A customized template often isn't necessary - especially if your data are from
similar populations to those that constitute the MNI-152 dataset.
Note that you really need lots and lots of scans to create reasonable prior
probability maps. Also note, that any functions that use customised prior
probability maps may need tweeking. If they are to be used from the SPM user
interface, then the defaults for segmentation will need changing.
For optimised VBM, it is the customised prior probability images that are
likely to make the greater difference. The average T1 image is only used in
the first pass for an affine registration in order to do the first
segmentation. The final spatial normalisation is done with to the prior
probability maps.
You really want templates and tissue probability maps that reflect your
spatially normalised data. If you plan to use nonlinear warping for spatial
normalisation, then the data should probably be generated using nonlinear
warping. The tissue probability maps (used as priors) would be a measure of
the probability of obtaining a particular tissue type at a point in the
spatially normalised data, given that a particular spatial normalisation
strategy has been used.
A single pass procedure may mean that the templates and tissue probability
maps are better registered with the MNI152 data, meaning that the coordinates
you report are closer to MNI152 space. A possible disadvantage would be that
the data are not registered with each other quite as well.
There are no really correct answers, and determining which is the least
incorrect is probably an empirical matter. Which do you believe are the
tissue probability maps that best reflect the prior probabilities of any
voxel being of a particular tissue class after spatial normalisation?
Best regards,
-John
> I have some questions about the first steps in optimized VBM: the creation
> of a customised template and priors. I am an absolute SPM beginner so
> please forgive me any stupid questions! After studying protocols that I
> found in the archives and in several papers the following points are still
> unclear to me:
>
> 1. Creation of a customized whole brain T1-template (for use in the initial
> (implicit) affine registration step in Optimized VBM)
> Option a: affine-only registration of all subjects to MNI-152-template,
> smoothing (8 mm), mean image
> Option b: Do not create customized T1 at all, and use the standard template
> for the following steps instead. What would be the advantage of that?
>
> 2. Creation of customized GM, WM, and CSF priors and of a customized
> brainmask:
>
> Option a: Affine-only registration of the images to my customized
> T1-template, segmentation, smoothing (8 mm), mean image. The new images are
> used as own priors, brainmask, and GM/WM-template (for estimating the
> optimized normalization parameters) in Optimized VBM.
> Using this option I do not have exactly corresponding prior images in the
> segmentation process from above. Would that be a serious drawback? Should I
> stick to the MNI-152-template?
>
> Option b: Same as in a, but also affine AND non-affine registration to my
> customized T1. Can you comment on that?
>
> Option c: Same as in Option a. but this time all images are registered
> (affine-only) to the MNI-152-template. Later I will use my GM/WM-templates
> for the estimation of the optimized normalization parameters only. For the
> initial (implicit) affine registration step to come in Optimized VBM the
> MNI-152-template will be used.
> I have some doubts if this process actually does create GM, WM etc. priors
> that are customized “enough” as registration is to the MNI152-template.
> However, would the standard priors be better suited as in Option a. for the
> segmentation process during creation of the customized priors (because they
> correspond more exactly with the template)?
>
> Option d: Same as in Option c. However, this time for the initial
> (implicit) affine registration step to come in Optimized VBM the customized
> T1 will be used! Also, I will use the new images as own priors, brainmask,
> and GM/WM-template (for estimating the optimized normalization parameters)
> in Optimized VBM, as in Option a.
> Being derived from the same template (MNI-152), might the customized priors
> created in this option better fit my customized template in the
> segmentation steps that follow in Optimized VBM (segmentation before
> estimating the optimized normalization parameters etc.) than in Options a
> and b?
>
> Do you have any recommendations for which protocol to use? I realize there
> are scripts available for this but I would like to understand what I’m
> doing before proceeding further. Thank you for any comments and
> suggestions!
|