FEAT does not transform the FMRI data directly into standard space but carries all statistics out in the original (low resolution) space and then transforms the final statistics images into standard space. The transformation from original space into standard space is normally carried out (automatically) in a two-step process; first an example functional image (the one which was used as the reference in the motion correction) is registered to the subject's structural image (normally a T1-weighted image which has been brain-extracted using BET [20]) and then the structural image is registered to a standard space template (normally the MNI152). The two resulting transformations are concatenated resulting in a single transform which takes the low resolution statistic images into standard space. This is the default FEAT registration procedure, and is what was used for the analyses presented above.
We investigated whether, for this data, FEAT's two-step process (using FLIRT) is an improvement over registering the example functional image directly into standard space (using FLIRT). The two-step registration resulted in a slight decrease in cross-session fixed and random effects overall variance (by approximately 3%). The number of activated voxels in general stayed the same, but the peak Z statistic improved (again by approximately 3%) when two-step registration was used, and the activation appeared qualitatively to contain more structural detail (i.e., was less blurred). The conclusion therefore is that, even for this within-subject across-session analysis, the two-step registration approach was of value in the FEAT analyses.