Hi Siewmin (and I guess this should answer Sid's questions as well),
If you run FSL's eddy_correct on your data, then this will, as you
know, produce a text file called blah.ecclog
You can then use the attached matlab script which will read your
ecclog file and bvecs file, and apply the appropriate rotations to
your bvecs.
The top-left 3x3 of the flirt matrices do not only contain rotations,
so you need to do a little trick to extract the rotation part (this is
done for you in the matlab script attached).
In my experience, I did not notice a huge difference in my data when
comparing e.g. FA before and after bvecs rotation, but I guess it
might make a difference on other data...
The relation between the amount of bvecs rotation and the output of
dtifit and/or the end result of tractography is not straightforward,
so I am not sure you can objectively assess whether a correction for
the bvecs rotation is necessary or not just by looking at MCFLIRT
results.
Cheers,
Saad.
On 28 Feb 2009, at 01:40, Siewmin Gan wrote:
> Hi, I have a couple questions about this procedure which I came
> across in the discussion
> forum.
>
> I have 2 repetitions of the same DTI sequence to increase the signal
> gained. I have
> concatenated my bvals and bvecs and ran eddy_correct on the merged
> file ( lining the
> brain s up with the -t option in fslmerge).
> It is recommend in the post here to " then rotate the bvecs by the
> rotation part of the flirt
> transformation matrices that are stored in the ecclog. I think you
> should be able to
> do this by simply multiplying each bvec by the top 3x3 matrix in the
> flirt matrix for that volume"
>
> May I ask if you mean multiply each bvec number by the TOP LEFT 3x3
> matrix as there
> are 4x4 matrix in the ecclog for each volume?
>
> Secondly would this step be mandatory and would it be less accurate
> in my
> DTIFIT/BEDPOST procedure if I did not rotate the bvecs as mentioned
> above? It is noted
> on the email post that it is presumed the two brains are at
> different angles at each
> repetition of the sequence. How much motion would be required (from
> one repetition to
> another repetition) to do this above step of rotating the bvecs?
> Visually when I check the
> B0 images from each repetition, some subjects have a noticeable
> shift in between the
> repetitions while others has same positioning of the brain in both
> repetitions.
>
> I used the MCFLIRT function on the FEAT option to check the level of
> motion in the merged
> file. The rotation is upto 0.04 rad and the translation 2-5mm. It
> mentioned high level of
> motion detected with the merged file. Is the rotation more important
> than the translation
> movement detected, to decide if I need to perform this step of
> multiplying each bvec by
> the 3x3 matrix in the ecclog for DTIFIT/BEDPOST procedure?
> If the pre-stats report did not have high level of motion detected,
> would it be safe to
> concatenate the same bvecs (without rotating) to do the the DTIFIT/
> BEDPOST procedures
> required for tractography analysis?
>
> Many Thanks in advance
>
> Siewmin
>
Saad Jbabdi
Oxford University FMRIB Centre
JR Hospital, Headington, OX3 9DU, UK
+44 (0) 1865 222545 (fax 717)
www.fmrib.ox.ac.uk/~saad
|