Hi Jon,
realigning the images prior to segmentation improves the starting
estimates, as the input image is then closer to what the segmentation
routine expects. However, you should do this using a rigid body
procedure, which does not involve scaling (exactly for the reason you
describe). Therefore, yes, realigning images is a good idea, and yes,
scaling does change voxels sizes, but you do not need the latter to
achieve the former. Try simply coregistering the anatomical with a
template image: it will not be larger afterwards, but better-aligned,
which is what you want. In order to do this manually, play around with
the values for pitch, roll, and yaw (rotations) and right, forward, and
up (shifts) when displaying an image.
Hope this helps,
Marko
> Pre-aligning images to the MNI template prior to segmentation for
> volumetry has been discussed on the list before and can improve
> segmentation depending on the particulars of the data set. I tried
> this with a rigid body + linear scaling transform, using 'Check Reg'
> and 'Reorient'. After applying the transform/reorientation to the
> image the transform is saved in the nifti sform fields as expected.
>
> I didn't expect the voxel size in the header to also change - is this
> correct?
>
> For example, applying a scaling of 0.9 to a template image changes
> the voxel size, sform and qform as follows:
>
> nifti_tool -diff_hdr -infiles single_subj_T1.nii
> single_subj_T1_original.nii name offset nvals
> values ------------------- ------ ----- ------ pixdim
> 76 8 -1.0 1.8 1.8 1.8 0.0 0.0 0.0 0.0 pixdim
> 76 8 -1.0 2.0 2.0 2.0 0.0 0.0 0.0 0.0 qoffset_x
> 268 1 81.0 qoffset_x 268 1 90.0 qoffset_y
> 272 1 -113.400002 qoffset_y 272 1 -126.0
> qoffset_z 276 1 -64.800003 qoffset_z
> 276 1 -72.0 srow_x 280 4 -1.8 0.0 0.0
> 81.0 srow_x 280 4 -2.0 0.0 0.0 90.0 srow_y
> 296 4 0.0 1.8 0.0 -113.400002 srow_y 296 4
> 0.0 2.0 0.0 -126.0 srow_z 312 4 0.0 0.0 1.8
> -64.800003 srow_z 312 4 0.0 0.0 2.0 -72.0
>
> Since the voxel size has changed, volumetry measurements in native
> space are affected and there doesn't appear to be a way to separate
> the original voxel dimensions (for volumetry) from the voxel to
> template mapping (when this mapping is 9 dof rather than 6).
>
> thanks for your comments, Jon
>
--
=====================================================================
Marko Wilke (Dr.med./M.D.)
[log in to unmask]
Universitäts-Kinderklinik University Children's Hospital
Abt. III (Neuropädiatrie) Dept. III (Pediatric neurology)
Hoppe-Seyler-Str. 1, D - 72076 Tübingen
Tel.: (+49) 07071 29-83416 Fax: (+49) 07071 29-5473
=====================================================================
|