Hi,
On 27 Mar 2009, at 23:58, Lin Nga wrote:
> Hi all,
>
> I have a few questions regarding the thresholding in FEAT:
>
> 1) I understand that with cluster thresholding, the voxels are first
> thresholded by the Z threshold and then the resulting clusters are
> tested to
> meet the Cluster P threshold.
>
> What I want to know is, how do I find the p-value corresponding to
> the Z
> threshold? Do I simply use the ztop (and ptoz for the reverse)? Just
> to
> verify, this function calculates the p-value for that Z-score and
> NOT the
> Cluster P threshold. Is that correct?
You mean the uncorrected p-value corresponding to the Z threshold?
Yes, that's right, use ztop.
> Also, if I'm conducting a whole brain analysis, I would want to use
> the #of
> resels given at the bottom of the smoothness file under each COPE's
> stats
> folder and not the resel count (# of resel/volume ?) for the ptoz
> function.
> Am I misunderstanding this?
I'm not following you - do you mean how are the cluster corrected p-
values calculated? That's not so simple as calculating the numbers of
resels directly - it involves Gaussian random field theory - see the
literature on that for more.....
> 2) On a similar note but a different analysis, I have a set of
> localizer
> data to find the FFA and I was hoping to get some advice on the
> threshold
> settings. I am defining the FFA as a sphere (of 8mm radius) centered
> on the
> most significantly activated voxel (within a certain range in the
> x,y,z
> directions) from the contrast ( Faces - Items ). At the default
> settings for
> the cluster thresholding ( Z threshold = 2.3 and Cluster P threshold =
> 0.05), I am only able to find FFAs in 31 out of the 47 participants.
> I was
> hoping that I would be able to localize more FFAs with less stringent
> thresholds.
>
> I tried the no thresholding option under the Post-stats tab of FEAT
> but the
> Post-stats report was blank except the the Post-stats title. Is this
> supposed to happen?
If you don't do thresholding you don't get thresholded outputs....so
yes that's correct.
> I was considering Voxel thresholding but I need coordinates for the
> most
> significant voxel; not sure how I would find it if it's not reported
> in the
> results.
You can use the "cluster" program, or more simply if you just want the
peak within an ROI you can use Featquery or fslstats. If you have a
standard space liberal ROI for FFA then the easiest way to go is via
Featquery because that handles inverting the standard space
registration transforms for you.
In answer to the questions below - if you are just trying to find the
peak voxel and NOT prove that it is significant then there's not
necessarily any advantage in using cluster thresholding. Just follow
the above, for example - do voxelwise thresholding and find the peak
within some ROI.
Cheers.
> Since the FFA is a rather small area, I'm thinking that I should
> lower the Z
> threshold and increase the Cluster P threshold but from previous
> posts, it
> sounds like lowering the Z threshold below 2.0 isn't a good idea.
> I'm new to
> GRF theory so I don't quiet understand why that is yet but from the
> ztop
> function, a Z threshold of 1.65 corresponds to a p-value of 0.049471.
> Doesn't that mean a Z threshold of 1.65 would still give me
> significant results?
>
> Lastly, I am planning to re-run the post-stats with Z threshold=2.3
> and
> Cluster P threshold=1. Does that sound like a good idea? Does anyone
> have
> any suggestions?
>
> Thanks in advance everyone!
>
> Have a great weekend,
> Lin
>
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|