It is true that the *_seg_sn.mat files contain an affine transformation
that is actually rigid-body. Any global size differences that would
previously have been accounted for by the affine transform are now
encoded by the nonlinear deformations instead. "Modulation" should
still be correct (you can try this by summing up the volume of a
native-space grey matter image, and a modulated spatially normalized
grey matter image).
The reasons for doing things this way were as follows:
The deformations estimated by the segmentation are actually a mapping
from the native space image to the tissue probability images (normally
in the spm5/tpm directory). Writing spatially normalized images
requires a mapping from the TPMs (template) to the individual images.
Therefore, the deformations need to be inverted. Doing this properly is
not as trivial as often perceived. The way it is done here, is to
create a full deformation field from the parameters (affine and DCT),
invert this deformation field using the piecewise affine model described
in the appendix of one of my papers, and then re-parameterise with an
affine transform and a bunch of DCT basis functions.
A few people are interested in applying "Deformation-based Morphometry"
methods. One way to do this is to do a Procrustes decomposition of the
deformation in order to factor out the pose (and sometimes the size) of
the individual subjects brains. What is left in the deformations should
then reflect the shape and size of the individual brains.
Because the deformations were generated in full, I figured that it would
be reasonably trivial to incorporate this Procrustes decomposition.
This is why the affine transform encodes a rigid body mapping (i.e. the
pose of the subject). The advantage of this is that anyone wishing to
do some sort of DBM can analyze the coefficients of the DCT transform
using standard multivariate methods.
Best regards,
-John
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
On Behalf Of Christian Gaser
Sent: Thursday, January 18, 2007 10:17 AM
To: [log in to unmask]
Subject: [SPM] Volume change due to affine registration
Dear SPMers,
while trying to calculate the size/volume change due to affine
registration in the segmentation
approach of SPM5 I have noticed that the calculation of the volume
change in spm_write_sn.m
always results in a value of 1. I have checked this using the formula:
detAff = det(prm.VF.mat*prm.Affine/prm.VG(1).mat)
and indeed this always returns 1. However the normalisation function
(outside segmentation) is
providing the right value and the effect seems to be restricted to the
normalisation used in the
segmentation approach. This will affect the calculation of the modulated
images. Only local
volume changes due to non-linear normalisation will be applied and the
affine parts will be
omitted (which makes sense for most cases, because usually people are
interested in analyzing
local volumes which are corrected for whole brain volume). This is in
contrast to the SPM2
approach, where the affine (as constant value) and the non-linear volume
changes were included
in the modulation step.
If this is correct (and intended), the statistical analysis of modulated
images should be always
applied without total or GM volume as nuisance parameter (which makes
life easier). Is this
correct?
Best regards,
Christian
________________________________________________________________________
____
Christian Gaser, Ph.D.
Assistant Professor of Computational Neuroscience
Department of Psychiatry
Friedrich-Schiller-University of Jena
Philosophenweg 3, D-07743 Jena, Germany
Tel: ++49-3641-935805 Fax: ++49-3641-935280
e-mail: [log in to unmask]
http://dbm.neuro.uni-jena.de
|