Thilo,
I think the code outlined below is correct in general, but it relies
on the SPM95-esque technique of estimating smoothness from a statistic
image. This problematic (a) because a statistic image is (even under
the null) not a GRF, but a t random field, and so the smoothness
estimator will be biased, and (b) because non-null signal in the
statistic image will induce structure and alos bias the smoothness
estimate.
For more on this, see Kiebel et al, NI 10:756-766, 1999, Fig 4
especially.
If you had some principled way to estimate the FWHM a priori, using
say null data and SPM (and hence residuals), then I think this would
work OK.
-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
On Fri, 24 Feb 2006, Thilo Kellermann wrote:
> Dear all,
>
> assuming I somehow created an image-file containing, say T-values and I know
> the according degrees of freedom that belong to these values. In order to
> find a corrected threshold for an error-rate of alpha based on the
> random-field theory, is it then valid to use "spm_est_smoothness" and
> "spm_resels" to determine the number of resels in that image and use this
> information as input for "spm_uc_RF"?
>
> The procedure from which I would like to know if it is correct looks like this
> (in SPM2):
> % read T-image from disk:
> Timg = spm_vol(Timg_filename);
> % estimate smoothness of T-image (mask is provided):
> [fwhm,VRpv] = spm_est_smoothness(Timg.fname,'mask.img');
> VM = spm_vol('mask.img');
> % determine resolution elements with smoothness estimate:
> R = spm_resels(fwhm,VM,'I');
> % determine critical T-threshold (using df degrees of freedom and Resels):
> critT = spm_uc_RF(alpha,[1 df],'T',R,1)
> where Timg is the volume-mapped image of the T-image. The mask-image is a
> suitable mask for that T-image and critT will be the desired height threshold
> for that image (thresholded at alpha).
>
> Thanks a lot for any comments.
>
> Thilo
>
>
> --
> Thilo Kellermann
> RWTH Aachen University
> Pauwelstr. 30
> 52074 Aachen
> Tel.: +49 (0)241 / 8089977
> Fax.: +49 (0)241 / 8082401
> E-Mail: [log in to unmask]
>
>
>
|