Thank you! What we ended up doing is using the non-stationarity toolbox implemented in SPM (Hayasaka, http://fmri.wfubmc.edu/cms/NS-General). As it turns out, it's not hard to use a looped 3dcalc (AFNI, Cox) command to split the FSL-generated 4D VBM results into separate 3D datasets (one for each subject), then submit these as the dependant variable in a SPM model estimation, save the "SPM.mat", and use that model to submit to the "ns" toolbox. It spits out the NS-corrected clusters quite nicely. If anyone needs (more specific) steps on how to do this, you can email me at schwartd (at) ohsu.edu. Anyway, thank you!

-Daniel


-----Original Message-----
From: FSL - FMRIB's Software Library on behalf of Reza Salimi
Sent: Sun 6/7/2009 4:16 AM
To: [log in to unmask]
Subject: Re: [FSL] Non-stationary thresholding of VBM results
 
Hi Daniel,
currently we are adding this option to *randomise* for both TFCE and
cluster-based inferences,
but unfortunately, it is not yet publicly available ...
We will present the algorithm and the results in HBM 2009 in SF in case you
are going to be there ...

but in general,
If you want to adjust the inference for the nonstationarity, it is necessary
to use adjusted cluster-sizes (in RFT-based RPV units) in all of you *
randomise* permutations (see Hayasak et al. 2004). It means, you need to
modify the current randomise code in order to do that ... and cannot combine
other tools with *randomise *in order to do that ...

However, if you want to parametrically do this, you can use the T-stat image
from* randomise* plus the standardized residual error of the first
permutation's GLM fit. This residual is used in order to estimate the
geometry of the random Z-stat (or gaussianized T-stat) and hence its null
distribution under the null hypothesis, and results in a voxel-wise
roughness estimate (RPV, or roughness per voxel). Currently, this residual
is not written as an ouput volume; you need to add a line to the
randomise.cc file, which writes this 4D residual volume in the output (note!
just the first permutation) ...

I haven't used that matlab toolbox, but in the RFT-based adjustment for
nonstationarity (according to Wrosely's works, and later Hayasaka's works),
these are the only two volumes you need.

hope it helps ...

Best, Reza


On Wed, May 27, 2009 at 4:16 PM, Daniel Schwartz <[log in to unmask]> wrote:

>
> Hi everyone!
>
> I'm not sure if NS cluster based inference is available in FSL (I assume it
> isn't) so I've been trying to use Worsley's FMRISTAT MATLAB-based toolbox.
> However, I'm having some trouble implementing the correction. Does anybody
> either a) have a different program to suggest or b) have instructions on how
> to do this in FMRISTAT? Thank you in advance for your help!
>
> -Daniel
>



-- 
G. Salimi-Khorshidi,
D.Phil. Student, Dept. of Clinical Neurology, University of Oxford.
[log in to unmask]    http://www.fmrib.ox.ac.uk/~reza
FMRIB Centre, John Radcliffe Hospital,
Headington, Oxford,  OX3 9DU
Tel: +44 (0) 1865 222466  Fax: +44 (0)1865 222717