[FieldTrip] Prewhitening on Elekta MEG data

[Schoffelen, J.M. (Jan Mathijs)]

Dear Sanjeev,

I am following the new tutorial for source reconstruction from Elekta MEG data available on the fieldtrip website, (http://www.fieldtriptoolbox.org/workshop/paris2019/handson_sourceanalysis/).

Nice, then making the tutorials was not a one-off time investment. Go for it!

This new pipeline uses a new function called ft_denoise_prewhiten which eventually makes possible to use all 306 MEG channels in source reconstruction. While replicating this tutorial I am having few issues.

1. The tutorial uses a function named datainfo_subjects, is this a new function in fieldtrip or any custom made matlab function just for this tutorial? I searched the newest version of fieldtrip but couldn't find it.

Indeed datainfo_subjects was written specifically with the tutorial dataset in mind. As such it should be taken with a grain of salt, and it may serve a didactical purpose as to how subject-specific information as to the location of the datafiles etc. can be organized in an efficient (albeit somewhat over-engineered) way. I believe that you can download some of the tutorial data, as well as some additional code through our good old ftp-server: ftp://ftp.fieldtriptoolbox.org/pub/fieldtrip/workshop/paris2019/


2. Can we use the all the 306 channels (prewhitened data) for computing time frequency analysis at the sensor level as well?

Yes, in principle that’s possible. Yet, as a caveat, I wouldn’t know how to do (spatial) cluster-based inference on those data, since it’s not straightforward to define ’neighbours’ for MEG sensors with different pick up coil geometries.

3.Does this prewhitening somehow AFFECT the source reconstruction using DICS methods and further connectivity methods.

Yes, it does, but in theory only for the better of it. That is, the prewhitening operator places the different sensor types at a more equal footing, which allows a meaningful combination of the gradiometers and magnetometers in the inverse modelling step. Provided this prewhitener is also applied to the forward model. This is something that FieldTrip does for you automatically, provided you compute the leadfield (ft_prepare_leadfield) using a sensor description (grad-structure) that is obtained from data that has passed through ft_denoise_prewhiten.
Next to this, from a theoretical point-of-view, the beamformer algorithm is probably ‘working better’ with appropriately spatially whitened data, since the assumptions of the noise space being spatially white are more likely to be somewhat true, as compared to when the data is not whitened.

Good luck and keep up the good work,

Jan-Mathijs





I would really appreciate any support.

Thanks


Best regards
Sanjeev

Sanjeev Nara
Predoctoral Researcher BCBL
www.bcbl.eu<http://www.bcbl.eu>
https://sites.google.com/view/sanjeev-nara/
Tel: +34 943 309 300 (ext 314)
Legal disclaimer/Aviso legal/Lege-oharra: www.bcbl.eu/legal-disclaimer
_______________________________________________
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/20200914/a458334c/attachment.htm>