I am toying with the idea of localizing sources of environmental noise from MEG data. The idea is to better understand those sources, and hopefully find some way of eliminating them, as opposed to just filtering them out post-hoc.
Is there some easy way to produce an all-space forward matrix (possibly with cruder spatial sampling for more distant locations)? This matrix would be characteristic of the system, applicable to any recording. Does this make sense?
Hi @alain
In my opinion, the all-space forward model is doable.
You can place sources everywhere in the space around the sensors.
You also need to estimate the conductivity and permeability values of the surrounding space, and then you can estimate the lead field on all these points from the sensors.
However, I'm not sure that your estimated lead field will be relevant for noise source localization, there are many factors that need to be taken into account.
What you suggest @alain is doable but only with a forward model of MEG/EEG that considers that sources are placed in an infinite homogenous medium. The models in Brainstorm are for (brain) sources inside a conducting medium (the head). The modeling of an infinite homogenous medium is actually simpler (e.g., plain Biot & Savart) but it is not featured in the software. Now, you can "hack" some of Brainstorm code for forward modeling to substitute the forward modeling deviations accordingly.