Dear Feng,
Non-stationarity issues are usually considered only in the context of VBM, but the stationarity assumption does not really hold for fMRI either, see a conference contribution by Worsley (2002, http://www.math.mcgill.ca/keith/HBM2002/HBM2002_poster2.ppt ), although violations might be less severe compared to VBM data (see Fig. 1 in Hayasaka et al., 2004, Neuroimage). So it's never bad to consider non-stationarity when looking at fMRI data.
Concerning small volume corrections, actually it's quite simple, but there's a discrepancy between theory and practice. The idea is to look at a-priori regions with a threshold that takes into account the volume / resel count of these regions, not the volume / resel count of the whole brain (which should result in a more liberal threshold for a-priori regions). Thus, if you have 10 a-priori regions you should combine these into a single mask file and use this file for small volume correction. Afterwards, you would go with a more conservative threshold based on the whole brain to test whether there are any additional, potentially interesting findings in non-a-priori regions. In practice, people go the opposite, they look at whole brain results, find some sig. results for some regions and some non-sig. results for others, start to think about whether this "noise" could be interesting, if it could it becomes an a-priori region, and then the SVC is conducted just for that particular region. This is ridiculous, but well.
Best
Helmut
|