Dear Urs,

On Tue, May 17, 2011 at 10:50 AM, Urs Bachofner <[log in to unmask]> wrote:
> I have a question regarding the concept of EEG data Re-referencing in Spm8.
> The way I see it, re-referencing is a very important step prior to Source
> Analysis. To get useful data, the Signals should be re-referenced to
> average.
> According to the SPM8 Manual P. 110 we can use spm_eeg_montage to do this.
>
> The data I'd like to process is recorded with a 128-channel EEG-net with the
> reference electrode at CZ.
> According to the SPM8 Manual, for average reference the matrix should have
> (N-1)/N at the diagonal and -1/N elsewhere.
> In my case this results in 0.9921875 at the diagonal and -0.0078125
> elsewhere.
>

> Two questions:
> 1. These numbers are so close to the original matrix (1 at diagonal and 0
> elsewhere) which is said to not change anything. Is this matrix correct?
>


Yes, it's all OK. The more electrodes you have the closer the average
will be to zero but remember that the actual values also depend on the
data, not only on the weights so you should still re-reference.

> 2. Since electrodes that are closer to the reference (in this case CZ) have
> a higher amplitude recorded, shouldn't such electrodes be weighted
> differently than electrodes that are more distant to CZ? Am I missing the
> point here?
>
>

No, the average reference is just what it is, every channel minus the
average of all channels. It does not depend on where the original
reference is.

> And finally, besides Filtering, epoching, Artefact Detection (removing bad
> channels and bad trials), Re-referencing, and baseline correction are there
> other preprocessing steps that are necessary to get good data from my evoked
> potentials?

No, these are the basic steps. You might try using the robust
averaging option. Also there is an option in MEEGTools for correcting
eye blinks. You might or might not need it depending on your data.

> And is there a certain order in which these steps should be
> executed?
>

I've just presented a slide about that at the SPM course last week.
Basically it says the following:

----

Considerations for order of processing steps:

It’s better to filter before epoching unless only small part of the
data is relevant. As an alternative one could pad the epochs of
interest with extra data and discard it later.

Downsampling speeds up the other steps and reduces disk space usage,
but it involves low-pass filtering and...

Low-pass or band-pass filtering before high-pass can generate ringing
at the edges, which is especially problematic for epoched data. So
high-pass should come first, but...

Only some channel types (EEG, MEG, LFP etc.) are filtered. So channel
types should be set correctly first (Prepare, Montage).

---


Best,

Vladimir
> Thank you very much for you help.
>
>
>
>