Dear João,
> I'm running a VBM analysis with SPM8 and VBM8 toolbox.
>
> I have a couple of questions:
>
> - Is it better to report the voxel-level or cluster-level p-values?
It's not that one is "better"—they are telling you about different
levels of inference. For the voxel-level correction, it tells you
that every voxel you see is unlikely to have occurred by chance (and
thus, you could make some inference about that voxel). If cluster
level, it tells you that a particular cluster is bigger than you would
expect by chance. This entitles you to make some inference about the
cluster as a whole, but not subparts (which can get tricky if a large
cluster covers multiple regions of the brain). There are several
other emails in the archives that go into more detail.
> - If I use the "Results" button in SPM8, does it use the non-stationary
> cluster extent correction to calculate valid cluster p-values? Is the method
> of Satoru Hayasaka et al. 2004 implemented in SPM8 or should I use an
> external toolbox, like the NS extension by Hayasaka et al
> (http://fmri.wfubmc.edu/cms/NS-General) ?
The "results" button in SPM does not correct for non-stationarity
which is present in VBM images. You'll need to use the ns toolbox.
Unfortunately, as far as I know the ns toolbox is not available for
SPM8 (nor will it be), which means using SPM5 for looking at your
results. As far as I'm aware this should not present a problem, as
the statistical machinery has not changed appreciably (but I would be
interested to hear if I'm wrong about that!).
I don't know whether the VBM8 toolbox includes anything for
non-stationarity correction—it would be worth checking.
> - If I choose "none" for the "p value adjustment for control" it will aply
> the following threshold to each voxel, am I correct? But even if I choose
> "none" correction I still get the FWE-corrected cluster-level and I get
> FWE-corrected cluster-level p-values... why is this?
You are correct—if you select "none", it will apply that threshold at
the voxel level. You can think of this as defining features
(clusters) in your data. You have not corrected for multiple
comparisons at the voxel level (because you selected "none"), but you
can still correct for multiple comparisons at the cluster level.
Having defined your clusters using some threshold, SPM will therefore
give you the corrected p values for each cluster. These are perfectly
valid, provided that (a) you understand the level of inference you're
working at (see above) and (b) in the case of VBM, you have accounted
for non-stationarity in the data.
This previous message may also be of some help:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;8b567b91.1107
Hope this helps!
Best regards,
Jonathan
--
Dr. Jonathan Peelle
Department of Neurology
University of Pennsylvania
3 West Gates
3400 Spruce Street
Philadelphia, PA 19104
USA
http://jonathanpeelle.net/
|