Hello
a few years ago I posted a message on this list asking for advice about
the use of customised templates (copied below). After carrying out
analyses I found diffuse differences in both WM and GM between groups
and times. I am now asking for advice in responding to a query about
"group heterogeneity and patient population used to generate the
customized template a priori". How can I tell whether these differences
are truly anatomically significant at a voxel group level or small-scale
structural level? Or whether they rather reflect- globally and
topographically - that subjects' brains are very different to each other
and to those of controls?
Any advice on how to tackle these problems would be much appreciated.
Best wishes
Chiara Nosarti
__________________________________
Dr Chiara Nosarti PhD, Lecturer in Cognitive Neuropsychology,
Division of Psychological Medicine,
Room M5.01.05
Section of General Psychiatry, PO Box 63,
Institute of Psychiatry
16 De Crespigny Park, Denmark Hill,
London SE5 8AF, UK
Tel: 0207 848 0133
Fax: 0207 701 9044
> I am carrying out analysis of changes in brain structure at T1 (14
yrs) and
> T2 (18 yrs) in a group of individuals who sustained brain damage at
birth
> and age matched controls. For T1 I created local templates for the 2
> groups, by normalizing a set of T1 images, averaging the normalized
> images, and smoothing by 8 mm. If I do the same for the 2 groups
separately
> at T2 does it make sense to compare the 2 times with different
templates?
> or should I be creating a template for my cases and controls with
images
> taken at T1 AND T2 ? Any advice would be much appreciated.
Date: Thu, 4 Nov 2004 21:21:49 +0100
Reply-To: John Ashburner <[log in to unmask]>
Sender: "SPM (Statistical Parametric Mapping)" <[log in to
unmask]>
From: John Ashburner <[log in to unmask]>
Organization: FIL
Subject: Re: template query
Comments: To: Chiara Nosarti <[log in to unmask]>
In-Reply-To: <025801c4c282$372b5420$e28d53c2@ChiaraNosarti>
Content-Type: text/plain; charset="iso-8859-1"
Content-Disposition: inline
I would suggest using the same template for all subjects and times. I
would
approach the problem in one of two different ways:
1) Coregister Early and Late scans together (no reslicing).
2) Spatially normalise. This could be done via the optimised way, which
would
involve segmenting the Early, segmenting the Late, and averaging the GM
with
ImCalc. Estimate the spatial normalisation paramaters by matching the
averaged GM to the grey matter template, and write the original images
spatially normalised (changing the voxel sizes to 1mm in the defaults).
3) Segment the early and late spatially normalised images.
4) Modulate.
5) Smooth.
6) Do a paired t-test type of statistical analysis.
Alternatively you could use the Deformations toolbox.
1) Coregister Early and Late scans together (no reslicing).
2) High-dimensional within-subject registration between your early and
late
scans. This would produce a deformation field (y_late.img). This takes a
long
time to run, and you may need to tweek some of the settings (e.g. maybe
try
more iterations.
3) Then write the Jacobian determinants out as an image (jy_late.img).
4) Spatially normalise by matching the early.img to the standard space,
and
write spatially normalised versions of the early.img and the jy_late.img
(1mm
voxels).
5) Segment the wearly.img.
6) Use Imcalc, selecting the waerly_seg1.img and the wjy_late.img, and
enter
the following expression: i1.*(1-i2)
7) Smooth the output of ImCalc.
8) do the stats on the smoothed images.
It's a bit of a convoluted procedure, but if you wait for a paper by
Kipps et
al in JNNP, then it should all become more clear.
Best regards,
-John
|