Accuracy of EEG source localization

[mansoureh]

Dear Brainstorm,

I have been looking into a few comments regarding the accuracy of the EEG source localization:

"EEG does not have the spatial resolution we can expect from MEG, the information we collect at the surface of the head is a lot more smeared, so there is no hope to localize precisely the activity within a sulcus."

Source Model Dipole Orientation For EEG: constraint or unconstrained?

Dear Hasan, The positive/negative values are well explained in the Source estimation tutorial: http://neuroimage.usc.edu/brainstorm/Tutorials/SourceEstimation#Sign_of_constrained_maps People in EEG source modeling tend to use unconstrained models, as illustrated in the EEG/Epilespy tutorial, because it produces smoother maps. EEG does not have the spatial resolution we can expect from MEG, the information we collect at the surface of the head is a lot more smeared, so there is no hope to loca…

"the spatial resolution of MEG/EEG source imaging is very low (centimeter) compared with fMRI (millimeter)"

Beamforming Activation

fMRI voxels are usually a few mm3 large but the actual spatial resolution of the technique is certainly less, due to data smoothing for stats, group registration across subjects etc…

My question is: Considering the spatial resolution is so low, what is the point having dipoles with size as small as 0.35cm^2 when using the Brainstorm recommended number of dipoles? If we cannot accurately localize the activity within a sulcus, can we rely on results showing statistical difference in a small region inside a sulcus between two conditions?

Could we maybe assume that dipoles with higher activity have been localized more accurately, and then threshold for more reliable results?

On a similar note, in the following paper, it is mentioned that there needs to be an estimate of at least 6cm^2 to have detectable EEG

Cooper, Ray, et al. "Comparison of subcortical, cortical and scalp activity using chronically indwelling electrodes in man." Electroencephalography and clinical neurophysiology 18.3 (1965): 217-228.

but again the dipoles in Brainstorm show much smaller regions of activity than that. I realize that is also probably because the visualization window thresholds activity, but I've also done a simulation where I assign activity to only one dipole and the scalp plots still show EEG activity. This has me confused, is there a certain limit of current/cm^2 below wich activity will not be detectable in EEG anymore? I realize this might be a different question, but I thought it might still be related to source localization accuracy, that's why I combined them into one post. thank you very much for your help

Best wishes,
Mansoureh

[Sylvain]

Hello Mansoureh.

If you wish to consider source orientations as cortically constrained, as by default in brainstorm, then a proper sampling of these latter across the cortex needs to be performed. This parameter (orientation) is actually more critical than location, especially in MEG. There is the possibility to relax this constraint and use head models with free-orientation dipoles at each source location with brainstorm. This is a fine approach as well for most application, where local spatial resolution is of no primary interest to the study. Then you can build much coarser source models (says 6000 dipoles, maybe less); the free orientation parameter is then expected to capture the different local contributions.

Concerning your simulations, simulations with any dipole strength will produce a signal on sensors. What is crucial is to set the signal-to-noise ratio, to define detectability of sources. It is believed that a 10 nA.m source strength is typical and detectable in MEG. See background reading of key review papers suggested on brainstorm’s website.

Hope this helps,

Sylvain.