JISCMail - FSL Archives

On 3 May 2014, at 22:23, marco solmi <[log in to unmask]> wrote:

Dear Mark,

Thanks for the response. I would ask you some questions more..
I have images with more than 1700 time points..
- I will reduce the components, do you think 100 could be ok? or something like 35? Any criteria in this choice?

normally we let MELODIC decide on dimensionality for single run analysis. However you can limit this to a max of (eg) 200 with

-d -200

the minus sign before the 200 tells it to auto-estimate up to a max of 200

I know that FIX works automatically detecting noise components, but I can not use that because I have no R and Matlab.
- Do you have a sense on when a standalone version will come out?

not any time very soon. However - see the latest version - which can be run without a matlab license either using the precompiled-matlab or with octave. R is freely available so you can get that.

In the meanwhile I would ask you some suggestions in recognizing noise components. As far as I understood I should base my selection on anatomic shape, and power spectrum.

I would see the Salimi-Khorshidi and Griffanti FIX papers on this - and also the fMRI artefact/cleanup literature.

Cheers

- At what frequency do generally heart beat and respiration related artifacts map?

- For commonly used tasks with stimuli lasting between 3 and 10 seconds, where should they map in the power spectrum? - --
- What do I need to know to estimate a task related signal frequency in the power spectrum?

- For let's say HR 60bpm,can I expect the the heart beat to map around 0.0116 Hz? So in the FSL report I should find it all on the left around a value of 16. But there I also find signals that look like task related.

- Are signals with a wide window of frequency most probably to consider noise?

Back to anatomic criteria, can I consider as rule of thumb bad components those having:
-stripe shape

-medulla intense signal (maybe swallow related or heart beat or respiration related?),
-brain edges locations, with positive signals on a side and negative on the other side?
-Orbitofrontal or low temporal strong activation?

Thank you very much for your time and any hint!

2014-05-01 3:00 GMT-04:00 Mark Jenkinson <[log in to unmask]>:

Hi,

This is certainly a *lot* of components, and you wouldn't normally get this many.
The number is affected by the number of timepoints in your data, but also the amount of structured signal/artefact.
It might be more helpful to set the dimension (i.e. the number of components) to a smaller value manually (turn the automatic dimensionality estimation off within the Stats tab of the Melodic GUI) and see what you see in the data.
If you have strong artefact then it is likely that these will appear clearly in a reduced set of components and that you could use these to denoise the data.

All the best,
Mark

On 30 Apr 2014, at 22:04, Marco Solmi <[log in to unmask]> wrote:

> Dear FSL experts,
>
> I am having some problems with fMRI data with behavioral task that do not show expected results.
> I am familiarizing with different preprocessing options; in a subject with a lot of motion I tryed feat with motion_outliers (often runs out of memory-you already clarified this point-thanks!), and ICA exploration with no model within the feat - prestats only(with more than 900 IC detected).
>
>
> The questions are:
> How can I deal with more than 900 IC (hand search does not seem feasable-maybe I can increase the false negative-false positive ratio)-how could I chose the IC instead of a manual search? I am rerunning the same with the design matrix specified, should I expect something different from 900 IC? should it be better to run it with the already MCFLIRT motion corrected image?
>
> Thanks!

--
Marco Solmi