Auditory ROI

Mailing List Archive (read-only)
1

External Email - Use Caution

Hello all,
I work on auditory localizer data (when subjects listened to single stereo
tones), the goal is to define ROI.
For doing this, I found the top 0.5% vertixes, and used grow_labels
<http://martinos.org/mne/stable/generated/mne.grow_labels.html#mne.grow_labels>
   function.
I got nice maps, but when I made a plot of the time course on the left and
right ROI I got clear auditory ERP on the left hemisphere, and noisy and
flat signal on the right (figures attached). Those results repeat on
themselves in some other versions, and for some other similar data sets.
Does someone know this phanenomen? any idea?
Noa Guttman

AL1_al_r_roi.png
?
(55K)
<https://mail.google.com/mail/u/0?ui=2&ik=d7ffcb2e0d&attid=0.2&permmsgid=msg-a:r3861790589206292876&view=att&disp=safe&realattid=f_jpd092fc1>
AL1_al_l_roi2.png
?
(35K)
<https://mail.google.com/mail/u/0?ui=2&ik=d7ffcb2e0d&attid=0.1&permmsgid=msg-a:r3861790589206292876&view=att&disp=safe&realattid=f_jpd092ei0>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20181206/f731fde9/attachment-0001.html
-------------- next part --------------
A non-text attachment was scrubbed...
Name: AL1_al_l_roi2.png
Type: image/png
Size: 35232 bytes
Desc: not available
Url : http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20181206/f731fde9/attachment-0002.png
-------------- next part --------------
A non-text attachment was scrubbed...
Name: AL1_al_r_roi.png
Type: image/png
Size: 55474 bytes
Desc: not available
Url : http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20181206/f731fde9/attachment-0003.png

2

External Email - Use Caution

Dear Noa,

can you replicate this with the MNE-sample data that has auditory condition?

how did you get the label time course? with

https://martinos.org/mne/stable/generated/mne.extract_label_time_course.html

?

If so with what function?

Alex
?On Thu, Dec 6, 2018 at 8:48 PM ??? ??? <noa1hc at gmail.com> wrote:?

3

External Email - Use Caution

Hi, Noa and Alexandre:

This is cancellation phenomenon in signal processing. To be more specific
in neuroimaging of MEG/EEG, when you send the same sound simultaneously to
each ear of a subject, like you mentioned single stereo tones, there will
be temporally correlated sources on bilateral A1s in the MEG or EEG signals
recorded from the subject. Highly temporally correlated sources in an MEG
or an EEG will cause the problem you mentioned. Furthermore, this
phenomenon can be explained by math. So, if you want to know more, I can
give you some references.

Solution to this problem: Noa, you need to use advanced source localization
algorithms to localize the true source grids and reconstruct the
distortionless time-courses of correlated sources. But I am not familiar
with MNE, so I don't know which advanced algorithms are available in MNE
software to address this problem. However, I know some available algorithms
in other toolboxes.

Sincerely,
???
Bertram Liu

Alexandre Gramfort <alexandre.gramfort at inria.fr> ? 2018?12?8? ?? ??10:28???

        External Email - Use Caution

Dear Noa,

can you replicate this with the MNE-sample data that has auditory
condition?

how did you get the label time course? with

mne.extract_label_time_course — MNE 1.6.0 documentation

?

If so with what function?

Alex
?On Thu, Dec 6, 2018 at 8:48 PM ??? ??? <noa1hc at gmail.com> wrote:?
>
> External Email - Use Caution
>
> Hello all,
> I work on auditory localizer data (when subjects listened to single
stereo tones), the goal is to define ROI.
> For doing this, I found the top 0.5% vertixes, and used grow_labels
function.
> I got nice maps, but when I made a plot of the time course on the left
and right ROI I got clear auditory ERP on the left hemisphere, and noisy
and flat signal on the right (figures attached). Those results repeat on
themselves in some other versions, and for some other similar data sets.
> Does someone know this phanenomen? any idea?
> Noa Guttman
>
> AL1_al_r_roi.png
> ?
> (55K)
> AL1_al_l_roi2.png
> ?
> (35K)
> _______________________________________________
> Mne_analysis mailing list
> Mne_analysis at nmr.mgh.harvard.edu
> Mne_analysis Info Page

