Often spatial smoothing is performed before voxel-wise statistical
analysis:
Registered image volumes
|
| Spatial Smooth
|
V
Smoothed, registered image volumes
|
| Statistical Analysis
|
V
Smooth Z map
There's evidence (Lowe MJ & Sorenson JA, Magn Reson Med. 1997 37:723-9)
that the spatial smoothing can be deferred until after statistical
analysis, in which case one would instead smooth the Z map itself:
Registered image volumes
|
| Statistical Analysis
|
V
Nonsmoothed Z map
|
| Spatial Smooth
|
V
Smooth Z map
The variance of the resulting smoothed Z map would then need to be
adjusted (Worsley et al., Human Brain Mapping 4:74-90).
This would save computations, since instead of spatially smoothing all
fMRI image volumes one would need to spatially smooth only the one Z
map. It might also save disk space if one were saving the smoothed fMRI
image volumes to the hard drive.
Has anybody in the SPM community tried and confirmed this? Is there
some disadvantage? Is this perhaps an option somewhere in SPM99? I
couldn't find a previous thread in the archives discussing this.
Thanks,
Joe Maisog.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|