I agree with Gwenaelle's assessment of things. With 12 directions, you
really won't be able to resolve crossing fibers, and this can lead to
significant inaccuracies in many pathways (e.g. the arcuate, the corpus
callosum, etc). Additionally, if you are planning to use the JHU
tractography atlases included with FSL, these were generated with a
deterministic tensor-based algorithim which also fails to resolve crossing
fibers. For example, if I compare this atlas's probability map of the
arcuate to a group average of 30 subjects acquired with 30 directions of my
own data, the focus of frontal projection is completely different (it goes
to the precentral gyrus in the atlas, but to the inferior frontal gyrus in
my results, both in MNI standard space).
I am not clear, with the data you have, what the benefit to doing
tractography or comparison with that tractography atlas is at all. Why not
simply run TBSS and describe anatomically the locations of significance that
you find in your group? Are you trying to avoid multiple comparison
correction or something like that? Presumably, TBSS as it stands now isn't
as sensitive to the lack of directional information in the data you have,
and will be more sensitive to its SNR (so hopefully you have more than a
single 12 direction repetition).
Peace,
Matt.
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
Of Gwenaëlle DOUAUD
Sent: Tuesday, February 09, 2010 3:18 PM
To: [log in to unmask]
Subject: [FSL] Re : [FSL] creative use of TBSS?
Hi Susie,
> Hello,
> I am working with a data set of 50 subjects and am
> attempting to analyze the
> 12-directional DTI data. I have been told that 12
> directions is not enough
> to perform tractography--is this true?
I am not the best person to answer this question, but I would think that the
problem is that you would not be able to model more than one single fibre
population in each voxel with this dataset.
So, provided that you would not be looking at tracts with a complex
architecture (e.g. crossing other tracts), or just looking at the "core" of
it (cf corpus callosum, cingulum bundle), this should be ok... Anybody
else's got a take on this?
> Assuming that it is unwise/unreliable for me to conduct
> tractography on the
> 12-dir DTI set, I was wondering if it is possible to use
> TBSS to perform a
> similar analysis. What I am proposing is:
> 1. Mean FA skeleton from 50 subjects is generated
> using TBSS tools.
> 2. Skeleton is masked with tract of interest from
> probabilistic atlas.
> 3. Mean FA is found for skeleton voxels within the
> mask for each subject,
> representing an estimate of the FA in the cingulate gyrus
> for example.
>
> Is this defensible?
This sounds sensible to me *as long as* you are aware that the vast majority
of the voxels in the WM is a mixture of different tracts (as you can see if
you turn on some of the WM atlases in fslview, see also Behrens et al.,
2007). So this is true also with the voxels of the skeleton to some extent,
but this still would be a good approximation.
This approach would lead to absolutely unambiguous results for regions like
the cingulum bundle, or the most "midsagittal" part of the fornix and corpus
callosum etc.
Hope this makes sense,
Gwenaëlle
|