Hi Christian!
back to you again..and sorry for the delay!
in reply to your reply to my second question:
the best smooth is a filter that is of matched size to the activation we wish to detect. The anatomical variability between subjects will mean that the signal across subjects may be expected to be rather widely distributed over the cortical surface. In such a case it may be wiser to use a wide smoothing to detect this signal. In contrast, for a single subject experiment, it would be wiser to use a very narrow smoothing, or even no smoothing.....
THIS IS FOR GLM ANALYSIS!..THE SAME LOGIC COULD ALSO BE ASCRIBED TO ICA (i think so,...maybe you do not agree with me)
if we are using ICA:
& if we are analyzing data of ONE subject, there is no prob; as we can do it with least gaussian (or even no guassian..as you pointed) smoothing!
but if we need data across subjects; and we do the same, we will be in danger of lossing significant points!
what is your idea?
and... what are the practical steps for what you called "concatenating"?
Regards
Nima Dehghani MD.
Schizophrenia, Cognition and Imaging Labaratoy; Department of
Psychiatry; UBC; CA
Email: [log in to unmask], [log in to unmask]
----- Original Message -----
From: "Christian F. Beckmann" <[log in to unmask]>
Date: Thu, 2 Oct 2003 11:55:50 +0100
To: [log in to unmask]
Subject: Re: [FSL] ICA...Spatial smoothing
Re: Hi
Re:
Re: On Wednesday, Oct 1, 2003, at 23:58 Europe/London, nima dehghani wrote:
Re:
Re: > Hi
Re: >
Re: > i found that MELODIC, uses a guassian kernel for spatial smoothing!
Re: >
Re: > it seems to me that there are 2 problems with this:
Re: >
Re: > 1. ICA assumes that dara are NON-guassian....so, how we can use a
Re: > gaussian kernel on a non-guassian data? is not it a fault?
Re: >
Re:
Re: Spatial smoothing is used to locally increase signal-to-noise. Any
Re: pre-processing has an impact on the source distribution, but locally
Re: convolving with a Gaussian kernel will *not* render the *entire*
Re: spatial distribution being Gaussian - in fact, if 'signal' appears in
Re: spatial clusters, spatial smoothing can indeed result in spatial
Re: histograms with higher neg-entropy (i.e. in source distributions that
Re: are 'easier' to extract due to increased non-Gaussianity) Smoothing too
Re: much, however, is problematic (mainly for the mixture model inference
Re: stage, though)
Re: I generally advice to use very low amounts of smoothing (FWHM~ voxel
Re: size). One of the nice advantages of the Mixture Model Inference stage
Re: is, however, that (unlike GRF-Theory based inference) Gaussian
Re: smoothing is no longer required. So if you're concerned about the
Re: Gaussian smoothing, just switch it off or even use some more advanced
Re: non-linear edge preserving technique like SUSAN.
Re:
Re: > 2. when one tries to use ICA for multisubject analysis, spatial
Re: > smoothing by a "unique kernel for all subjects" becomes a more serious
Re: > problem!
Re: >
Re:
Re: Melodic does not yet include any multi-subject analysis options, I
Re: suppose you're talking about concatenating data? What makes sense and
Re: what doesn't will depend on what exactly you propose to do w.r.t.
Re: multiple subjects. Per se, however, I can't see any serious problem
Re: with a unique kernel - quite the opposite, in fact.
Re: Maybe you would like to make your question more specific?
Re:
Re: hope this makes sense
Re: ta
Re: christian
Re:
Re:
Re:
Re: > do any of ICA experts havd an idea or a solution to this?
Re: >
Re: > am i right or not, with the 2 questions
Re: >
Re: > Regards
Re: >
Re: > Nima Dehghani MD.
Re: >
Re: > Schizophrenia, Cognition and Imaging Labaratoy; Department of
Re: > Psychiatry; UBC; CA
Re: >
Re: > Email: [log in to unmask], [log in to unmask]
Re: >
Re: > --
Re: >
Re: > _______________________________________________
Re: > Sign-up for your own personalized E-mail at Mail.com
Re: >
Re: > CareerBuilder.com has over 400,000 jobs. Be smarter about your job
Re: > search
Re: >
--
__________________________________________________________
Sign-up for your own personalized E-mail at Mail.com
http://www.mail.com/?sr=signup
CareerBuilder.com has over 400,000 jobs. Be smarter about your job search
http://corp.mail.com/careers
|