Hi Simon,
Yes, we are planning on including TensorICA in the next FSL release,
which should be this summer we hope. This will be part of MELODICv3.
You could do what you suggest, or you could normalise each subject's
4D raw data into a standard space (you might want to use more reduced
resolution than 2mm though because of RAM concerns), and then
concatenate in space or time (depending on what question you want to
ask), across subjects, and then run MELODIC. You could also reduce
the temporal dimensionality of each subject first before doing the
above, just by taking the reduced PCA output from the first stage of
MELODIC. Any of this would then be similar(ish) to what you can do
with GIFT/FASTICA.
The difference between TensorICA and the above is that it is a
fundamentally different model, as it not only models temporal and
spatial aspects of a given component, but _separately_ models the
subject (or session) variation - the data is represented as the sum
of various components, each of which is formed as the outer product
of the 3 aspects (space, time and subject variation). In our
experience this gives richer and more easily interpretable modelling
of such 5D datasets than what we could achieve previously.
Hope this helps? Cheers, Steve.
On 29 Mar 2007, at 08:51, Simon Robinson wrote:
> Dear Christian Beckmann,
>
> Are there plans to release your Tensor ICA method to enable group ICA
> analyses? Will this eventually be an extension of MELODIC?
>
> Until this comes out, am I right in thinking that the best option
> with FSL
> is to normalise single-subject fMRI data to a standard space, do a
> MELODIC
> analysis on these, concatenate components to form a single set, and
> do a
> second MELODIC analysis on that? I have found this so demanding of
> RAM that
> I have had to limit the number of components severely or keep the
> number of
> subjects small. Does TICA offer advantages over rival group ICA
> methods
> implemented in GIFT and FastICA?
>
> Thanks,
>
> Simon
------------------------------------------------------------------------
---
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
------------------------------------------------------------------------
---
|