
3dFWHMx to estimate the intrinsic smoothness of the data after preprocessing and compare it to corresponding data preprocessed with FSL feat (fig. 6A of the preprint, fig 3a of the published paper).
Thank you for your response. As you said, I have critically read your paper in which you compared FSL’s and fmriprep’s output. However, there is an issue that confuses me, what do you really mean when you say explicit vs implicit smoothing? How could implicit smoothing occur in my data?
Every measurement device has a point-spread function (PSF), the smoothness (or spread) of this function is implicit to your settings. Considering the acquisition + all the preprocessing you run on your data as “the device”, then at the end of the pipeline you’ll have an implicit PSF that is characteristic of the ensemble.
When you explicitly apply some smoothing filtering to your data, besides the “denoising” you typically pursue, what you are also doing is modifying the intrinsic PSF of your data so that it meets certain assumptions you want to impose.
This is why some smoothing tools (e.g., AFNI’s tool for this) will ask you the width of the kernel you want to end up with, not the kernel you want to apply.
Oh, I see. Thank you very much for your time and detailed explanation.