EEG reference

Hello MNE users,

I noticed that in MNE inverse operations, all EEG data were forced by an
average referencing.
Does anyone has insight about how does the average reference influence the
inverse operations in MNE (I mean use dSPM or sLoreta)? Especially would
this average referencing influence functional connectivity measure in
source space (e.g. using phase locking)?

Thanks,
Mengting
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20170923/d04881a0/attachment-0001.html

Hi Mengting,

The average reference is the only valid scheme when performing ?source localisation?, no matter what you do afterwards. To first order, the localisation procedure is

- have measured (EEG) data
- build forward model that computes sensor-level readings for know sources
- compare measured and predicted (forward-projected) data to each other
- minimise prediction error under chosen prior/model

The predicted data are calculated relative to a hypothetical absolute reference potential of zero at infinity, whereas a real dataset could be referenced to a number of points on the head. To be able to compare the two datasets, both are re-referenced to their respective average: the average reference does not depend on the location of the on-line reference electrode (as long as it was functioning properly). After this re-scaling, the values can be compared directly.

Note that like any (proper) re-referencing procedure, taking the common average only shifts the zero-point; relative differences between electrode readings remain unaltered.

/Chris

Hi Chris,

Thanks for the info, this really helps in understanding the whole
procedures. So as I understand, reference is only a required step for
source localization, but not any other steps like measuring the
connectivity between brain regions. Sounds reference removed the DC
componets but remains the oscillation. If so, does it influence the power
in DC components? Also, does it matter the conventional source localization
methods such as dipole fitting?

Really appreciate for help,

Mengting

2017-09-25 4:55 GMT-04:00 Christopher Bailey <cjb at cfin.au.dk>:

Hi Mengting,

The average reference is the only valid scheme when performing ?source
localisation?, no matter what you do afterwards. To first order, the
localisation procedure is

- have measured (EEG) data
- build forward model that computes sensor-level readings for know sources
- compare measured and predicted (forward-projected) data to each other
- minimise prediction error under chosen prior/model

The predicted data are calculated relative to a hypothetical absolute
reference potential of zero at infinity, whereas a real dataset could be
referenced to a number of points on the head. To be able to compare the two
datasets, both are re-referenced to their respective average: the average
reference *does not depend on the location of the on-line reference
electrode* (as long as it was functioning properly). After this
re-scaling, the values can be compared directly.

Note that like any (proper) re-referencing procedure, taking the common
average only shifts the zero-point; relative differences between electrode
readings remain unaltered.

/Chris

Hello MNE users,

I noticed that in MNE inverse operations, all EEG data were forced by an
average referencing.
Does anyone has insight about how does the average reference influence the
inverse operations in MNE (I mean use dSPM or sLoreta)? Especially would
this average referencing influence functional connectivity measure in
source space (e.g. using phase locking)?

Thanks,
Mengting
_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis

The information in this e-mail is intended only for the person to whom it
is
addressed. If you believe this e-mail was sent to you in error and the
e-mail
contains patient information, please contact the Partners Compliance
HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in
error
but does not contain patient information, please contact the sender and
properly
dispose of the e-mail.

_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis

The information in this e-mail is intended only for the person to whom it
is
addressed. If you believe this e-mail was sent to you in error and the
e-mail
contains patient information, please contact the Partners Compliance
HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in
error
but does not contain patient information, please contact the sender and
properly
dispose of the e-mail.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20170927/1ef78c7c/attachment-0001.html

Hi Mengting,

So as I understand, reference is only a required step for source localization,

The _average_ reference is required for source localisation

but not any other steps like measuring the connectivity between brain regions.

I?m not sure I understand what you mean by ?brain regions? here. If you?re going to measure connectivity in source space, your need an (inverse) operator of some sort to project your electrode-data ?into the brain?.

If you want to do sensor-level connectivity calculations on EEG data, the effect of the reference might depend on the connectivity metric you use. For those time/frequency-domain measures I?m aware of (but have little first-hand experience with), it doesn?t matter which reference your data is in.

Sounds reference removed the DC componets but remains the oscillation. If so, does it influence the power in DC components?

Yes and no. Relative differences within and between channels remain the same in any (proper) reference. Note though that the time-domain is not involved in (re-)referencing, so possible ?DC components? still remain in the data.

Also, does it matter the conventional source localization methods such as dipole fitting?

