That's the "Jacobian scaled" spatially normalised image, which is now
generated slightly differently. Rather than sample from the original
image using the inverse deformation and scaling by the Jacobian
determinants, the spatially normalised images are generated by pushing
the original voxels to their new positions using the forward
transform. This exactly preserves the tissue volumes, but leads to
the slight aliasing effects that you see here. Usually, the data are
spatially smoothed for VBM analysis, so these high-frequency aliasing
effects disappear.
I was thinking that you'd be using Dartel to refine the warps (ie
wanted 1.5mm isotropic "imported" images). The regularisation effects
relate to the behaviour of Dartel.
Best regards,
-John
On 12 November 2010 18:02, Christian Kaufmann
<[log in to unmask]> wrote:
> Thank you John. Just to let you know: A segmentation with 1 vs. 1.5 mm
> resolution shows different segmentation results indicated with red arrows.
> Don't know whether this results from different behaviour of the
> regularisation. Hmm, seems to me that I have to dig into this one, if I
> really want 1 mm resolution :-)
> Best, Christian
>
>
>
> Batch job for this segmentation:
>
> None that I'm aware of, although I should really check the behaviour
> of the regularisation used by Dartel, and how this is affected by
> different voxel sizes.
>
> Best regards,
> -John
>
> On 12 November 2010 10:02, Christian Kaufmann
> <[log in to unmask]> wrote:
>
> Dear John and Christian,
>
> are there any apparent pitfalls to consider when using an isotropic
>
> resolution of 1 mm instead of 1.5 (apart from 'out of memory' problems)? I
>
> resliced ~/spm8/toolbox/Seg/TPM.nii and
>
> ~/spm8/toolbox/vbm8/*IXI550_MNI152.nii with Jimmy Shen's Nifti tools
>
> ( http://www.mathworks.com/matlabcentral/fileexchange/8797 ), and adapted
>
> spm_preproc_write8 to vx = 1.
>
> Thanks!
>
> Christian
>
>
>
>
|