Thanks for replying.
On Thu, Apr 7, 2011 at 9:06 AM, Stephen Smith <[log in to unmask]> wrote:
> Hi
> On 6 Apr 2011, at 09:41, Cornelius Werner wrote:
>
> Hi list,
>
> I am trying to complement a traditional GLM analysis of a rather
> lengthy and complex fMRI study with MELODIC results. I am prticularly
> interested in components correlating with event timecourses (six
> types), which were displayed randomly across subjects. This means that
> tensorial ICA is out. I ran individual ICA on all single subjects and
> am now identifying correlated maps manually in order to somehow
> group-average them later. This is where I need help:
>
> - can I get component numbers correlating with my EVs in an automated
> fashion for each subject?
>
> Yes- in the GUI, final tab, you can enter a design matrix for exactly this
> purpose - the results from correlating the component timecourses against the
> model you provide are included in the web-page reporting
That's what I did. Can I filter out significant ones via an existing
script? I suppose that there is some perl or python way of parsing the
html-file, but I am not that good at this. Well, I still can do it
manually.
> - which data should I feed into a group analysis?
>
> I guess you need to identify (ideally just one strongest) component from the
> above to feed into the group-level result…..
> Or - you could use concat-mode group-ICA and dual-regression, to maybe get
> the above without utilisation of the GLM
In this case, I would need to check the dr_stage1 output files for a
correlation with my EVs of interest, wouldn't I? Do I need to convolve
my EV with an HRF, as the timecourses in dr_stage1 are sampled from
the data?! Or is the dr_stage1 output already "deconvolved"? Actually,
I am not sure about DR in this case - there are no separate groups,
all subject got to see all stimuli, but in a randomised fashion. Sorry
if I didn't make that clear. In any case, thanks a lot for the help,
as always!
Cheers,
Cornelius
>
> For the latter part, I imagined fslmerge-ing maps that a) belong to
> one EV-type and b) show spatial similarity as determined by fslcc,
> meaning separate parietal and frontal maps relating to a particular
> would not be thrown together (but analysed separately). Then I'd be
> runnning randomise on the image stack. But: which data should enter
> this set: the raw component from melodic_IC or the probmap?
>
> Sure - or you can do what i suggest above. You could try either using
> melodic_oIC (which doesn't include normalisation using the 'noise') or the
> z-stat version melodic_IC, depending on your preference
> Cheers
>
> And how
> should I deal with subjects not exhibiting a fitting map? Did anyone
> perform something similar before?
>
> Thanks for any help,
> Cornelius
>
> --
> Dr. med. Cornelius J. Werner
> Department of Neurology
> RWTH Aachen University
> Pauwelsstr. 30
> 52074 Aachen
> Germany
>
>
>
> ---------------------------------------------------------------------------
> 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
> ---------------------------------------------------------------------------
>
>
>
>
--
Dr. med. Cornelius J. Werner
Department of Neurology
RWTH Aachen University
Pauwelsstr. 30
52074 Aachen
Germany
|