Hi Zach,
we looked into such a comparison a while ago (Keller et al, NeuroImage
2004) and found very little differences between the custom template and
the standard template in a group of morphologically-almost-normal
adults. This was back in spm2-days, and the issue, as you correctly
suppose, is even less pressing in spm5.
However, one issue that may be of relevance to your project is that you
are looking at elderly subjects, whose brain even in the normal state
will be quite different from the MNI standard template (average age is
27 years, if I remember correctly). As a rule of thumb, the further away
you are from the population that contributes to the standard template
the more relevant a custom template may be, which would argue in favor
of looking into this. In this case, I would probably try to first create
my custom template and the to do the exact same procedure twice, once
with your priors and once with the standard ones. You can then compare
either the results (effect of age in one vs. the other group) or
contrast the processed data directly.
As to your procedure, there is a long discussion on whether one should
only use an affine normalization when creating a template (I think so).
As to the smoothing, the priors for unified segmentation are not
smoothed anymore (see the standard priors in spm5/tpm), and you do not
necessarily need a T1 anymore. Moreover, one could argue that using the
standard priors for your initial segmentation already introduces a bias,
which you could circumvent by segmenting without priors, for example
using Christian's VBM5 toolbox. Also, I would not normalize to the T1 as
this will be less accurate than normalizing via the segmentation
routine. Other than that, no more comments :)
Best,
Marko
THRELKELD, ZACHARY D schrieb:
> Hello,
>
>
>
> I’m working on a VBM comparison of regional grey matter volume
> differences between normal and mild Alzheimer’s groups. I am using SPM
> and have read in previous posts that creating a custom T1 template and
> custom GM priors is not considered necessary in SPM5 due to the new,
> optimized iterative segmentation/bias correction/spatial normalization
> procedure.
>
>
>
> However, we would still like to compare results obtained from segmenting
> images with the standard SPM priors versus normalizing using a custom
> (group-specific) T1 template and segmenting using custom priors. I was
> thinking of adopting the following steps. Can you please tell me if
> these steps are reasonable in SPM5?
>
>
>
> Thanks in advance,
>
> Zach
>
>
>
>
>
>
>
> Create a custom T1 template:
>
>
>
> A) Normalize all subjects’ structural images to the ICBM/MNI template
>
> B) Smooth normalized scans using 12mm FWHM Gaussian kernel
>
> C) Average all the normalized, smoothed images into a custom T1 template
>
>
>
> Create custom tissue probability maps (priors):
>
>
>
> D) Warp all native T1 scans to the custom T1 template
>
> E) Segment normalized scans into GM/WM/CSF images using standard SPM priors
>
> F) Smooth GM/WM/CSF images using 8mm FWHM Gaussian kernel
>
> G) Average smoothed GM/WM/CSF images to obtain custom GM/WM/CSF priors
>
>
>
> Segment the original structural MRI scans using custom template and priors:
>
>
>
> H) Normalize original images to custom T1 template using affine,
> nonlinear transformations, medium regularisation, re-sampling to 2 mm^3
> and no masking
>
> I) Segment normalized images into GM/WM/CSF using custom priors
>
> J) Smooth normalized, segmented GM classes with 8mm FWHM Gaussian kernel
>
> K) Proceed with statistical analysis of smoothed, normalized GM classes
>
>
>
--
=====================================================================
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
=====================================================================
|