Some background information: The default 8 mm for "source image smoothing" are applied to equalize the smoothness of your data and the template (the human T1 template has been smoothed by 8 mm). As the template is stored as a smoothed version no additional smoothing is applied (template image smoothing 0 mm).
Looking at the Karolinska rat template, the template seems to be much smoother than your raw data. In the methods paper (Schweinhardt et al., 2003, J Neurosci Methods) they mention a smoothing kernel of 0.8 mm^3 during template generation ("Finally, the mean MR rat brain template was spatially smoothed using a filter with a full width half maximum (FWHM) of 0.8 mm^3"). This would correspond to 8 mm^3 for SPM (as 1 mm in the final template corresponds to 0.1 mm "in real life"). Actually it's 8 mm in each of the three directions (smooth the canonical_T2 with 8 mm and it looks very similar to the template, smooth with 2 mm and it's much rougher).
In short, you might go with the default SPM settings, except the bounding box. For the smoothing step of the functional data (after normalisation) you typically want to use a kernel two or three times the voxel size (human raw fMRI data ~ 3 x 3 x 3 mm^3, smoothing ~ 8 mm). Animal data is a little tricky as voxels might be far from isotropic, with thick slices on the one hand and high resolution in-plane on the other. You might still consider an isotropic smoothing kernel, two or three times the in-plane resolution, which would effectively result in no smoothing along the third dimension, or go with a non-isotropic kernel.
Finally, there might be some more information provided with the rat templates. There are some additional m files and others to replace SPM's default files. However, you would have to try with SPM99 then as they are not compatible with later versions.
|