Hi

Great, software, and 0.6 looks like a very useful update.

I'm wondering how best to manage the gradient table for nifti files that include a rotation. As far as I can tell (at least in 0.5), the nifti rotations are ignored, which means that the vectors in the gradient table have to be "unrotated" to match the imported DTI images. So I guess a request would be to support nifti rotations, but in the meantime, is there a way I can use the patient orientation parameter flag (or some other means) to correct the gradient table, and/or adjust the DTI images?

 

I've got some details here using SPM code that show what's going on:

% Read nifti header information (in SPM format)

V = spm_vol( niftiFilename );

% Read translations/rotations/scaling/shears in nifti header; the rotations are the 4th, 5th and 6th elements of prm

prm = spm_imatrix(V.mat);

If prm (4:6) are 0, there is no problem. However, if prm(4:6) are not 0, that means the nifti header indicates rotations, but these appear not to be accounted for when the data are read by dti_recon. That means the DTI images might be at a funy angle, which is not the end of the world, but the vectors in the gradient table have to have these rotations "removed" otherwise they will be wrong. FYI, here's how I do it in SPM:

% Create -ve rotation matrix

Mrot = spm_matrix([0 0 0 -prm(4:6) 1 1 1]);

% Given vectors in variable "bvecs" for each image

for i = 1:size(bvecs,1)

bvecs_norot(i,:) = (Mrot(1:3,1:3)*bvecs(i,:)')';

end

 

Best wishes,

Paul

 

 

Post edited 11:26 pm – March 24, 2010 by pmmacey

---

Hi Ruopeng

Sorry I was unclear.

The gradient tables are read from the scanner without problem. The issue is one of reading in the nifti files, and accounting for the rotations in the affine transformation matrix in the nifti header.

Let's say I rotate my DTI images to better match the template space, or because the subject was slightly tilted. I can perform this rotation by changing the parameters in the nifti header. In other programs (SPM, MRIcron), the image will then be displayed rotated , but it appears that DT does not read/apply those rotations.

The issue with the gradient tables is that we need to account for the rotations not being read in, so it is a secondary problem.

I hope that makes sense!

Best wishes,

Paul

 

 

Post edited 1:17 am – March 30, 2010 by pmmacey

---

Hi Ruopeng

I take it that by "local space" you mean that DT reads in the

slice-by-slice data, and accounts for voxel sizes and origin shifts as

read from the nifti header. If so, that makes sense to me, and I can work with that.

I think there are times when we want to align our data to another space. Once scenario is motion correcting two series from the same subject. If there is movement between the series, then we can rotate/shift the second series to the spaced of the first, and combine the data from both series when reading into DT. The problem arises if the motion correction is done by changing the nifti header parameters to do the shift/rotations (as with SPM), as opposed to reslicing. We could reslice the data, but it would be nice to avoid reslicing because of partial volume effects and so on.

MRIcron can display without or without reorienting (it's an option under Preferences – the option takes effect after restarting the software).

I think the question of reorienting gradients is secondary and probably not worth bothering about – we read them in from Siemens without problem, and can use them with DT as long as we make sure they are in local space (i.e., not reoriented). As long as I have the correct understanding of how DT reads the data in local space, then I can figure a way to register the DTI data to different spaces when needed.

Thanks,

Paul

 

 

Hi again

Thats very helpful. And yes you're right, I will need to reslice if I want to combine series. And the "local coordinates" is also clear to me now.

Time to get back to analyzing data!

Best wishes,

Paul