sLORETA with identity matrix as noise covariance

Dear all,

I am working with some resting-state EEG data and trying to find out what is the most sensible approach to compute sources. I was opting for sLORETA using (as suggested in the tutorial) identity matrix as noise covariance, obtaining very noisy signal at deeper sources.

As far as I understood, sLORETA standardizes the source data by the noise, which in this case would be the identity matrix. Is it then a sensible approach? Should I maybe compute the covariance matrix on long resting-state recordings (e.g. 40s) and choosing the diagonal at the source computation step in order to obtain a subject specific
noise covariance?

Many thanks for your help

Ramtin

@Sylvain @John_Mosher

1 Like

If your signal of interest is that of resting-state, then using resting-state data for noise covariance estimation would not be logical, in my opinion (even with just keeping the diagonal elements). Hence a non-informative identity matrix would be the safest choice.

Many thanks for your recommendation.

In fact, would this approach correspond to a (non standardised) LORETA?

You would still have an empirical model of data covariance, which LORETA does not use.

1 Like

I understand, many thanks for your help!