At 10:39 AM 2/22/2005 -0800, Matthew Brett wrote:
>Thanks very much for this - it's very helpful. Do you have a view on
>the extent of the problem if you simply assume that the image is
>stationary, at 8mm, 12mm smoothing?
I happened to have a VBM data set, smoothed with 12mm. I took a look at the
RPV image, and I can tell that my data set is not stationary. The smoothest
spot has FWHM about 30mm and the least smooth spot has FWHM about 12mm. So
I don't think a heavy smoothing induces stationarity, and simply assuming
stationarity could lead to biased results in a cluster size test. The
p-values are underestimated for clusters in smooth areas, and are
overestimated for clusters in rough areas. You are more likely to detect
clusters in smooth areas than the ones in rough areas.
A voxel-height test, on the other hand, seems to be robust to
non-stationarity. But if it is to be used in an SVC context, I suggest
re-calculating the average smoothness in an ROI, just in case that the ROI
happens to be in a smooth / rough spot.
-Satoru
Satoru Hayasaka ==============================================
Post-Doctoral Fellow, MR Unit, UCSF / VA Medical Center
Email: shayasak_at_itsa_dot_ucsf_dot_edu Phone:(415) 221-4810 x4237
Homepage: http://www.sph.umich.edu/~hayasaka
==============================================================
|