Reverse Polarity of SSP decomposition

In one of the trials; the SSP1 is inverted for VEOG signals.

Can someone help me in fixing it ?

The sign of SSP components is arbitrary. Note that SSPs comprise a time series (shown in black here) and a basis vector (a form of topography of how much of each SSO is contained across sensors). The product of these two matches the original data, including their original polarity. In short, there's nothing to fix here :wink:

Oh that's interesting! I wasn't sure on how SSP components were going to be subtracted from the time series. I am interested in knowing the maths - Do you have any reference?

Also as an extension of blink artifacts.
In few trials, SSP decompositions' weren't good at all. I resorted to ICA decomposition. I understand that
the total number of ICA components = total number of channel(recording points) ;
each ICA component attempts to represent the behaviour of a source which contributes to the recording trace (here VEOG).
So, ICA1 is activity of a source somewhere near the eyes.

Is my understanding right?

Also, Is there a way in figuring out how much of ICA1 contributes to VEOG? This would help me gauge whether to remove just ICA1, or ICA1+ICA2 ?

SSPs is like a principal component analysis of signal contents around certain events of interest (such as eye blinks). It then projects the data away from the components that capture the artefact the most, as selected by the user.

ICAs are also mathematical entities which are not derived from physiological modeling. It is you as the user who decides which components contain signal elements that capture artifacts the most.

There is plenty of literature on ICA, PCA and EEG and I believe Brainstorm's documentation cites some of these references.

Some relevant links:
In Brainstorm documentation:

In the MNE manual:

1 Like

Thank you Raymundo!