Regarding fixing issues with flipped coordinates: 2 of the 3 planes will
always have incorrect diffusion lines. The plane that has correct diffusion
lines does not contain the row that needs to be flipped in the bvecs file.
Therefore, if the XY and YZ planes are incorrect, however the XZ plane is
correct, that means you need to flip the Y (2nd) row in the bvecs, which was
your case.
Regarding rotating the FOV when you acquire the images: depending on the way
you generate your gradient table, you may need to account for any rotations
of the FOV. If you simply take the gradient table that is in the sequence,
you will have to correct it for the rotations. If you use some software to
generate the table based on the DICOMs (e.g. MRIConvert), it may already
take into account these rotations. Changing the angle of the FOV is fine as
long as you properly correct the gradient table for it.
Regarding rotating the images during post-processing: if you rotate the raw
DTI data significantly at any time during post-processing, you will have to
make corresponding changes to the gradient table. I posted a shell script
that helps with this previously; it takes a FLIRT matrix and transforms your
bvecs file according to the rotational components of the matrix.
Peace,
Matt.
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
Of Michael Scheel
Sent: Friday, November 28, 2008 9:12 PM
To: [log in to unmask]
Subject: Re: [FSL] question re meanf2samples
thanks Saad, you were right my y-coordinates were flipped. I figured
that out since in the occipital callosal fibers the tensors were
looking from anterior-lateral to posterior-medial instead from
anterior-medial to posterior-lateral.
However I had some issues with wrong bvecs already in the past - so I
am wondering: Do you have an approach on how to check you have the
right bvecs?
E.g. we correct for angulation to ac-pc-line - it's mostly only some
degrees but for tracking it will matter I guess.
1) Would you discourage to angulate to ac-pc on acquistion as an extra
uncertainty factor?
2) How do you make yourself confident that you're working with the
right bvecs - any quick test for it?
cheers, michael
On 28-Nov-08, at 1:13 AM, Saad Jbabdi wrote:
> Hi Michael,
> I suspect this is a problem with your bvecs orientations. Even when
> the colour coded V1 image looks fine, there can still be a sign flip
> in either the x, y or z component of V1 (which in turn can be due to a
> flip in the same component of bvecs).
> So in order to ensure that this is not the case, you need to look at
> V1 in "line mode" in fslview (not RGB mode), and make sure that e.g.
> the corpus callosum runs fine on both coronal and axial views.
>
> Cheers,
> Saad.
>
>
> On 27 Nov 2008, at 18:06, Michael Scheel wrote:
>
>> Hi everybody,
>>
>> I have a question regarding the meanf2samples.nii.gz output of
>> bedpostx. As i understand it shows in which voxels the data support
>> a second fibre orientation. I have been looking at my data and
>> compared them to the data in the tbss practical dataset. At first I
>> was very happy since it seems that the data support a lot of second
>> fibre orientation.
>> But when I use probtrackx on the dataset I just don't get reliably
>> tracking. Even V17-left-right connections (with Callosum as
>> waypoint) I don't get any results. I checked on the tutorial-data
>> set - where it runs fine and gives expected results. Therefore I
>> think that that the problem lies in the data. (my data have 32
>> directions, b-value 700, 3 averages - dtfit works excellent with
>> good FA and correct color coding - so bvals/bvecs should be ok as
>> well)
>> Any idea what's going on? Is the meanf2samples.nii.gz showing that
>> my data is too noisy?
>>
>> Thanks a lot, Michael.
>>
>
> Saad Jbabdi
> Oxford University FMRIB Centre
>
> JR Hospital, Headington, OX3 9DU, UK
> +44 (0) 1865 222545 (fax 717)
> www.fmrib.ox.ac.uk/~saad
|