Also, if you really wanted to constrain probtrackx with a specific FA
threshold, simply threshold the FA with fslmaths -thr, binarize (-bin) and
then use that mask in place of nodif_brain_mask.
I wouldn't recommend this, however (for one reason, FA can be reduced
because of crossing fibers, and bedpostx may well be able to pick out two
robust fiber orientations in white matter regions of low FA). If you feel
you are getting poor results because of tracking outside of white matter,
you can probably get better results by segmenting a T1 and transforming that
segmentation to diffusion space with nearestneighbour interpolation (using
FLIRT), and using that in the place of nodif_brain_mask.
Peace,
Matt.
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
Of Saad Jbabdi
Sent: Friday, May 15, 2009 5:14 AM
To: [log in to unmask]
Subject: Re: [FSL] probtrackx with anistropy to constrain tracking
Hi Kenji,
See below:
On 15 May 2009, at 09:00, Kenji wrote:
> Dear FSL user.
>
> I have a question about "anistropy to constrain tracking".
>
> Use anisotropy to constrain tracking:
> Use this option if you want the fractional anisotropic volumes to
> influence the
> tractography.
> The tracts stop if the anisotropy is lower than a random variable
> between 0 (low anisotropy) and 1 (high anisotropy).
>
> Why is the threshold decided by not a fixed value but random numbers?
> What is the advantage?
There is no "threshold", in the common sense. The method used here is
formally equivalent to integrating over all thresholds between 0 and
1, using rejection sampling on the following distribution:
probability(stop tract) = int_0^1( prob( f < t | Y) dt
where prob( f < t | Y ) is equal to the area under the curve of the
posterior on f between 0 and t.
You can have a feel about how this works in the following cases:
when p(f|Y) = delta(0), the integral is 0, as prob(f<t|Y) is always 0.
so the tract would never stop.
when p(f|Y) = delta(1), the integral is 1, as prob(f<t|Y) is always 1.
so the tract would always stop.
when p(f|Y) is flat, the integral is 1/2, as prob(f<t|Y) = t. so there
is a 50/50 percent chance to stop.
Cheers,
Saad.
> I would like to reconstruction white matter pathways, but,
> if the value of random number is large, the tracking will stop in
> spite of areas
> of high anisotropy (white matter).
> Or if the value of random number is very small, the tracking will
> not stop in
> spite of areas low anisotropy (gray matter).
>
>
> Thanks
> Kenji.
>
Saad Jbabdi
Oxford University FMRIB Centre
JR Hospital, Headington, OX3 9DU, UK
+44 (0) 1865 222545 (fax 717)
www.fmrib.ox.ac.uk/~saad
|