Hi
My data comprises vertexwise geometrically-driven scalar values
spanning the whole cortical/subcortical meshes, which is reshaped to a
vector. Now, in contrast to performing a massive univariate analysis
to find correlation with a variable (a genotype) across population, I
have hypothesized that, generally, genetic factors are likely to leave
some "co-variating" fingerprints on the brain geometry, which can be
extracted/blindly separated by melodic. In other words, I was planning
to use subject weightings associated with each component as a summary
phenotype and see if any genetic factor modulates each component's
"timecourse" across population, hence find out which IC(s) are
statistically influenced by a genetic variation.
The problem: The subject-wise weighting value per component "mixes"
the contribution of both noise and useful signal in that IC, so it
doesn't sound an optimal summary phenotype for feeding into next level
statistics. How can I achieve a summary value for each IC which is not
confounded by spatial regions that are absent in the thresholded maps?
(Actually, some other reasons exist that I can not incorporate a
simple permutation-based univariate analysis in the study design, not
fitting in a short email.)
Thanks,
Sourena
On 11/14/12, Stephen Smith <
[log in to unmask]> wrote:
Hi - I'm afraid it's hard to answer as it's not clear what your question is
- i.e., what you are trying to achieve with your ICA-based analysis of
non-FMRI data?
Cheers, Steve.
On 14 Nov 2012, at 10:27, Sourena Soheili wrote:
Dear Steve
I understand it, and actually that is the problem. But, is there a
pipeline to do something like a "back reconstruction" of the time courses
after thresholding the components? At least, does it mathematically
sound?
Sourena
On Wednesday, November 14, 2012, Stephen Smith <[log in to unmask]>
wrote:
The mixture modelling and following thresholding are done at the end on
the spatial maps - so none of that affects the "timecourses" (subject
weights per component).
Steve.
On 13 Nov 2012, at 19:43, Sourena Soheili wrote:
Hi Professor Smith
Thanks for clarifying. Another related issue, I guess that doing
statistics on loading parameters of each IC across population is
relevant. But, using this original parameter will lead to ignoring an
important part of MELODIC, that being Gaussian mixture modeling and
noise thresholding of components. Can there be a method to also
incorporate this de-noising step for production of a more robust IC
representative? I was thinking of simply using thresholded IC maps as a
weighted ROI, but wonder if a better strategy exists.
Thanks for your time,
Sourena Soheili-Nezhad
On Tue, Nov 13, 2012 at 8:59 PM, Stephen Smith <[log in to unmask]>
wrote:
Hi - yes - but MELODIC doesn't do temporal prewhitening by default
anyway.
Cheers.
On 13 Nov 2012, at 14:45, Sourena Soheili wrote:
Hello FSL community
This has been previously discussed in the forum that MELODIC can be
used for analysis of structural (non-fMRI) data like a single subject
2D approach where the temporal dimension is substituted by #subjects.
But, as there should be no "temporal" autocorrelation present here,
does application of temporal pre-whitening sound irrelevant in this
type of analysis?
Thanks in advance,
Sourena Soheili-Nezhad, MD
---------------------------------------------------------------------------
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
---------------------------------------------------------------------------
---------------------------------------------------------------------------
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
---------------------------------------------------------------------------
---------------------------------------------------------------------------
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
---------------------------------------------------------------------------