Choosing scout function, before or after


I first computed source distribution and then normalized the source file to z-score and then compute connectivity test between ROIs.

I get two mixed results: while choosing between [B]When to apply the scout function a. before b. after[/B]

Going with option a i.e. before, all the diagonal values of the adjacency matrix are zero.

However, with option b i.e. after, some of the diagonal values (several ROIs at last) are non-zero.

What is the problem and could you please explain the significance of choosing between applying scout function before and after.

Thank you.


In the computation of the NxN connectivity matrices, the diagonal values are set to zero, because we are not interested by the connectivity measure between a signal and itself.
If we keep these high values, it may distort the display by forcing the scaling of the colorbars to a range of values that is not representative of the values of interest.
Let’s consider N scouts that all include M dipoles.

[B]Apply scout function before[/B]: First it computes the average of the sources time series by scout (N scouts), then computes the connectivity matrix (NxN).
This connectivity matrix contains zero values on the diagonal, and is saved directly as the output file.

[B]Apply scout function after[/B]: First it estimates the connectivity between all the possible sources in all the scouts (NM sources, [NMxNM] connectivity matrix), this matrix contains zeros on the diagonal. Then it applies the scout function to the connectivity values by pair of scouts: computes NN times the average of MM values. Each average of these MM values, corresponding to the connectivity value for a pair of scouts, has very little chance to be zero (even if there are zeros on the diagonal). The final connectivity measure saved to the output file for a scout with itself is usually not zero. This is why you don’t observe zeros on the diagonal in this case.

Does it make sense?

Hello Francois,

A heads up for a possible bug in the orthogonlised AEC matrix calculation.

On export of individual-subject connectivity data to Matlab workspace I get correct values along the diagonal for Coherence ( exactly “1”), non-orthoginalized AEC ( exactly “1”) and PTE (exactly ‘0.5’) using the ‘before’ scout function.

However for the same process when running the orthogonalized AEC (amplitude envelope correlation) I get quite large (but none close to 1 or 0) values along the diagonal, which makes me wonder if the orthogonalized AEC is calculated correctly (the matrix is symmetric nonetheless). This is despite the fact that the diagonal is correctly displayed (zero values/blocked out) on the matrix visualisation through the Brainstorm GUI.

Incidentally, any work planned in the near future on improving network thresholding/statistics ?

Many thanks!