As nobody has answered yet, I will share my thoughts on this.
To me it seems important that you have acquired your data for patients
and controls in a randomised mixed manner. This means that you
hopefully have an equal number of controls and patients scanned with
head coil A and with head coil B. Therefore differences in SNR will
contribute equally to both groups, meaning that your differences
between patients and controls are not driven by head coil.
I do not know on how to transform your data as you have asked but I
would consider testing head coil effects using different GLMs. If you
do not find differences due to head coil on your condition effects you
might not need to transform your data.
I would test GLMs:
1) include head coil as a covariate
2) model every head coil plus subject group as a different group, this
will give you 4 groups (patients A, patients B, controls A, controls
B) in your statistical design and you can investigate if there is a
main effect for head coil in your data and test for condition and head
coil interactions (see for contrast settings the tutorial from Jan and
Darren). The idea to model different sequences/scanners/head coils as
separate groups was posted earlier by Christain Gaser for VBM
data-analysis. But I think it is also might be useful in your case to
investigate head coil driven main and interaction effects. With a
conjunction you can test for an overlap between head coil A and B.
However, if you have acquired your data for patients with head coil A
and your controls with head coil B you might find yourself in a tricky
situation as differences found between controls and patients might
simply be due to the SNR differences between head coils. Even if you
only take within-subject-effects to the second level, you can never
find out what the interaction is between head-coil and differential
BOLD responses. Or am I too pessimistic now?
Simone Reinders, PhD
King's College London
Institute of Psychiatry (IoP)
De Crespigny Park
On Mon, Apr 7, 2008 at 7:45 AM, Amit Etkin <[log in to unmask]> wrote:
> Hi all,
> I have two data sets (both with patients and controls) acquired for the same
> task on the same scanner, but with two different head coils (due to an
> annoying technical issue with co-acquired physiological measures). Is there
> any way to transform to transform the data (to effect size?) to account for
> the possibly slightly different SNR and variance across in order to pool
> across these data sets?