[FieldTrip] Magnetometers and gradiometers for source localisation

[Sarang S. Dalal]

Hi Casper,

This is an important question, but unfortunately there is not yet consensus on the best approach to combine different sensor types (magnetometers and gradiometers, or MEG and EEG, for example). Indeed, the basic conundrum is that each sensor type records data of different scale, with different units, and different noise levels, so it's not immediately obvious how to incorporate them "fairly" into the beamformer.

Most researchers will indeed simply select either magnetometers and gradiometers, and to be honest, this is the easiest thing to do in most software packages, including FieldTrip. As to which is better -- yes, theoretically magnetometers ought to have sensitivity to deeper sources, but in practice, it'll likely depend on how much noisier the magnetometers is compared to the gradiometers. If you're lucky enough to have an especially clean recording environment, then magnetometers would indeed probably be the better choice... but at many sites, the magnetometers pick up a lot of noise, so for them, the gradiometers may be a better choice. You may simply need to try both ways and evaluate the resulting performance.

If you want to get into ways to actually combine the gradiometers and magnetometers, it would involve coding your own functions to create special covariance matrices. (As far as I know, there isn't a pre-made FieldTrip function that implements these techniques yet, but maybe there's a new or hidden function in there.) The two major approaches that I know of are prewhitening and zeroing out the "cross terms" in the covariance matrix (i.e., where a magnetometer channel would be multiplied with a gradiometer channel). This is experimental territory, so we don't yet know for sure how these approaches behave in the wild, but in my limited experience with limited datasets, they appear to do the right thing -- i.e., give a small improvement in output SNR over selecting one channel set alone, without messing things up completely. :-)

If you're interested in giving that a try despite the caveats, I and perhaps some others on this list could help you with code and advice. Just note that it's very much venturing into DIY territory and it'll be up to you to figure out whether it actually helps for your data or not. Especially if you're interested in deeper brain sources, you'll need to evaluate whether this approach is better or not for that purpose.

Hope that gives some insight!

Cheers
Sarang



On Fri, 2019-11-15 at 09:18 +0000, Casper Kerren wrote:
Hi,

I am currently source localising an effect using LCMV on Elekta MEG data. I have both gradiometers and magnetometers and have until recently treated these as the same when preparing the leadfield. However, after having had discussions with my colleagues about the different units and strength for grads and mags, we are hesitant as to what might be the best way of selecting channels. I have searched the fiedltrip mailing list, and have found some answers, but nothing that I can say is very convincing. Could someone point me in the right direction regarding this? Perhaps someone who has done some checks on using both grads and mags and using only grads or mags. We are interested in a deeper source, and naively thought that magnetometers would be necessary to include. However, is this just as I say: naïve?

In the ft_prepareleadfield function I can see that gradiometers is the only channels that are thought of being included, so does fieldtrip actually expect us to exclude magnetometers?

Kind regards,

Casper Kerrén, PhD student
School of Psychology
College of Life and Environmental Sciences
University of Birmingham
Edgbaston
Birmingham B15 2TT
e-mail: cxk699 at student.bham.ac.uk<mailto:cxk699 at student.bham.ac.uk>




_______________________________________________

fieldtrip mailing list

<https://mailman.science.ru.nl/mailman/listinfo/fieldtrip>

https://mailman.science.ru.nl/mailman/listinfo/fieldtrip


<https://doi.org/10.1371/journal.pcbi.1002202>

https://doi.org/10.1371/journal.pcbi.1002202

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20191115/269fbfe9/attachment.html>