Just to add to Michael's comments. You can figure out voxel sizes
from the matrices in the NIfTI header by:
vox = sqrt(sum(M(1:3,1:3).^2))
where M is the matrix. A matrix of
M = [-0.9988 -0.0287 0.0390 124.3288
-0.0201 0.9785 0.2051 -97.2591
0.0441 -0.2041 0.9778 -13.5133
0 0 0 1.0000]
gives vox = [1.0000 1.0000 0.9998], which indicates 1 mm
isotropic voxels. The 0.0002 mm discrepancy in the size along z is
simply because of the limited precision with which the matrix is shown
in the original email.
Best regards,
-John
On 23 April 2012 14:31, Michael Harms <[log in to unmask]> wrote:
> Hi John,
> Your acquisition was apparently acquired with an oblique orientation, in
> which case you expect there to be off-diagonal components in the q/sform
> matrix in the nii header.
>
> If T is your matrix, and
> w1 = T*[i j k 1]' is the position of voxel index (i,j,k) in "world
> space" and
> w2 = T*[i j k+1 1]' is the position of voxel index (i,j,k+1)
> then
> norm(w1-w2) = 1.0,
> which is exactly what you'd expect if the data at hand was collected
> with a 1 mm isotropic spacing.
>
> That is, any structural measures should be fine. When you display the
> data in "Voxel space" you are showing it aligned with the original
> acquisition axes. When you show it in "World space" you are rotating it
> according to the matrix T.
>
> cheers,
> -MH
>
> On Mon, 2012-04-23 at 04:07 +0100, John Smith wrote:
>> -0.9988 -0.0287 0.0390 124.3288
>> -0.0201 0.9785 0.2051 -97.2591
>> 0.0441 -0.2041 0.9778 -13.5133
>> 0 0 0 1.0000
|