Dear Linda,
> I have analyzed fMRI data from a number of subjects and I'm about to look at
> the activity maps. I've read in a number of papers that a common
> thresholding approach is to first apply a height threshold of p = 0.001
> (uncorrected) and then apply a cluster extent threshold of 10 voxels. Is
> this approach the most common in fMRI or is there any other approach that is
> more common?
Just to expand on what Sasha said (all of which I agree with):
fMRI data are always spatially correlated, which means that you are
likely to get a certain size cluster just by chance. How big this
cluster is really depends on the properties of your actual data. So
for some data, 10 voxels might be a totally appropriate extent
threshold, whereas for others it may need to be 500 voxels to be
statistically significant. Using a principled method of correction
for multiple comparisons (i.e. using the cluster-level p values,
either FWE or FDR) deals with this by looking at the properties of
your actual data set.
(Of course, you can also use a corrected threshold at the voxel level,
but this is typically less sensitive.)
For more I would suggest the following article:
Bennett CM, Wolford GL, Miller MB (2009) The principled control of
false positives in neuroimaging. Social Cognitive and Affective
Neuroscience 4:417-422.
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/
|