The clinical toolbox is based on SPM. SPM expects modality
specific templates (e.g CT, PET, MR T1, MR T2, etc) that match
tissue contrast. The clinical toolbox provides a CT-specific
template that matches the size, shape and alignment of other SPM
templates. These templates provide a good average shape and size of
a European adult brain.
There are several caveats:
1. While SPM uses average sized brain templates, other
tools like ANTS and FSL use MNI templates that have roughly the
same size and orientation but are larger than typical. This
difference is illustrated by Figure 1 or Horn et al. This also
explains why MRIcroGL comes with both the "spm152" and "mni152"
brains - note that these two images are slightly different in size
to each other.
https://www.ncbi.nlm.nih.gov/pmc/article...
In general, the normalization provided by the clinical toolbox
will be similar to other SPM tools, but will not exhibit the size
dilation of FSL and ANTs. Therefore, you need to be careful that
you are consistent on the tools you use within an analysis.
2. Our CT template was intentionally acquired in individuals who
had similar age to older adults who have strokes in our region.
Therefore, the ventricles are a bit more dilated than the SPM
templates that are based on young adults and you can observe some
sulcal atrophy. Further, our template is representative of the
diverse population of South Carolina and does reflect the diet and
genetics of our region. Features like post-bregmatic depression and
wide diploic space were common in our dataset. While we aligned all
of our population to match the shape of the SPM templates, it does
make our template different than others. I actually think it is a
good idea to capture diversity in our templates, but I must
recognize that our template was by design less homogeneous than
others. If your sample is large enough, you can use our templates
as a starting point, and use DARTEL to create a template that
better reflects your population.
In general, the choice of template matters. Our template
provides a robust normalization for stroke and is appropriate for a
study where you use a SPM for all processing. Mixing modalities
(e.g. CT and T2-FLAIR), time post injury, multiple sites iwth
different populations, etc. will not only create differences
in template, but also the extent of the injury that you see. In
general, the prime consequence of these sources of variability will
be to lower statistical power, even if they can aid the clinical
generalizability. These concerns and comments also apply to
atlases.
