Variations on how to estimate sample noise covariance

Discussions
, ,
1

Hello there Brainstorm community,

I have several questions regarding forward and inverse modeling below:

In the tutorial section Variations on how to estimate sample noise covariance, there is a shorter sub-section explaining how to estimate noise covariance when the target brain activity is resting-state, shown below.

If the target brain activity to your experiment is resting, resting segments cannot be used to obtain noise statistics. For MEG, use empty-room measurements as explained above. For EEG , there are possible options: use sensor variance estimates, or avoid estimating empirical noise statistics.
Option #1 : Calculate noise covariance over a long segment of resting recordings, but save only the diagonal elements, i.e. the variance measured at each sensor. This option is available in Brainstorm's advanced options of source computation: select the option "Diagonal noise covariance".
Option #2 : Select "No noise modeling" in the contextual menu. This option uses the identity matrix as noise covariance matrix, and therefore assumes equal, unit variance of noise on every sensor. In inverse modeling, this is equivalent to assuming that noise is homoskedastic, and equivalent on all sensors. With this latter option, if data quality is not even on all electrodes, a higher noise level on some sensors may be explained with stronger, spurious source activity.

I understand that in this situation you cannot use resting segments to obtain noise statistics and therefore two options were given; #1) diagonal noise covariance or #2) "no noise modeling".

While I understand option #2) "no noise modeling", I am not sure about option #1), selecting "diagonal noise covariance".
To get to be able to select "diagonal noise covariance" you first need to right-click the recording and select compute sources (2018). However, without noise covariance, this is not possible. Therefore you first need to generate "no noise modeling" anyway regardless. Only then can you get to the stage of compute sources (2018) and then select "diagonal noise covariance".

Therefore in the case of resting-state would you first generate "no noise modeling" to compute sources and then further select "diagonal noise covariance"?
Not sure if it is correct to say option #1) diagonal noise covariance OR #2) "no noise modeling" when you need to generate "no noise modeling" regardless.

2

You should not use the covariance between channels computed in resting state, but you can use the channel variances (diagonal) computed from resting state.

Not quite, let me clarify:

Option#1: Compute noise covariance matrix from resting-state recordings (right-click on recordings Noise covariance > Compute from recordings option), then when computing sources select diagonal noise covariance

Option#2: Set the identity matrix as noise covariance matrix (right-click on recordings Noise covariance > No noise modeling (identity matrix) option), then when computing sources select diagonal noise covariance (although it's already a diagonal matrix).

3

Thanks for the reply Raymundo,
If both methods involve diagonal noise covariance and they differ by either compute from recordings OR no noise modeling what would be the pros and cons of each method?

4

From the covariance tutorial:

In inverse modeling, this is equivalent to assuming that noise is homoskedastic, and equivalent on all sensors. With this latter option, if data quality is not even on all electrodes, a higher noise level on some sensors may be explained with stronger, spurious source activity.

As such computing from recordings and using only the diagonal takes into account the differences in noise level in the sensors. These are some post that you may find interesting: