Hi Tom,
Tensor-ICA assumes identical temporal characteristics across different
subjects/sessions for each component. This is not the case in your
setup - though, if timing is only 'slightly' different it might still
work in practice. Anyhow - my guess is you'd be best advised to run
the 'concatenation' approach and then use GLM analyses on the separate
time courses to check the association with your design. The individual
temporal responses are saved in the tXX.txt files (where XX is the
component number). For N subjects/sessions the temporal responses are
the last N columns in that file.
hth
Christian
On 5 Mar 2008, at 10:35, Tom Johnstone wrote:
> I have a few questions on how I'm setting up a multi-session ICA.
>
> I have 5 scan sessions from the same individual. In each scan session,
> the task was slightly different (e.g. different stimuli/response) and
> the timing was also slightly different.
> What I would like to do is identify components that are:
> i) common across all sessions
> ii) unique to specific sessions and correlate with stimulus timing for
> that session
> iii) unique to certain combinations of sessions and correlate with
> stimulus timing for those sessions
> iv) for ii) and iii), it would be even better if I could show the
> uniqueness of the components to those sessions in some way (i.e. who
> that they exist for those sessions but not for the others)
>
> So far, I've done a pretty stock standard multi-session Tensor ICA,
> but two questions arise:
> a. how do incorporate a different time series model for each session
> (in the post-stats part)?
> b. what's the best way of getting at ii and iii above - when I set up
> a session-subjects model with multiple higher-level contrasts, I don't
> seem to get any output regarding those contrasts?
>
> Would I be better off doing separate ICAs on each session and then
> performing higher level GLM on the resulting components? If I do this
> though, how do I match up components across sessions? It seems to me
> that multi-session tensor ICA should in principle allow me to answer
> these questions - just need a bit of guidance perhaps.
>
> Many thanks,
> Tom
>
>
> --
> School of Psychology and CLS
> University of Reading
|