kentman234 Wrote:
-------------------------------------------------------
> nick Wrote:
> --------------------------------------------------
> -----
> > kentman234 Wrote:
> > > - I want to go data-driven. That's why I
> > mentioned
> > > ICA, to descriminate the components of the
> two
> > > classes i.e. one component should indicate a
> > > unique class. You said that this is not
> > possible
> >
> > I said I did not see how that could work - feel
> > free to prove me wrong, that means I would have
> > learnt something interesting.
>
> A very good article about classification of fMRI
> data with Independent components can be found
> when we google the sentence
> "classification of fMRI independent components".
> It is the first hit.
>
> Each component is characterized using some
> parameters of a certain metric. When calculating
> the parameters of a metric for each component, we
> can determine to which class the component is
> related. The training step requires user's
> intervention in labeling the ICs. The testing step
> works automatically and consists in classifying
> the remaining components using the trained
> classifier.

It is a nice paper indeed - but their approach and research question seem quite different from what you want to achieve.

> I understand the other parts of your reply.

Glad to hear that.

best,
Nick