Dear SPMers,
I'm doing analysis of PET scans and the results I'm getting are pretty puzzling. I'm doing a simple two-sampe t-test of two Group of scans and I'm getting a small but extremely widespread (3/4 of the brain. in absolutely random places of each kind) increase of signal in one of the two groups compared to the other.
Now, since FDG PET activity values are meaningless (they depend on injected dose, acquisition time, patient weight etc...etc..) I'm telling the batch to scale each image to its global mean in the factorial design section.
Ideally, I would expect the con image average to be 0. If I force the mean of every image to be the same, the average difference should be zero (or am I wrong here??). So I've tried reading the con file (with spm_read_vols) and computing the mean of all the non-masked pixels in the matlab command line. mean(myConVolume(~isnan(myConVolume))). Turns out that this mean is 0.3, which is very close to the median of the histogram of the absolute con values. How can that be? Am I missing something about proportional normalization?? (If I histogram the con voxels I see a gaussian distribution which is more than a standard deviation away from zero!).
Can this be something resulting from the masks??? (Just wondering, I don't have any clue!!)
My settings are: two sample t-test, Independence: yes, variance: unequal, Grand Mean Scaling: yes, ANCOVA: no, Masking: threshold, relative, 0.8, global calculation: mean, Global normalization: overall grand mean scaling, yes, value: 6.5. Normalization: proportional.
Thank you,
Luca
Rispetta l’ambiente: non stampare questa mail se non è necessario.
Respect the environment: if it's not necessary, don't print this mail.
|