The website says E=0.5 (is this voxels, mm or what?)
And H=2 (I assume the height of the t statistic I am using),
Why this discrepancy? Have you found out something empirically that was
not obvious when the website was created? Or are these parameters better
suited for TBSS skeletons? Stefano
-----Original Message-----
From: Steve Smith [mailto:[log in to unmask]]
Sent: Thursday, March 27, 2008 4:41 AM
To: [log in to unmask]
Subject: Re: [FSL] mask usage and duration of calculations in randomise
for TBSS
Hi - depending on what options you used for the template specification
in TBSS this is not necessarily surprising. If you used the "find the
most typical subject" then this template is only affine-registered
into MNI space and so will not exactly correspond to all labels in all
the FSL atlases. Even if you register directly to the FMRIB58_FA,
exact matching is not guaranteed as the atlases are defined with
different templates, different registration programs etc - so will be
close but not necessarily exact.
Your approach for resolving this is indeed the most obvious sensible
thing to do - e.g. dilate the atlas mask of interest sufficiently to
overlap your skeleton. Yes, dependingon how you then extract
information from the new mask, it probably makes sense to AND the
dilated mask with your mean_FA_skeleton_mask.
Yes, cluster-based thresholding inside randomise is much slower than
voxel-based, but is also more sensitive to extended regions of signal.
You might also try TFCE (E=1 H=1, nperm=500 to start with) instead of
cluster-based.
Variance smoothing does probably slow things down even more, though I
don't have a feel for by how much. It seems that if your Nsubjects>~15
you may not see much (if any) benefit from variance smoothing.
Cheers.
On 26 Mar 2008, at 20:30, Marenco, Stefano (NIH/NIMH) [E] wrote:
> I am trying to use a mask for the ant limb of the internal capsule
> after running TBSS. I am using randomise for my stats.
> One thing I noticed is that the AnteriorLimbInternalCapsule mask
> available in FSL does not cover fully the ant limb of the skeleton
> (obtained with MNI152) in the lower portions of the structure. I was
> perplexed by this discrepancy, I don't quite understand how this
> arises. I addressed this by a modal dilation of the mask with
> fslmaths.
>
> One question is the following: to take full advantage of the mask,
> do I have to do an AND operation between this mask (that exceeds the
> boundaries of the skeleton) and the skeleton itself, and obtain a
> more restricted mask?
>
> I noticed that when running randomise the duration of the
> computations increases tremendously when a cluster correction is
> requested of the program. However, in the same analysis I also set
> the variance smoothing to 6. Does variance smoothing contribute to
> the duration of calculations? And is the size of the variance
> smoothing kernel related to the extra time it takes to do the
> analysis?
>
> Does tfce take even more time than simple cluster extent?
>
> Stefano Marenco, MD
> NIMH/GCAP
> Building 10, room 4S235
> 10 Center Drive
> Bethesda, MD 20892
> tel 301 435-8964
> fax 301 480-7795
> email: [log in to unmask]
>
------------------------------------------------------------------------
---
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
------------------------------------------------------------------------
---
|