Hi
On 24 Apr 2007, at 12:10, Finn Lennartsson wrote:
> Hi all,
>
> I have a question about prescribing gradient directions to FDT.
>
> Searching the mailing list I reckon that many posts concern this
> question, but I cannot really figure out what to apply on my question.
>
> We have acquired a DTI set on a GE scanner (v.12) with 45
> directions and 6 b0. The slices were axial slices with the
> Frequency Echo (FE) in R/L and Phase Echo (PE) in A/P (as defined
> on the scanner). For FDT I define a bvecs based on the tensor.dat
> file used for feeding the gradient directions into the GE-scanner
> (or actually I have a user-specified file with Jones directions).
>
> When running DTIfit then I get a V1-vectorfield which seems to be
> flipped R/L. ( e.g. wrong principal directions on axial images in
> Corpus Callosum or coronal images in Internal Capsule). The FA-maps
> appears to be correct, as one would expect if there is just a R/L
> problem.
>
The x -orientation in you images may well be defined opposite to the
scanner orientations. Particularly if you are using radiological
format. The bvecs should be defined relative to the image, not to the
scanner for dtifit.
>
> I tried to flip the x-row in bvecs (x->-x), and then the fibers
> were correct in the Corpus Callosum and the Internal Capsule.
> However, I am not certain of what this problem-solving flipping
> means in practise...? So my question is whether my images are
> defined incorrectly, whether I am feeding them with non-congruant
> directions or whether the images and the bvecs (as they are now)
> are just incompatible ( e.g. defined in different spaces)? Or if
> something else is wrong...?
>
It sounds to me like you have done exactly the right thing. Don't
forget to check the vectors in all three planes though (axial,
coronal, sagittal).
> I do the conversion of dicom->nifti with spm5 ("dicom import"), and
> this seems to work fine as for all the avworient-stuff
> (radiological, sform/qform are scanner anat with LAS). I read
> somewhere on the web that the gradient directions on a GE-scanner
> are relative to FE, PE and SS-axis. Is this true? What if I use
> oblique slices or have a set for which FE/PE/SS does not correspond
> to x/y/z? How should I convert a gradient set into a bvecs file so
> that FSL can interpret them correctly for these cases?
> Thanks,
In this case, I'm afraid you will have to rotate your bvecs by hand
using the rotation matrix defined by your slice orientation. It's
easier to acquire in pure axial!
T
>
> Finn Lennartsson
