Hi FSL community
Following up this kind of analysis...
I have two groups of controls and patients, all acquired in the same scanner, but in two different moments, and between the first group and the second an upgrade on the scanner's protocols was performed. Is there any hope I could combine these groups?
Thanks in advance for your attention...
Nancy
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Michael Harms
Sent: Thursday, April 21, 2011 9:31 AM
To: [log in to unmask]
Subject: Re: [FSL] Multi-scanner study
Hello Jon,
In principle, you can only ignore the possible effect of scanner if each
group was balanced across the three scanners, and even then, accounting
for scanner might soak up some residual variance and thus improve your
power to detect group differences.
However, it sounds like the patient groups were not at all balanced
across scanners, but in fact each patient group was acquired on a
completely separate scanner. In this case, the effect of patient group
and scanner is hopelessly confounded, and cannot be disentangled.
Sorry, but if that is indeed the case, there is nothing you can do, and
I would say that the data is not suitable for group comparisons.
Best,
-MH
On Thu, 2011-04-21 at 09:49 +0100, Jon Brock wrote:
> Hi,
> Any advice appreciated.
>
> I have data from 4 groups of subjects (Controls and 3 Patient groups)
> acquired on 3 different scanners. If possible I would like to combine
> these data together but I am unsure how to model scanner as a covariate. If I
> combine the Control data together then all three scanners will be represented
> in this group. Would it be correct to assume that I can safely compare any
> patient group with the control group and not worry about any potential effect of
> scanner? What happens if I want to compare between patient groups? In this case
> each Patient group has been acquired on a different scanner so I think I need to
> make sure there are no main effects of scanner or scanner x group interactions? Any advice
> on what a suitable design matrix and contrasts would look like would be greatly appreciated.
>
>
> Cheers
CLAIB 2011 www.claibhabana2011.sld.cu
V Congreso Latinoamericano de Ingenier?a Biom?dica/ V Latin American Congress on Biomedical Engineering/ V Congresso Latino-Americano de Engenharia Biom?dica
Organizado por/ Organized by/ Organizam:
Consejo Regional de Ingenier?a Biom?dica para Am?rica Latina (CORAL)
Sociedad Cubana de Bioingenier?a
Mayo/May/Maio 16 - 21, 2011, Palacio de las Convenciones, La Habana, Cuba
|