[Neuroimaging] Obtaining power spectrum for each component

[bthirion]

On 06/08/2019 19:55, Silvia Russo wrote:
> Hello!
>
> I am relatively new to fMRI analysis and trying to use open-source 
> python-based methods to do my work.
> I am supposed to run single-subject ICA analysis on 25 subjects, 
> identify 5 resting state networks and look at the averaged low 
> frequency across networks in each subject.
>
> I am using preprocessed images from the human connectome project and 
> running them through nilearn’s module DictLearn.
> I asked for 10 components (as the images are already preprocessed).
>
> I had three questions regarding this:
>
>  1. Is it correct to create a ‘for’ loop that runs DictLearn on each
>     of the 25 subjects and call this a single-subject analysis? If
>     not, what would be a good way to do single-subject ICA without
>     having to use FSL Melodic?
>
That sounds right.
>
>  1. How can I automatically identify the specific network each
>     component may represent? E.g. component 1 is 90% likely to
>     represent the default mode network, component 2 is 92% likely to
>     represent the salience network, etc
>
I'm not aware of such an automated labelling in the Python ecosystem. 
You probably want to do this manually.

>  1. FSL automatically creates power spectrums for each component, how
>     can you do the same in python?
>
You would have to create a small utility for that using scipy, e.g.

https://docs.scipy.org/doc/scipy-0.19.0/reference/generated/scipy.signal.periodogram.html

Best,

Bertrand Thirion

PS: I would advise to ask such question through the Neurostars interface.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/neuroimaging/attachments/20190806/6154d882/attachment.html>