Hello,
I have a question about the voxelwise analysis with SPM5 of some DTI
data (FA/MD images aligned already).
I want to do a one-vs-others analysis to detect regions of FA
reduction for the 1 patient with respect to the group of 32 normal
controls. With age and sex as confounding variables and the ANCOVA
selections shown at the end (which trigger the inclusion of the last
two columns in the design matrix), the design matrix would look like:
1 0 age1 sex1 ANCOVA_1_1 ANCOVA_2_1
1 0 age2 sex2 ANCOVA_1_2 ANCOVA_2_2
...
1 0 age32 sex32 ANCOVA_1_32 ANCOVA_2_32
0 1 age33 sex33 ANCOVA_1_33 ANCOVA_2_33
If (b1, b2, b3, b4, b5, b6) would give the fitted values of the GLM
model, then b1 would represent the level of the FA for the group of
32, while b2 the FA level for the single patient. So a t-contrast like
c = (1 -1 0 0 0 0)
would measure the value b1-b2 and, since the t-test is one-sided, it
would give the areas/voxels where the FA in the single patient is
significantly (to whatever degree) lower than those in the group of
controls, hence my areas of FA reduction.
Is this right or am I missing something very important? I am asking
this because I do not detect areas where FA would be reduced with this
contrast (to a certain degree of significance), but if I chose the
contrast for areas where FA would be higher in the single patient than
the rest, i.e. c=(-1 1 0 0 0 0), then I would get quite a few
significant areas (to the same degree of significance), and I am a
little puzzled by that.
=========== ANCOVA - related selections in the GUI ===========
(is there concern about these choices on the 2-nd level design specification?)
Design:
Two-sample t-test
ANCOVA - yes
Covariates:
age
sex
(for each of them I chose "Centering -> As implied by ANCOVA")
Global normalisation:
Normalisation -> ANCOVA
==============================================================
Thank you very much,
Silviu Podariu
UNMC, Omaha, NE
|