Hyo-Jeong,
> I'm new at SnPM and I'm struggling to analyze FDG-PET data of 3
> groups (1scan/sub) using SnPM. Number of subjects are
> 7(Gr1-patient1), 9(Gr2-patient2) and 16(Gr3- controls). Gr1 and Gr2
> have 2 disease-related covariates, and Age(for all 3 groups) will be
> treated as a nuisance variable. I installed snpm3 with spm2,
> matlab7 in windows XP machine.
>
> As I have no idea how to put three groups and three covariates(Gr3
> doesn't have two of them!) in one model, firstly I tried '>2 groups'
> plugin provided in SnPM3. I took 1000 permutation. And as usual for
> all SPM beginners, I got error message, as below.
To answer the covariate question, there is no good answer: When you
include a covariate you have to specify a value for every measurement.
Entering "zero" for the subjects that don't have them seems bad, as
you are then assuming that those subjects have a possibly very extreme
value. Better is to assign the subjects without covariates the mean
covariate value, as this should have less of an effect the slope
estimate.
There are two ways of setting the missing covariate values. First is
to compute the mean of the covariates you have and enter the mean for the
missing values. The other way is to take the covariate values you
have and center them (subtract off the mean) and then enter a value of
0 for the missing covariates. Both should yeild the exact same
results (in terms of P-values and F/T-stats).
> SnPM: snpm_cp
> ========================================================================
> Initialising...
> Working on correct permutation...
> ??? Error using ==> snpm_cp
> No voxels in brain
>
> I tried 2groups two-sample t-test to compare two patients group(Gr1-Gr2)
> and got same error again.
This sounds like an analysis threshold problem. First of all, have
you run the exact same data in SPM successfully? While SnPM offers
something SPM doesn't (nonparametric corrected and uncorrected
P-values and thresholds) it isn't as feature-rich in terms of results
interrogation. So I always start out by running SPM on my data (in,
say, adirectory named 'anlaysis-P'), then run the equivalent analysis
in SnPM (i.e. with no variance smoothing, in a 'analysis-nP'
directory) and then, if my DF are much less than ~20, I'll also do a
run with variance smoothing (in a 'analysis-nPs' directory.)
Second, what answers did you supply for
'Select global normalisation'
'Select global calculation'
'Threshold masking'
'Grand mean scaling'
(and, if you ran SPM, how did they differ).
And, as a last effort, have you reviewed all your data using CheckReg?
I always do this with as many as possible (up to 12 is usually feasible).
This will check that you don't have origin/world space problems. To
see if you have scaling problems, right-click in side one of the
images and select 'Image->Window->Global->auto'. If one of the images
is super-dark, it could suggest there are problems.
-Tom
-- Thomas Nichols -------------------- Department of Biostatistics
http://www.sph.umich.edu/~nichols University of Michigan
[log in to unmask] 1420 Washington Heights
-------------------------------------- Ann Arbor, MI 48109-2029
|