All localisation methods require the average reference, otherwise the recorded and predicted (forward model) data cannot be compared.

Best,

Chris

Hi Mengting,

let me chime in here. Whenever you are doing EEG analysis, the choice of
reference is important. EEG measures voltage, which is the difference in
electric potential between two points. This means, there's technically
no such thing as "the voltage at electrode Cz", there is only "the
difference in electric potential between electrodes Cz and the one I
stuck on the nose" (or whatever reference electrode you used, could be CMS).

After the recording, it is easy to "re-reference" the data, which is
choosing a difference electrode as a reference. For example, to move
from the nose reference to channel Pz as reference, you can just do: Cz
= Cz - Pz and you have a voltage at the Cz channel with Pz as reference.

You can imagine your data often looks completely different, depending on
the reference you choose. If you choose Pz as a reference, the signal at
Pz will be 0 (Pz - Pz) and all channels surrounding Pz only have a very
small signal. If Pz contained oscillatory activity (maybe alpha
rhythms), you would see alpha power at every electrode *except*
surrounding Pz. For example you would see "negative" alpha at Fpz
(remember, we're looking at Fpz - Pz).

So maybe Pz is not a very useful reference. We would like to have some
"neutral" reference that doesn't contain signals of interest. This is
why nose and mastoids are popular locations for the reference. But
there's another option: choose the average of all electrodes as "virtual
reference". With this reference, at any time, mean(Cz, Pz, ...) = 0.
This is the reference assumed by forward models, which is why when doing
anything that involves the forward model (source localization,
connectivity analysis at the source level, etc.) your data needs to be
in this reference.

One thing you need to be aware of when re-reference to an average
reference is that the reference is, you know, the average of all
electrodes. If one of the electrodes is noisy (maybe it got loose during
the experiment), this ruins the reference signal, and since the
reference is subtracted from all sensors, it ruins the signal at *all*
sensors.

So, first use a sensors which you know for sure has a good signal as a
reference (preferably a "neutral" reference like the noise, mastoids or
earlobe), then inspect the data to spot all sensors that have problems.
Either remove them from the data or have MNE-Python reconstruct the
signal at the bad sensor by interpolating the neighbouring sensors.
Then, when the data is squeaky clean, move to the average reference.

regards,
Marijn.

Hey Marijn,

Many thanks for this very basic but really important message, I think it
makes me feel much better why select average reference.
Thanks,

Mengting

2017-09-28 3:54 GMT-04:00 Marijn van Vliet <w.m.vanvliet at gmail.com>:

Hi Mengting,

let me chime in here. Whenever you are doing EEG analysis, the choice of
reference is important. EEG measures voltage, which is the difference in
electric potential between two points. This means, there's technically no
such thing as "the voltage at electrode Cz", there is only "the difference
in electric potential between electrodes Cz and the one I stuck on the
nose" (or whatever reference electrode you used, could be CMS).

After the recording, it is easy to "re-reference" the data, which is
choosing a difference electrode as a reference. For example, to move from
the nose reference to channel Pz as reference, you can just do: Cz = Cz -
Pz and you have a voltage at the Cz channel with Pz as reference.

You can imagine your data often looks completely different, depending on
the reference you choose. If you choose Pz as a reference, the signal at Pz
will be 0 (Pz - Pz) and all channels surrounding Pz only have a very small
signal. If Pz contained oscillatory activity (maybe alpha rhythms), you
would see alpha power at every electrode *except* surrounding Pz. For
example you would see "negative" alpha at Fpz (remember, we're looking at
Fpz - Pz).

So maybe Pz is not a very useful reference. We would like to have some
"neutral" reference that doesn't contain signals of interest. This is why
nose and mastoids are popular locations for the reference. But there's
another option: choose the average of all electrodes as "virtual
reference". With this reference, at any time, mean(Cz, Pz, ...) = 0. This
is the reference assumed by forward models, which is why when doing
anything that involves the forward model (source localization, connectivity
analysis at the source level, etc.) your data needs to be in this reference.

One thing you need to be aware of when re-reference to an average
reference is that the reference is, you know, the average of all
electrodes. If one of the electrodes is noisy (maybe it got loose during
the experiment), this ruins the reference signal, and since the reference
is subtracted from all sensors, it ruins the signal at *all* sensors.

