Dear Lakshmi,
correction; the matlab command below should of course read
XUA = XA+B*BETA(n+1:n+m,:);
I was being a little too fast there, sorry.
Good Luck Jepser
Dear Lakshmi,
>
>Dear Jesper:
>
>Here are the details of the design. It was a PET study with 10 subjects. Each
>subject was scanned twice under two different conditions. The images were
>normalized to SPM96 template, and the differences between the two conditions
>were computed using SPM 96, mulitsubject-different conditions design with two
>contrasts, no global normalization and no covariates.
>
>The matlab codes for extracting the vaules from XA.mat file and the images
that
>were used for analysis seem to be working fine. They both seem to be getting
>the values from the same location and the column, and this was verified using
>various routines.
>
>We are wondering whether the differences could be due to block effect. Could
>you comment on this. Any other ideas you might or others might have would be
>appreciated.
>
You are absolutely right, the differences you describe are due to "block"
effects. The term "block" effects comes from ANOVA jargon, but is
matematically no different from any covariate you may enter into your
design. These block effects are in this case simply, for any given voxel,
for subject i the difference between the voxel mean (across scans) in
subject i and the voxel mean across all subjects. I don't quite understand
what you want to do with these values, but if you absolutely need the
"unadjusted" values you may do as follows
load XA.mat
load BETA.mat
load SPM.mat
%
% Now assume that you have n conditions in m subjects. That means
% that you will have n+m rows in BETA (which is where the fitted
% effects reside). To get the unadjusted values you should do the
% following.
%
XUA = XA+B*BETA(n+1:m,:);
where XUA will have the same dimensions as XA, but contain data unadjusted
for block effects. Remember also to use no "grand mean scaling".
Having said this, I would think that whatever processing you intend to do
to this data your life would be a lot easier if you used the adjusted data.
Good luck Jesper
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|