On Wed, 28 Feb 2001 Hernando Ombao wrote
>
> How is the voxel corrected p-value computed in SPM? The formula for the
> p-value obtained from Worsley, et al (92) paper is approximately
>
> Resel * {(4 ln(2))^(3/2)} * {(2*pi)^(-2)} * (t^2 -1) * exp( - t^2/2).
>
> The Resel value is given in the SPM output. Also the voxel t or
> Guassianized statistics. I tried plugging these values in the formula
> above and do not get the same answer as the SPM output.
Hernando,
The corrected p-values in SPM are based on the results in
Worsley KJ et al (1996)
A unified statistical approach for determining significant
signals in images of cerebral activation.
Hum Brain Mapp. 4:58-73
http://euclid.math.mcgill.ca/~keith/unified/unified.abstract.html
The term you cite is just the first term in a sum; there are subsequent
terms that relate to different aspects of the topology of the search
region. See section 3 of the above paper for details.
The relevant SPM99 code is spm_P.m (master corrected p-value function)
and spm_resels_vol.{m,c} (computes the 4 resel measures).
One important subtlety (pointed out to me by John A): The volume used
when determining the RESEL count is *not* the number of voxels; rather it's
the *volume* of the search space.
A 2D example:
x--x--x
| | |
x--x--x
| |
x--x
This 8-pixel search region only has a volume of 3.
-Tom
-- Thomas Nichols -------------------- Department of Biostatistics
http://www.sph.umich.edu/~nichols University of Michigan
[log in to unmask] 1420 Washington Heights
-------------------------------------- Ann Arbor, MI 48109-2029
|