Print

Print
Dear Gwenaëlle, thanks a lot. I see your point but I guess it might be only  the case when I am looking at bivariate correlations between the subject course of the component and the subject variable ( eg. Age). In my case this bivariate correlation is near to zero, however when I control for effect of Total brain volume and sex in a partial regression model, I get a correlation value around 0.2 which I think in the case of my subject variables and expected effect sizes could not be neglected. 
We know ( and also it has been shown in linked ICA 2012 paper ) that Total brain volume (TIV) of subjects do correlate significantly with area measure ( and in my data actually TIV bivariate correlation with subject course of this component is 0.8) this might mean that by controlling for TIV I am taking out the effect of area measure on this component to a very high extent. If this is true then I guess I should somehow be able to show it that really the vbm changes, although explaining a very tiny part of variance of this component, are related to my subject variable and not the area. 
These are only my thoughts and concerns. I will be very happy to read your comments on it and if you have any idea how I can manage to test them.
Thanks a lot again 
Shahrzad
Sent from my iPhone

On 14 May 2015, at 15:41,Gwenaëlle DOUAUD <[log in to unmask]> wrote:
Hi Shahrzad,

After discussing this with Adrian (Groves), it seems like it actually wouldn't be such a good idea to try disentangling the different contributions of each modality to the subject load.

The key here is to look at the overall modality weightings for your component of interest (say 54% for freesurfer area versus 8% for subcortical VBM-derived volumes).

Cheers,
Gwenaelle
 
------------------------------------------------------------------------------------------------------------------------
Gwenaëlle Douaud, PhD
Associate Professor & MRC Career Development Fellow

FMRIB Centre, University of Oxford
John Radcliffe Hospital, Headington
OX3 9DU Oxford UK
Switchboard: +44 (0) 1865 222 493
Fax: +44 (0) 1865 222 717

www.fmrib.ox.ac.uk/team/principal-investigators/gwenaelle-douaud
www.fmrib.ox.ac.uk/research/fmrib-interface-analysis-clinical-neuroscience-group
-----------------------------------------------------------------------------------------------------------------------

De : shahrzad kharabian <[log in to unmask]>
À : [log in to unmask]
Envoyé le : Mercredi 13 mai 2015 8h54
Objet : [FSL] Fw: flica.......separate contributions of different modalities?

Dear Group, Since I received no answer to my previous question, I  will resend it and describe in more detail what I need.

My question:
Linked ICA provides a subject course which is shared among the modalities and a modality-specific spatial map for each component. 
I am wondering if, it is possible to get subjects-courses of each modality for each component separately, as well?
So, by this I mean if it would be possible to separate the contributions from each modality to have for each modality slightly different estimate of the particular subject course? (as it is written in Groves 2012 paper?)
I need this since I have some components that show for example increased area (freesurfer) and at the same time decreased subcortical volume (VBM). In order to be able to interpret correlation of the subject course of this component with a subject variable (e.g. age), I am a bit lost which of the modalities are deriving the association.
Do you have some idea how one can access the subject-courses per modality? I appreciate any help on this issue.
thanks a lot, 
shahrzad



----- Forwarded Message -----
From: shahrzad kharabian <[log in to unmask]>
To: FSL - FMRIB's Software Library <[log in to unmask]>
Sent: Monday, May 11, 2015 3:56 PM
Subject: flica.......separate contributions of different modalities?

Hi, 
I am wondering if it would be possible to "separate contributions of different modalities, yielding slightly different estimates of the particular subject-course", as is mentioned in the Groves 2012 paper?

Thanks a lot
Shahrzad