Dear Sean,
I suspect that your signal at present mainly represents artifact due to
imperfect normalisation of your Alzheimer patient group. The spm template is
expecting normal brains and when it attemps to match a shrunken brain to the
template (guided by reducing the least-squares difference between the
subject's brain and the template), it will interpolate cortex where there is
none. The normalisation programme concentrates on areas of maximal differece
between subject and template voxel values. I'm assuming that your patients
have big venricles (especially if some of them have,so-called, sub-cortical
dementia).This would explain why you are getting signal here.
I don't belive that there is a simple fix to this. I've been working with
others on a programme that will reduce the effect a stroke has on the
normalisation process, but this only works for brains that have focal
damage. There has been talk of seperate templates for older subjects in
recent spm e-mails; it may be that this will help, but I think that there
are fundamental problems when globally abnormal brains are submitted to
spatial normilazation in an attempt to compare them with normal subjects'
data.
Yours,
Alex.
-----Original Message-----
From: Sean Colloby
To: [log in to unmask]
Cc: [log in to unmask]
Sent: 19/04/00 12:17
Subject: Can anybody help?
Dear SPM community,
I am relatively new to the realms of SPM, I would be
extremely grateful if anyone you could help me on a few issues
regarding my analysis.
Until now I have been using the windows version of SPM to make a
group comparison between 20 controls and 40 Alzheimer patients
(SPECT Tc-HMPAO Images - voxel size 5.40mm).
I have normalised all images to the PET template, using the 12
parameter affine transform, MNI bounding box and resliced to the above
voxel size of 5.40mmx5.40mmx5.40mm. The images were then
smoothed to a FWHM of 15mm x 15mm x 15mm.
For the analysis, I then choose the option of Scaling, with a single
study and single subject containing two conditions. The controls
condition and the AD patients condition. I then set my contrast to
(1, -1) and ran the analysis.
When I looked at the glass brain data, the only regions of
significance where in the ventricles? Nowhere else, i.e.,in the
frontal, temporal or parietal regions? Not even when I set very
liberal uncorrected p values and number of voxels per cluster to 1.
Am I doing this analysis correctly? Is it my reslicing, or have I set
up the PET statistics wrongly, or maybe I have very poor data? Surely
there must be some differences!
Can anybody possibly help me and give me some useful pointers?
Thanks.
Sean Colloby
Research Associate
Wolfson Research Centre
Newcastle General Hospital.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|