Dear JL,
Many voxels in FA do not have a gaussian distribution at the group level. Smoothing "force" or "help" having a more gaussian distribution but it is not perfect (I recommend Jones et al. NeuroImage (2005) vol. 26 (2) pp. 546-54). Therefore, it is better to use non parametric statistics like snpm (Nichols and Holmes. Human Brain Mapping (2002) vol. 15 (1) pp. 1-25), XBAM (http://www.brainmap.co.uk/xbam.htm) or Tract based statistics from FSL (Smith et al. NeuroImage (2006)vol. 31 (4) pp. 1487-505).
Smoothing also help to reduce the inter-individual variability as no brain normalization algorithm is perfect. Hence it is recommended to use a smoothing according to the level of details you want to interpret (ie: if you smooth at 8*8*8, theoretically you can only make inference on significant cluster >512 voxels, 4*4*4 --> > 64 voxels).
hope it helps !
cheers,
michel
On Aug 14, 2010, at 9:04 AM, Jung L Hsu wrote:
> Dear SPM experts,
> I wounder if I can use spm two group t-test statistic inference in such a way. I perform DTI analysis and I have 2 groups' FA images. I try to compare their difference. If I did not perform spatial smoothing, could I still use spm statistic inference ? Should I still need to perform spatial smoothing in data image even I use snpm analysis? than is, no smoothing data can not enter into snpm analysi ? thanks for suggestion
> best
> JL
Michel Thiebaut de Schotten, PhD
NATBRAINLAB
ANR-CAFORPFC/ANR-HMTC
CRICM-INSERM UMRS 975
Pavillon de l'Enfance et l'Adolescence
47 Bd de l'Hôpital
75651 Paris Cedex 13, France
www.natbrainlab.com
+33 613579133
|