Hi - you probably mean larger veins - you're right that you need to be careful - if you just regress out the full variance of such components, you may "hurt" the RSNs - this is equivalent to the "aggressive" cleanup option in FIX.   Alternatively you can do the default FIX thing:  run the full regression against all timecourses (good and bad), and just remove the space of the bad ones after that (see inside FIX scripts for further details, if that's not obvious).  

Cheers.

On 16 May 2014, at 10:10, Andreas Lidström <[log in to unmask]> wrote:

Ok, thanks. One last thing; Would you regress out components that are neatly capturing say extra cerebral tissue? That is, the time and frequency plots looks like good components but they are not components of interest, or does this hurt the rest of the data?

2014-05-16 9:31 GMT+02:00 Stephen Smith <[log in to unmask]>:

Hi - I would still recommend using auto-ICA-dimensionality for cleanup, even if you're not going to use FIX.

Cheers.

On 16 May 2014, at 09:26, Andreas Lidström <[log in to unmask]> wrote:

Thanks for your answer. I know that automatic determination of components is recommended for cleanup, but our issue is that since we have run MELODIC on data that has not been processed in FSL we do not have the files that FIX needs to run, so I will have to classify the components myself. Is there a reasonable number of components to limit the automatic determination to since I have ~200 subjects?

2014-05-16 7:57 GMT+02:00 Stephen Smith <[log in to unmask]>:

Hi

On 15 May 2014, at 17:29, Andreas Lidström <[log in to unmask]> wrote:

Hello,

I want to do manual cleanup of my data after running MELODIC and I have two questions:

For our future group-analysis we will use a threshold of 0.99 and 25 components, but I am wondering if using 0.5 may be better if you want to detect noise-components?

I'm afraid I don't really understand the question - I don't understand what thresholding you are referring to, that would interact with detection of noise components?

Allso, is it ok to use only 25 components in this first 'cleanup-ICA' or will this produce components that are mixed with noise, and therefore not regressable?

We recommend using auto-dimensionality estimation for within-run ICA and artefact removal.

Cheers.

I appreciate any insight

//Andreas

---------------------------------------------------------------------------
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

---------------------------------------------------------------------------

Stop the cultural destruction of Tibet

---------------------------------------------------------------------------
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

---------------------------------------------------------------------------

Stop the cultural destruction of Tibet

---------------------------------------------------------------------------
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

---------------------------------------------------------------------------

Stop the cultural destruction of Tibet