Dear Xiang,
> Simplify, for 2 subjects in each of the two groups, TIVs are:
>
> group1 group2
> sbj1 2.01 2.06
> sbj2 2.11 1.99
>
> I want to do a two-sample t-test.
>
> Question1
> John's tutorial says to input TIVs in "Global values"->"User"->"Global values". But spm says "Global values" should be a X-by-1 array. And there's no option to input two TIVs for the two groups respectively. How should I input TIVs?
In your design matrix, you have one image per row; this image can be assigned to either group 1 or group 2. So don't think of subject 1 and subject 2 for each group, but think of subject 1 and subject 2 in group 1, and subject 3 and subject 4 for group 2.
Each group will have a column in the design matrix that indicates whether a subject belongs to that group (1) or not (0). So:
group1 group2
sub1 1 0
sub2 1 0
sub3 0 1
sub4 0 1
So when including TIV, you need a single array, with one image per subject. This is also true for the covariate case (see below).
>
> Question2
> Some posts said traditionally, TIVs are input as covariates. Is it correct I do as below?
>
> Covariate
> Vector: 2.01 2.11
> Name: group1
> Interactions: None
> Centering: (default) Overall mean. I have no idea what this is, should I change it to "Factor 1 mean", or "No centering"?
> Covariate
> Vector: 2.06 1.99
> Name: group2
> Interactions: None
> Centering: (default) Overall mean.
You'll just want a single covariate, called perhaps "TIV", that has values across both groups:
group1 group2 TIV
sub1 1 0 1.734
sub2 1 0 2.171
sub3 0 1 1.827
sub4 0 1 2.091
Mean-centering on the overall mean is probably what you want. If you have some values in a vector X, mean centering is just: Xcentered = X - mean(X). The mean of Xcentered will then be 0. This will make the contribution of this term to your model somewhat easier to interpret.
For a 2 sample t-test, I would expect 3 columns in your design matrix, as above. Then, to test group1>group2 (having factored out effects of TIV), you would run the T contrast:
[1 -1 0]
or conversely, gor group2>group1:
[-1 1 0]
If you are curious as to the effect of TIV (what you are accounting for by including it), you could run a contrast on that column:
[0 0 1]
There is a longer discussion of TIV as a covariate in Barnes et al. (2010).
Barnes J, Ridgway GR, Bartlett J, Henley SMD, Lehmann M, Hobbs N, Clarkson MJ, MacManus DG, Ourselin S, Fox NC (2010) Head size, age and gender adjustment in MRI studies: a necessary nuisance? NeuroImage 53:1244-1255.
> Question3:
> I am using John's script (below) to calculate TIV. I figure "Vols" is the returned volume. For one subject, I got GM/WM/CSF as 0.92/0.71/0.38, and TIV was the sum, 2.02. Are these values normal? (I am just wondering the sum may be near 1).
These seem reasonable to me. In the Good (2001) paper, average TIV values were probably closer to 1.6 liters, but there is quite a lot of variability (Fig. 3g). In a recent study we've done, using standard segmentation average values are closer to 2 liters, and using the "new segment" on the same subjects gives average TIV values closer to 1.6 liters. Again, there is substantial variability across subjects, so given only 2 subjects, it's hard to know where yours lie, but they definitely seem reasonable.
Good CD, Johnsrude IS, Ashburner J, Henson RNA, Friston KJ, Frackowiak RSJ (2001) A voxel-based morphometric study of ageing in 465 normal adult human brains. NeuroImage 14:21-36.
Hope this helps. Good luck!
Jonathan
--
Dr. Jonathan Peelle
Department of Neurology
University of Pennsylvania
3 West Gates
3400 Spruce Street
Philadelphia, PA 19104
USA
http://jonathanpeelle.net/
|