_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
Mne_analysis Info Page

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20181209/5cca670f/attachment.html

4

External Email - Use Caution

hi,

This is cancellation phenomenon in signal processing. To be more specific in neuroimaging of MEG/EEG, when you send the same sound simultaneously to each ear of a subject, like you mentioned single stereo tones, there will be temporally correlated sources on bilateral A1s in the MEG or EEG signals recorded from the subject. Highly temporally correlated sources in an MEG or an EEG will cause the problem you mentioned. Furthermore, this phenomenon can be explained by math. So, if you want to know more, I can give you some references.

I think that Noa is using MNE-dSPM so I am not sure it is a problem of
correlated sources.

Solution to this problem: Noa, you need to use advanced source localization algorithms to localize the true source grids and reconstruct the distortionless time-courses of correlated sources. But I am not familiar with MNE, so I don't know which advanced algorithms are available in MNE software to address this problem. However, I know some available algorithms in other toolboxes.

see what MNE has in store:
http://martinos.org/mne/stable/auto_examples/index.html

you can use MNE/dSPM/sLORETA/eLORETA with make_inverse_operator
you have most beamformers (make_lcmv, make_dics) as well as
sparsity based solvers that output dipole sets such as mixed_norm,
tf_mixed_norm, music, gamma_map etc.

Alex

5

External Email - Use Caution

Hi,

Thanks for your reply, Alex.
Yeah, you might be right.
Noa said "single *stereo* tones", so it may be the correlated sources
problem.

correlated sources in M/EEG. But the original LCMV beamformer will not be
able to localize and reconstruct the correlated sources because of its
unit-gain constraint and the cancellation phenomenon. Moreover, recently,
some scholars are tackling this problem, so if some people need very
accurate source localization, I would recommend newly advanced algorithms
to them.

Bert

Alexandre Gramfort <alexandre.gramfort at inria.fr> ? 2018?12?9? ?? ??10:27???

        External Email - Use Caution

hi,

> This is cancellation phenomenon in signal processing. To be more
specific in neuroimaging of MEG/EEG, when you send the same sound
simultaneously to each ear of a subject, like you mentioned single stereo
tones, there will be temporally correlated sources on bilateral A1s in the
MEG or EEG signals recorded from the subject. Highly temporally correlated
sources in an MEG or an EEG will cause the problem you mentioned.
Furthermore, this phenomenon can be explained by math. So, if you want to
know more, I can give you some references.

I think that Noa is using MNE-dSPM so I am not sure it is a problem of
correlated sources.

> Solution to this problem: Noa, you need to use advanced source
localization algorithms to localize the true source grids and reconstruct
the distortionless time-courses of correlated sources. But I am not
familiar with MNE, so I don't know which advanced algorithms are available
in MNE software to address this problem. However, I know some available
algorithms in other toolboxes.

see what MNE has in store:
Examples Gallery — MNE 1.6.0 documentation

you can use MNE/dSPM/sLORETA/eLORETA with make_inverse_operator
you have most beamformers (make_lcmv, make_dics) as well as
sparsity based solvers that output dipole sets such as mixed_norm,
tf_mixed_norm, music, gamma_map etc.

Alex

_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
Mne_analysis Info Page

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20181209/27a159dd/attachment-0001.html

6

Hi,

MNE or dSPM should not have a problem with correlated sources like LCMV. However, if the auditory responses from a previous tone extend to the period which is used in the noise covariance calculation interesting things may happen. What happens if you use an empty room noise covariance. What do the MNE signals (not dSPM) look like?

- Matti

7

External Email - Use Caution

Also sometimes in with bilateral auditory ROIs you can have the head position affect the SNR asymmetrically.. For example if a subject with a smaller head resting on the helmet such that their left is touching and have a larger gap on the right, you'd something like what you are seeing. Measuring and correcting for head position can help get the right localization but won't help recover the lost SNR in such cases.

Best,

Hari

?