Print

---

'lo, Christian.

Thanks for getting back to me. I can certainly stipulate some number of
ICs, but how many? I don't have an a priori hypothesis regarding the
expected number, though maybe not 1 or 50. My other thought was simply to
down-sample my data to about 900 time points - still many more than typical
resting state, I know, but the auto estimation still gives me only one
component (explaining some 76% of the variance). Of course, it may be that
this one component actually does describe my data best....is there
something I'm missing here, some kind of threshold below which the
automatic estimation is most robust?

Again, many thanks,
Paul

### Me code snippet (5 ICs - why not?):
fsl5.0-melodic -i 23All_dsamp2Hz.nii.gz -o ICA_23All_2Hz --nomask --nobet
--vn --report --tr=.05 -v --Ostats --dim=5


On Wed, Oct 3, 2012 at 3:33 PM, Christian F. Beckmann <
[log in to unmask]> wrote:

> Hi
>
> Just switch off the conversion (i.e. *not* use the --pbsc option which
> switches on conversion to percent BOLD signal change ;)
> I suspect you have a great number of time points? If so, you should
> consider bypassing the dim. estimation and setting a number explicitly with
> the -d option.
> hth
> Christian
>
>
> On 1 Oct 2012, at 20:40, Paul Robinson <[log in to unmask]> wrote:
>
> > Greetings, FSLers!
> >
> > I've got some non-fMRI data in I'd to get ICA maps for, and am wondering
> if I can't do it with MELODIC. Well, melodic, really ;)
> >
> > I try to run (from within a subject's data folder):
> >
> > melodic -i my_4D_data.nii -o ICA --nomask --nobet --vn --pbsc --report
> --tr=.05 -v
> >
> > and standard output still tells me "Converting data to percent BOLD
> signal change". Do I need to supply boolean values to these flags (e.g.,
> "--nobet=1"; not that it seems to make a difference when I do, as I get the
> same output). My data are sampled at 20Hz.
> >
> > Any help or pointers would be most appreciated.
> >
> > Thanks,
> > Paul
> >
>