Hi
Hard to say, have you QCed the data? What do you get from looking at
the intensity ranges in the data, using fslstats input file -r -R -
is it the case that that subject has unusual intensities?
Also, you might want to try running this with 'variance normalisation'
switched off
hth
Christian
On 29 Jul 2009, at 23:59, M Mather wrote:
> Steve,
>
> Thanks for the information about what to expect in terms of number of
> components. We went ahead and re-did the ICA analysis prespecifying 20
> components. The subjects were responding to pictures during the scan
> (it was
> not a resting state scan) and so it makes sense that the first
> component
> involves visual regions, but what concerns me is that all 20 of the
> components appear to be driven by one person, based on the output of
> the "sessions/subjects mode" chart. Here is a PDF of Component 1:
> http://www.usc.edu/matherlab/component1.pdf
>
> and here is the rest of the output in a zip file:
> http://www.usc.edu/matherlab/group_clean_20compnts.gica.zip
>
> Sessions 37 and 38 were from the same subject, and
> all of the components show the same pattern as the one attached
> here, where either Session 37 or 38 (or both) show high values for the
> component and everything else is near baseline (one component shows
> a high
> value for one other session as well, but that is the only exception).
>
> With results like these, should we be excluding that one person or
> is this
> imbalance not a concern?
>
> Many thanks,
> Mara
>
>
>
> On Jul 25, 2009, at 6:38 AM, Steve Smith wrote:
>
> Hi,
>
> If you are analyzing multiple subjects together then this isn't
> necessarily
> an unreasonable number. The number of components that MELODIC
> estimates
> depends on many factors, including effective resolution, smoothing,
> quality
> of subject alignments, amounts of artefacts, subject homogeneity,
> SNR, etc
> etc. I would actually be a little surprised if it estimated much
> lower than
> this in a multisubject analysis.
>
> However, if this number is larger than is useful for your purposes
> (e.g. if
> you want to get group results that gives you 5-10 RSNs) then there
> is no
> reason why you shouldn't hand-set the dimensionality to anything you
> like -
> e.g. 20 is what we used in the 'low-dimensional' analyses in our
> BrainMap vs
> RSNs paper that's just come out on earlyview.
>
> Cheers, Steve.
>
>
>
> On 24 Jul 2009, at 17:37, M Mather wrote:
>
> We are analyzing some data using MELODIC (temporal concatenation
> option)
> that we did not collect ourselves and are baffled by the number of
> MELODIC
> components that result (over 100). This is much higher than we've
> seen in
> our own data and are wondering what the causes might be, or if it is
> within
> the bounds of what we should expect. There are 492 TRs per scan,
> with almost
> 40 scans included in the original analysis (each participant
> contributed one
> or two scans). Many of the MELODIC components show very high values
> for one
> person or scan and very low values for the rest of the participants.
> We have
> tried a variety of things and everything just ends up leading to
> even more
> components:
>
> 1. We went back and ran MELODIC for each individual separately and
> denoised
> their data by removing components that appeared to be motion or other
> artifacts, then reran the overall set. This did not appear to
> improve matters.
>
> 2. Because it looked like one person in particular was driving many
> of the
> components that showed a large effect size for one person and not
> the rest,
> we reran the analysis without that person. That led to a result with
> even
> more components, many driven by another person. We excluded that
> person, and
> got a result with even more components.
>
> 3. We added 10 mm to the spatial smoothing. That was also no help.
>
> Does this suggest a problematic dataset or is there something we
> might be
> doing incorrectly?
>
> Many thanks,
> Mara
_______________________________________________
Christian F. Beckmann, DPhil
Senior Lecturer, Clinical Neuroscience Department
Division of Neuroscience and Mental Health
Imperial College London, Hammersmith Campus
Rm 419, Burlington Danes Bldg, Du Cane Road, London W12 0NN, UK
Tel.: +44 (0)20 7594 6685 --- Fax: +44 (0)20 7594 6548
Email: [log in to unmask]
http://www.imperial.ac.uk/medicine/people/c.beckmann/
Senior Research Fellow, FMRIB Centre
University of Oxford
JR Hospital - Oxford OX3 9DU
Tel.: +44 (0)1865 222551 --- Fax: +44 (0)1865 222717
Email: [log in to unmask]
http://www.fmrib.ox.ac.uk/~beckmann
|