This is an interesting question to me, as we just finished processing
data with multiple shells (with less than 60 directions at the highest
b-value) through the probabilistic fiber tracking method. The results we
get for thalamic segmentation look very similar to those published in
your Nature Neuroscience paper, and appear to be fairly reproducible, at
least for frontal and temporal connections to the thalamus, however I am
not sure if the estimates are more noisy than otherwise. We do not have
a comparison with one b-shell for the same subjects. Tim, would you like
to look at a sample of these data? Would it be possible to compute a
bi-exponential fit in bedpost?
Stefano Marenco, MD
GCAP, Clinical Brain Disorders Branch,
NIMH
10 Center Drive, room 4S235
Bethesda, MD 20892
tel. (301) 435-8964
fax. (301) 480-7795
email: [log in to unmask]
-----Original Message-----
From: Tim Behrens [mailto:[log in to unmask]]
Sent: Tuesday, March 27, 2007 9:07 AM
To: [log in to unmask]
Subject: Re: [FSL] intermediate b-value: how useful?
Hi - I guess this is not an easy question - here is my take.
If you are fitting a single diffusion tensor, the diffusivity (and
FA) values that you measure will depend on b-value, as apparent
diffusion is not mono-exponential in the brain, as the diffusion
tensor model assumes.
Therefore, if you collect at more than a single b-value, your model
fit to a diffusion tensor model will be worse. You will measure
something between what you would have measured with each one of the b-
values. My guess is that the error in FA/MD will be higher than for a
single b-value shell acquisition, but I have not spent a long time
thinking about it. However, remember that for single shell
experiments, your measurements are necessarily conditioned on you
choice of b-value.
If all you are interested in is diffusion orientation (for
tractography), I again think you are better off spreading your
orientations equally on a single shell. The model in bedpost also
assumes mono-exponential decay of each compartment, so I expect that
the multi b-value experiment will look like more noise to the
bayesian estimation (as the model fit will be worse) and therefore
will inflate uncertainty.
Having said this, we do not have practical experience.
T
On 26 Mar 2007, at 20:43, Stephen Towler wrote:
> For TBSS and probtrack analysis, what benefit (if any) does one
> derive from acquiring three b-values (b0, bmax/2, bmax) instead of
> two b-values (b0 & bmax)? When collecting 2-3 mm isovoxels at 32
> or sixty directions, omitting the intermediate b-value creates
> substantial time savings which is a huge help with compromised
> subjects.
>
> Thanks!
> Stephen
>
>
> --
> (please note email change from mbi.ufl.edu to ufl.edu)
>
> Stephen Towler
> [log in to unmask]
> 352-294-0048 office
> 352-258-6409 mobile
> 352-392-8347 fax
>
> Leonard Lab
> Department of Neuroscience
> PO Box 100244
> University of Florida HSC
> Gainesville, FL 32610
|