Hi - there's no fixed rules on this.

I would say that if your patients are not that different* from the controls then you may as well use all subjects in the group ICA.

If they are quite different then you might want to do the group-ICA on just the controls.

Cheers.

* you might not know this until you've run the dual-reg for a first time - but you may be able to tell from the structurals.  If the structurals look very different between the groups you might want to use those as confounds in the final randomise - see feat_gm_prepare.





On 1 Aug 2011, at 23:41, Minjie Wu wrote:

Hello,

We are trying to analyze the resting state data using group ICA (melodic and dual regression).  I have one question regarding the group ICA.

Do the groups in Melodic need to be balanced in order to find the common spatial independent components?
In our case, we have 14 controls, 15 early patients and 15 advanced patients. Do we need to randomly choose 7 early patients and 7 advanced patients to match the 14 controls ( 14 patients vs 14 controls) for Melodic?

Or we could simply run Melodic with all subjects (14 controls, 15 early patients and 15 advanced patients)?

Thank you very much for your help.

Minjie



---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director,  Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
+44 (0) 1865 222726  (fax 222717)
[log in to unmask]    http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------