We have a question regarding the computation of covariance matrices with
EEG and MEG in MNE:
The inverse is always computed on the average reference data (for EEG).
Therefore the covariance matrices should be computed on the average
reference too. Although the inverse solution, theoretically, does not
depend on the choice of reference electrode, one has to make sure that
the leadfield, covariance matrix, and data are referenced in the same
way when performing the computations in MNE. Apparently, MNE always uses
average reference for the leadfield and the data, but not necessarily
for the covariance matrices.
This means that we have to use the option --projon when we are computing
the covariance matrices with mne_process_raw.
If we use the --projoff option, then the covariance matrices are
computed with the acquisition reference, and therefore, the inverse
solution will be incorrect.
We have a question regarding the computation of covariance matrices
with EEG and MEG in MNE:
The inverse is always computed on the average reference data (for
EEG).
Therefore the covariance matrices should be computed on the average
reference too. Although the inverse solution, theoretically, does
not depend on the choice of reference electrode, one has to make
sure that the leadfield, covariance matrix, and data are referenced
in the same way when performing the computations in MNE. Apparently,
MNE always uses average reference for the leadfield and the data,
but not necessarily for the covariance matrices.
This means that we have to use the option --projon when we are
computing the covariance matrices with mne_process_raw.
If we use the --projoff option, then the covariance matrices are
computed with the acquisition reference, and therefore, the inverse
solution will be incorrect.
Are we right?
You are bringing up an important point. However, this is automatically
accounted for in MNE.
The average reference is equivalent to a projection operator. When the
inverse operator is put together, this projection is automatically
applied to the noise covariance matrix as well. Therefore, it is OK to
leave the projection off when the noise covariance matrix is computed.
Incidentally, I have found that sometimes the (for reasons I do not
exactly know) the EEG noise covariance is not as "good" when you look
at its eigenvalue spectra as the MEG one. Therefore, especially when
computing combined MEG/EEG estimates I usually regularize the noise
covariance using --megreg 0.1 and --eegreg 0.1 options to
mne_do_inverse_operator.
I'm currently using the function mne_make_combined_event_file from the mne
matlab toolbox. This function gets the time of an event on a specified
channel and put it in a text file. But I've noticed that it doesn't get the
onset of the event but the offset of this event (the time when the channel
goes back to zero).
Why is that ? Is there any reason to look at the offset instead of the onset
?
Thanks in advance,
Lucie
2009/4/16 Matti Hamalainen <msh at nmr.mgh.harvard.edu>
Hi Elisabeth,
Dear all,
We have a question regarding the computation of covariance matrices with
EEG and MEG in MNE:
The inverse is always computed on the average reference data (for EEG).
Therefore the covariance matrices should be computed on the average
reference too. Although the inverse solution, theoretically, does not depend
on the choice of reference electrode, one has to make sure that the
leadfield, covariance matrix, and data are referenced in the same way when
performing the computations in MNE. Apparently, MNE always uses average
reference for the leadfield and the data, but not necessarily for the
covariance matrices.
This means that we have to use the option --projon when we are computing
the covariance matrices with mne_process_raw.
If we use the --projoff option, then the covariance matrices are computed
with the acquisition reference, and therefore, the inverse solution will be
incorrect.
Are we right?
You are bringing up an important point. However, this is automatically
accounted for in MNE.
The average reference is equivalent to a projection operator. When the
inverse operator is put together, this projection is automatically applied
to the noise covariance matrix as well. Therefore, it is OK to leave the
projection off when the noise covariance matrix is computed.
Incidentally, I have found that sometimes the (for reasons I do not exactly
know) the EEG noise covariance is not as "good" when you look at its
eigenvalue spectra as the MEG one. Therefore, especially when computing
combined MEG/EEG estimates I usually regularize the noise covariance using
--megreg 0.1 and --eegreg 0.1 options to mne_do_inverse_operator.
I hope this helps,
Matti
-------------
Matti Hamalainen, Ph.D.
Athinoula A. Martinos Center for Biomedical Imaging
Massachusetts General Hospital
Building 149, 13th Street, Mailcode 149-2301
Charlestown, MA 02129-2060
USA
e-mail msh at nmr.mgh.harvard.edu
Tel +1 617 726 0323
FAX +1 617 726 7422
_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu Mne_analysis Info Page