So, first use a sensors which you know for sure has a good signal as a
reference (preferably a "neutral" reference like the noise, mastoids or
earlobe), then inspect the data to spot all sensors that have problems.
Either remove them from the data or have MNE-Python reconstruct the signal
at the bad sensor by interpolating the neighbouring sensors. Then, when the
data is squeaky clean, move to the average reference.

regards,
Marijn.

Hi Mengting,

So as I understand, reference is only a required step for source
localization,

The _average_ reference is required for source localisation

but not any other steps like measuring the connectivity between brain
regions.

I?m not sure I understand what you mean by ?brain regions? here. If you?re
going to measure connectivity in source space, your need an (inverse)
operator of some sort to project your electrode-data ?into the brain?.

If you want to do sensor-level connectivity calculations on EEG data, the
effect of the reference might depend on the connectivity metric you use.
For those time/frequency-domain measures I?m aware of (but have little
first-hand experience with), it doesn?t matter which reference your data is
in.

Sounds reference removed the DC componets but remains the oscillation. If
so, does it influence the power in DC components?

Yes and no. Relative differences within and between channels remain the
same in any (proper) reference. Note though that the time-domain is not
involved in (re-)referencing, so possible ?DC components? still remain in
the data.

Also, does it matter the conventional source localization methods such as
dipole fitting?

All localisation methods require the average reference, otherwise the
recorded and predicted (forward model) data cannot be compared.

Best,

Chris

Hi Chris,

Thanks for the info, this really helps in understanding the whole
procedures. So as I understand, reference is only a required step for
source localization, but not any other steps like measuring the
connectivity between brain regions. Sounds reference removed the DC
componets but remains the oscillation. If so, does it influence the power
in DC components? Also, does it matter the conventional source localization
methods such as dipole fitting?

Really appreciate for help,

Mengting

2017-09-25 4:55 GMT-04:00 Christopher Bailey <cjb at cfin.au.dk>:

Hi Mengting,

The average reference is the only valid scheme when performing ?source
localisation?, no matter what you do afterwards. To first order, the
localisation procedure is

- have measured (EEG) data
- build forward model that computes sensor-level readings for know sources
- compare measured and predicted (forward-projected) data to each other
- minimise prediction error under chosen prior/model

The predicted data are calculated relative to a hypothetical absolute
reference potential of zero at infinity, whereas a real dataset could be
referenced to a number of points on the head. To be able to compare the two
datasets, both are re-referenced to their respective average: the average
reference *does not depend on the location of the on-line reference
electrode* (as long as it was functioning properly). After this
re-scaling, the values can be compared directly.

Note that like any (proper) re-referencing procedure, taking the common
average only shifts the zero-point; relative differences between electrode
readings remain unaltered.

/Chris

Hello MNE users,

I noticed that in MNE inverse operations, all EEG data were forced by an
average referencing.
Does anyone has insight about how does the average reference influence
the inverse operations in MNE (I mean use dSPM or sLoreta)? Especially
would this average referencing influence functional connectivity measure in
source space (e.g. using phase locking)?

Thanks,
Mengting
_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis

The information in this e-mail is intended only for the person to whom it
is
addressed. If you believe this e-mail was sent to you in error and the
e-mail
contains patient information, please contact the Partners Compliance
HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you
in error
but does not contain patient information, please contact the sender and
properly
dispose of the e-mail.

_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis

The information in this e-mail is intended only for the person to whom it
is
addressed. If you believe this e-mail was sent to you in error and the
e-mail
contains patient information, please contact the Partners Compliance
HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you
in error
but does not contain patient information, please contact the sender and
properly
dispose of the e-mail.

_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis

The information in this e-mail is intended only for the person to whom it
is
addressed. If you believe this e-mail was sent to you in error and the
e-mail
contains patient information, please contact the Partners Compliance
HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in
error
but does not contain patient information, please contact the sender and
properly
dispose of the e-mail.

_______________________________________________
Mne_analysis mailing listMne_analysis at nmr.mgh.harvard.eduhttps://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis

The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine athttp://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.

_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis

The information in this e-mail is intended only for the person to whom it
is
addressed. If you believe this e-mail was sent to you in error and the
e-mail
contains patient information, please contact the Partners Compliance
HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in
error
but does not contain patient information, please contact the sender and
properly
dispose of the e-mail.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20171001/4d63161e/attachment-0001.html