Dear freesurfers,
How is spatial smoothing performed during a conventional fmri processing stream?
I used SPM to (1) run the 1st level analysis on each subject, (2) transformed the resulting volumetric statistical parametric map to the surface space, (3) normalized each individual's surface map onto fsaverage (which involves surface-based smoothing), and then (4) ran 2nd level analysis on these data.
The question is:
3D-spatial smoothing is normally done (and I believe recommended) at step (1) above when estimating the GLM at the 1st level. However, performing 3D smoothing on raw functional images seems to be detrimental if the data will eventually be projected on to a surface space.
I then thought about NOT 3D-smoothing at step (1), that is, estimating GLM on unsmoothed data, and then using 2D smoothing after step (2) when statistical parametric maps are projected onto the surface space. However, this sounds problematic since smoothing followed by model estimation might give different results from model estimation followed by smoothing.
The third approach is to convert all the volumetric functional images into surface space, and then start model estimation in the surface space using FreeSurfer. But this means that I won't be able to use SPM or other packages to run model estimation (that might have superior functionality when it comes to fmri data analysis). Is this what's usually done to integrate fmri data with freesurfer procedures?
Or, alternatively, is there a better integration protocol between FSL and FS (I don't know much about FSL) to address this issue?
I'd be grateful to know how other people are doing this. Thank you!
Joon