> we have 29 SPECT scans of 29 Alzheimer patients (one scan for each
> patient) and 7 neuropsychological scores which are intercorrelated.
> We'd like to perform a correlation analysis between SPECTs and the
> neuropsychological scores by taking their intercorrelation into account.
> How can we do it? Is it feasible with SPM?
\\\ ///
\\\ ///
\\\|///
\\ ~ ~ //
( @ @ )
---------------oOOo-(_)-oOOo--------------
If your scores are inter-correlated, then you could save
yourself some degrees of freedom by reducing them with a
principle components analysis (PCA). If the scores are
all scaled about the same, and stored in a 29x7 matrix (X),
then you PCA to reduce them to the first n PCs by:
n = 3; % for example
[U,S,V] = svd(X,0);
reduced = U(:,1:n)
If you want to see how much information is lost this way, then
the data can be approximately re-constituted and compared with
the original by:
X1 = U(:,1:n)*S(1:n,1:n)*V(:,1:n)';
disp(X - X1)
If the scores can be adequately represented by a single component,
then you can use a t test in order to look for positive and negative
correlations. ( Note that you may need to negate the component).
Another alternative would be to try an F contrast, which would show
you regions where any combination of the scores, will explain some
of the differences among the data.
Best regards,
-John
------------------------------------------
--
Dr John Ashburner.
Functional Imaging Lab., 12 Queen Square, London WC1N 3BG, UK.
tel: +44 (0)20 78337491 or +44 (0)20 78373611 x4381
fax: +44 (0)20 78131420 http://www.fil.ion.ucl.ac.uk/~john
|