Christian,
I think we are under the impression that ICA is looking for significant points of activation (be they physiological or the actual behavior we are looking for), and that given that we have 117 (I don't know why I put 180- it is 117) events in a session, we should be seeing 117 components. If this is not correct, then what are we seeing from the ICA? If anything I think we expected to see more than 117, to account for 'other' activations not related to the tasks we were presenting, so to see fewer seemed odd. Also, 'assuming our assumption' was correct about the components, and since Melodic can't/doesn't label them per event (& how could it know?), I re-ran it, manually entered 117 components to look for, and then matched our components to those as they occured in time (thinking the component are temporal regardless). Obviously, if our understanding of ICA is incorrect, the last bit is also incorrect as well, so it may just be that we need to understand what ICA is act
ually giving us (and can you tell me that?)
Regards,
Emily
----- Original Message -----
From: "Christian F. Beckmann" <[log in to unmask]>
Date: Friday, June 13, 2008 1:11 pm
Subject: Re: [FSL] Melodic Components: Interpretation
> Hi
>
> Don't quite understand what you mean by 'should have 180
> components':
> melodic will estimate a rough ball-park figure of how many
> components
> are estimable relaibly from your data. Looks like melodic thinks
> that
> it doesn't make sense to extract beyond 108 components. Or do you
> meant hat the corresponding time courses are of length 108 only,
> even
> though your original data is of length 180? If so, something is
> wrong,
> most likely with the input file you;re feeding into melodic. Check
>
> using fslsize that the input really has as many volumes as you
> think
> it should have.
> hth
> Christian
>
> On 11 Jun 2008, at 21:27, Emily Stoneham wrote:
>
> > Hello all!
> >
> > I have been searching the archives, and am not finding exactly
> what
> > I am looking for, so I will see if
> > I can explain it here with specifics.
> >
> > I have run melodic on a single subject that we know 'should'
> have
> > 180 components (one for every
> > image response). There are 3 different types of images that are
>
> > presented (randomly), and the
> > design requires the 3 column format to run in FEAT because it is
> a
> > self-paced task. My question is:
> >
> > We a) only see 108 components, and b) I cannot figure out how
> to
> > correlate them with the response
> > tasks. Is there a way to do that?
> >
> > Regards,
> > Emily
>
|