EEG Spatial smoothing comparison

Hello there Brainstorm team,
The following question is regarding the processes "spatial_smoothing" and "spatial_smoothing [2024]".
I share screenshots of applying the "old" spatial_smoothing process (bottom left, C) versus the [2024] spatial_smoothing process (bottom right, D) to zscore_abs transformed sLORETA source map (top right, B) of cortical auditory evoked potential (top left, A). In both types of spatial smoothing processes, the below options were used.

My question is twofold, one of clarification and one of pros and cons.

  1. I understood spatial smoothing to smooth out focal activity often seen in constrained source maps (like this example). Therefore when I applied the two spatial smoothing processes I had expected the resulting source maps to have smoother and cleaner source map activations compared to panel B (i.e. lower patch-like brain activity). However, it seems like what happened in practice is that there is an increased spread of activity to other cortical parts (shown by red circles/ovals). And that the general strength of activation (in zscore) is less in panels C and D. Is this what you might expect to see from such processes?
  2. What are the pros and cons of the two different spatial smoothing methods? I am assuming that the "recommended" process to be used would be the [2024] version as it is more recent?


I'll try to answer as best I can and I'll let Raymondo complete and correct me :slight_smile:
As a preamble, I would say that Spatial smoothing with the option fixed FWHM and spatial smoothing 2024 with geodesic should lead to similar results. Using the option 'before 2023' would lead to a smoother map than the two others at similar FWHM inputs.

It seems from the input that you used different FHWM; can you confirm that you used the same for the screenshot (either 3mm or 10mm?)

For question 1, I am not very familiar with EEG source localization so I'll let Raymondo comment, but we expect the activity to spread spatially as the smoothing is applied.

You can check Surface Smoothing by Edouard2laire · Pull Request #645 · brainstorm-tools/brainstorm3 · GitHub to know more about the change in smoothing. here is a picture comparing the different options for smoothing:

The main difference between smoothing > FWHM fix, and smoothing [2024] > geodesic is that in the first one, we assume that the length between two adjacent nodes is constant however, because of the geometry of the brain, some brain regions will have more nodes and a smaller distance between each nodes. This means that the smoothing applied across the brain is not uniform. You might be applying more smoothing in some regions of the brain.

The new version alleviates this by actually using the actual distance between each vertex for the smoothing. Calculating this distance can, however, sometimes, be computationally intensive and in practice doesn't change the results too much when the density of the original mesh is close to being uniform.

You can see here the distribution of the distance between vertexes showing that they are quite all grouped around the median used for the computation:

In the end, this is mainly important when including subcortical structure:



@edelaire, thanks for the detailed explanation!

@mcp0228, in addition to the explanation above, I would recommend you to compare the plots panelB and paneD using no tresholding (or a lower threshold), to appreciate the effects of smoothing.

For the second question, the methods in the Smoothing 2024 are preferred.

@edelaire thank you kindly for that detailed answer! Much appreciated :smile:.
Regarding your question below, I present another screenshot below, this time making sure that the FWHM is 3 mm for both panels B and C (same time point as the above screenshot 89.8 ms). And thank you @Raymundo.Cassani for your suggestion I've also lowered the threshold to better appreciate the effects of smoothing.

In the above screenshot I just used the default settings which resulted in the spatial smoothing (old) to have 3 mm and the [2024] version to have 10 mm FWHM. Panels B and C clearly do look different while panel C actually looks quite similar with panel A (which has no spatial smoothing applied). Therefore, I have concluded that panel B looks more like if it was applied with spatial smoothing. I then compared these source maps to panel D which is spatial_smoothing [2024] with the default setting of 10 mm FWHM (geodesic). This to me looks like if it was spatial smoothed the most. Thus I am not convinced that the option of fixed FWHM and spatial smoothing 2024 geodesic should lead to similar results (for this specific condition/participant). But regardless, it does highlight that if you just used the default settings, spatial smoothing [2024] geodesic would produce the "best" results in terms of smoothing source activations.