Hi Alvaro
> I'm conducting a research of dyslexia in children, trying to compare patient group activations against a control group.
>
> I've performed first level analysis to several patients and controls, and now I'm trying to see the differences through a two sample t-test, which I understood exists for that purpose.
>
> 1-The problem is that we don't know how to interpret the results of this two sample t-test. For example, if I use the contrast [1 -1] to check the differences between Patient and Control, what do the resulting clusters mean? Are they areas of the brain that are activated on the patient group which are not active in the control group? Or simply areas which get more significant signal intensity?
The [1 -1] contrast tells you areas where the first group are more
active than the second group; i.e., where the average parameter
estimate for that group is significantly higher than the average
parameter estimate of the second group, given the variability in the
data.
> I have attached the Fixed effects analysis of both groups, and the results of this 2sample t-test. I was expecting to see more activated regions, because group 2 almost shows no clusters. On the other hand, if I use a [-1 1] contrast to see G2>G1, I see no suprathreshold voxels. Could somebody please help me understanding this?
There are several things going on here. First, you don't have very
many subject, and with so few subjects in each group, it is unlikely
you will see very much. So, that is probably the main thing.
Second, you're using a fairly conservative threshold of FWE < .05
voxelwise. If you do something like .001 or .005 (uncorrected)
voxelwise, and then correct for cluster extent (which is also a
reasonable/valid correction), you may see something.
Third, you've done fixed effects analyses on both of the groups,
rather than random effects. With a small N things get tricky, but
it's not surprising that you would find differences between fixed and
random effects (e.g., huge difference in degrees of freedom). I
suspect if you did a random effects analysis on both groups, you would
see less significance (the utility of doing a random effects analysis
on so few subjects is questionable for science, but might help you
figure out what's going in in your data).
Finally, it is possible to have group 1 differ from 0, and group 2 not
differ from 0, but still be no differences between group 1 and group
2. That's why it is always necessary to explicitly test the
significance of group differences (rather than using the
presence/absence of a cluster to infer group differences). Sometimes
plotting the effect sizes and variability for each group can help
understand if this is going on (using, for example, MarsBar to extract
parameter estimates for a region for a region of interest).
> 2-Is there a way of comparing the fixed effects resultant con*.img of both groups to see the differences?
You can open up both in CheckReg and visually compare, or use ImCalc
to create a difference image (i.e. select both con* images and simply
i1-i2), and look at that. (But in general, as I said above, you would
probably want to do a random effects analysis on each group.)
hope this helps,
Jonathan
|