No. The idea behind the way that Dartel does things is that images of
the same study are more likely to match each other than they are images
from different studies (or some template constructed from other people's
data). For this reason, there is a within study non-linear alignment,
and a separate study-to-MNI alignment based on an affine registration.
These two transforms are combined by the 'Normalise to MNI space'
option.
The reasons why intra-study alignment may be preferable are:
1) Image contrasts differ among sequences, scanners etc, so what the
segmentation considers as GM from one sequence may not be consistent
with what is considered GM from another sequence. Closer alignment of
tissue classes should be possible when there is more agreement on which
tissues are homologous.
2) Subject populations may differ. For example, if the aim is to align
brains of children together then the accuracy of alignment across
subjects in the study may be more accurate if the brains are aligned
with a template derived from on kids' brains. The MNI templates are
based on European (or American) subjects of a particular age range. If
I was working in Korea or Singapore (for example), then I'd be pretty
reluctant to base my inter-subject alignment on such an atlas.
The current MNI space is really only defined for affine registered data
(so the study-to-MNI alignment is only an affine transform). At one
time, there were plans to have an ICBM brain model (atlas/template)
based on more precisely aligned images. At this point, the
study-to-template alignment in SPM would need to become non-linear.
Note also that the objective function for matching the study average to
MNI space is based on KL-divergence between the templates. There is
only a simple least squares matching behind the 'Normalise' button. The
choice of objective function depends on the properties of the data. The
distribution of the noise in tissue class data is not well modelled by a
Gaussian distribution, so least-squares is not really appropriate.
Best regards,
-John
Dear SPMers,
I was just wondering, does this imply that the 'Normalise'
button in SPM (without non-linear iterations) is exactly the
same as the affine bit of DARTEL Normalise to MNI Space?
cheers,
ying
--
John Ashburner <[log in to unmask]>
|