[FieldTrip] Source localization of resting state EEG data using lcmv beamformer

[Stefan Dvoretskii]

Dear Hanna,
as to my best knowledge, you can assume the data is stationary in the wide
sense for *short time periods*, e.g. 2 second time frames.
Because of this, one legitimate way we do it in our lab is to cut long
resting-state data (e.g. 5 minutes rs experiment) in small overlapping
windows (e.g. 2 second time frames with 1 second overlap). These are then
your "quasi-trials" you can calculate the covariance matrix on.

I have not applied common filters much but we had this idea too. I also
believe it is explained in Jaiswal, Amit, et al. "Comparison of beamformer
implementations for MEG source localization." NeuroImage 216 (2020): 116797
in some detail. Somebody else on the list might be more helpful on this.

I hope it helps you somewhat. Don't hesitate to ask any further questions,
if I can be of assistance.

Best regards,
Stefan
Research Assistant
Painlab Munich

On Thu, 24 Feb 2022, 23:15 Honcamp, Hanna (PSYCHOLOGY) via fieldtrip, <
fieldtrip at science.ru.nl> wrote:

> Dear Fieldtrip community,
>
> ​​
>
> I am analyzing continuous EEG resting state (RS) data and I have a couple
> of questions about the beamformer source reconstruction methods and the
> best application of it in this context. Firstly, I have not seen many
> tutorials/papers concerned with RS source analysis using the lcmv
> beamforming method. Since I am interested in the analysis and properties of
> the reconstructed source time courses, the lcmv seemed a viable option.
> Further, in van Veen et al. (1997), it is described that the computation of
> the covariance matrix assumes that the data is “wide sense stationary”.
> This description seems to be contradictory to the non-stationarity of RS
> data.
>
>
> Q1: Is the lcmv beamformer the recommended source analysis method for
> resting state data for the purpose of extracting the source time courses?
>
>
> Q2: What is the best way to compute the covariance matrix in the context
> of RS data? Specifically, should I use the whole data or a subset of
> timepoints? What is the reasoning behind that?
>
>
> Lastly, I understood that the common filter approach is recommended for
> within-subject analysis, e.g., comparing conditions. However, it is not
> clear whether the common filter is also feasible to use in a multi-subject
> context.
>
>
> Q3: In order to compare the reconstructed source time courses of multiple
> subjects, do I need to construct a common filter and apply it to all
> subjects? If so, how does that affect the covariance computation, i.e.,
> should I use all subjects (e.g., in a concatenated format) for computation
> of the covariance matrix, or a subset of subjects?
>
>
> Many thanks in advance - any advice is much appreciated!
>
> Best,
>
> Hanna
>
>
>
> Best,
>
> Hanna
>
>
>
> *Hanna Honcamp*
> PhD Candidate | BAND Lab
> Faculty of Psychology and Neuroscience
> Dept. NP&PP | Maastricht University
> _______________________________________________
> fieldtrip mailing list
> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
> https://doi.org/10.1371/journal.pcbi.1002202
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20220225/e2156567/attachment.htm>