Problem with surfaces vs volumes

[rjjj8]

Hi, thanks to you both Hossein & Francois for the help

@hossein27en

First of all, try to compare the PSD of raw EEG, scouts' time series for the surface, and scouts' time series for the volume. If you apply the same approach for surface and volume source reconstructions, you should at least see a similar pattern for surface and volume.

Unfortunately the PSD for the scout time series is essentially the same as that for coherence, here are the figures:

@hossein27en

Here, you have computed 1 x N connectivity, in which you need to specify one scout as the source. Since you wanna compare surface with volume, your source is not the same in two cases, besides the thing that all sinks are also different. So, if you want to perform a comparison in this way, select the scouts as the source, in which they have a high overlap between surface and volume.

Unfortunately those were NxN coherence calculations![quote="hossein27en, post:5, topic:3261"]
At this time, I recommend using magnitude-squared coherence instead of imaginary coherence to you.
[/quote]

Tried magnitude squared and I don't know quite how to interpret it - but it still seems to have a 20Hz peak unfortunately

@hossein27en

Please note that coherence is a measure of pairwise correlation in the frequency domain. This measure is divided by the power of each variable. I do agree that if you see an alpha peak in all channels, you expect a high coherence between all pairwise channels, but don't forget the correlation and normalization.

I think this might be what Francois is referring to in that the normalization is having an effect?

@Francois

What the process will do is compute separately the connectivity measure for each orientation and each scout (3x3=9 values for each pair of scouts), then get the maximum measure for each pair (max of 9 values). This is arbitrary and unstable, but this is what is done for lack of a better solution at this time.

Here is the spectrum when I use that method:

So that's great that I have a spectrum that corresponds! It is returning 90x3 = 270 values - Is there a method in Brainstorm where I can get a coherence measure that is the maximum i.e. just 90 scouts? I tried to add the coherence results to the pipeline but I couldn't see how I could reduce it to just 90 scouts.

I would rather use Surfaces because they are clearly easier to interpret, but I was under the impression that they weren't valid for deep structures - is this right? If I can use surfaces, I would prefer it if it is OK.

A question is: how does this analysis compare to the CARTOOL software Volume analysis - is it the same?