Hi
Two things:
(i) you should be running non-parametric randomisation on the DR output (as is done by the DR script) for best (most trustworthy) results.
(ii) what you observe is quite possible. See http://imaging.mrc-cbu.cam.ac.uk/imaging/PrinciplesRandomFields for an in-depth explanation. In a nutshell: GRF cluster size inference compares the size of any actual post-threshold cluster against the size of what you'd expect from thresholding a smooth random field (i.e. no signal, just noise). At a low threshold you expect to see larger clusters to survive while at high thresholds you expect that only small clusters survive by chance. Your actual DLPFC cluster might be large wrt what you would expect to happen by chance if the inital cluster forming threshold is set high (i.e. it might have a low probability of appearing by chance and therefore be labelled significant) but it might be too small to still be considered large relative to what you'd expect to happen by chance at a low cluster forming threhsold. In this case it survives significance testing at Z=2.3 but fails at z=1.9
hth
Christian
On 20 Aug 2012, at 19:13, Johnson Hampson <[log in to unmask]> wrote:
> Hello experts
> I ran a FSL dual regression melodic ICA analysis to get FEAT copes and Varcopes. Now when I run higher level feat analysis, I results are really strange... My cluster level threshold was z=2.3 and P=0.05 and I saw a nice dorsolateral prefrontal cortex at p<0.05 (only one cluster). But when I reduce my threshold to z=1.9 and P=0.05, I cannot see this same DLPFC region, but I see other clusters in the prefrontal cortex. Can anyone explain why this might be?
> How can I see a region at higher thresholds (z=2.3) but not at lower thresholds (z=1.9)??
> (my Feat version is v5.98)
> Thank you for any help!
> John
|