The most likely explanation is that FSL interprets your gradient table
differently than does DTIStudio, and so the table needs to be adjusted for
FSL. You should display the V1 output of DTIFIT as lines in FSLView and
look at the lines in all three orthogonal planes in structures like the
corpus callosum and corona radiata. The lines will likely be off in two of
the three planes (XY, XZ, YZ) and the one in which the lines are correct
will not contain the row of the gradient table that needs to be flipped.
For example if the XY plane is correct the z row will need to be flipped (it
is the third row). By flipped, I mean you need to change all of the signs
to the opposite of what they currently are. You will then need to rerun
DTIFIT and BEDPOSTX.
Peace,
Matt.
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
Of Tugan
Sent: Tuesday, November 18, 2008 3:19 PM
To: [log in to unmask]
Subject: [FSL] Q about probabilistic tractography
Hello everyone,
I would like to ask your experience in general in generating the tracts
using probabilistic tractography as implemented in FSL (bedpost/probtrackx).
I have a group of DTI data from a 3T MRI system with relatively good spatial
and directional resolution (1.75mm in plane, 2mm slice, 32 directions). I
run this data through all FDT 2.0 steps and try to generate tracts. Major
tracts can be generated fine but there are occasional -somewhat unnatural-
zigzags in the tracts. I tried cortico-spinal tracts and ILF and they
produced OK. But cingulum generally produced very poorly most of the time;
sometimes no cingulum can be produced between two masks I generated near
splenium and genu of CC. Data SNR is fine and I tried several subjects with
no motion or other artifacts. I run the same image data through DTIStudio
and the tracks produce beautifully. I was thinking that probabilistic
tractography would be more robust against noise in pixels and produce tracts
more reliably. But, for some reason, streamline techniques using brute-force
approach generated the tracts better. The same data, the same gradient
tables. ECC and head motion is corrected by 12 parameter affine
transformations by both programs. Does anyone have any insights why I might
be experiencing this? By the way, I am using the default values for
bedpost/probtrackx steps and using the same mask images for both probtrackx
and dtistudio.
kind regards,
Tugan Muftuler
Univ. California, Irvine
|