Sources estimation - Sensors type

Mattia_Pagnotta · March 5, 2015, 10:52am

Good Morning all,

I’m working on MEG recordings acquired using a Neuromag Vector View 306 Channel MEG (Elekta AB).
There are three options for “sensors type”: MAG works on the 102 magnetometers, GRAD works on the 204 planar gradiometers, and ALL.
How does this third option work? How does it combine the different types of channels?

Thank you for your help and time.

Best regards,
Mattia

John_Mosher · March 5, 2015, 5:32pm

Hello Mattia,

I’m assuming you’re asking how we combine these two MEG arrays, when the units are apparently so different? The magnetometers in the Elekta instrument are stored in units of T (as are the other MEG machines), but the planar gradiometers are stored in units of T/m, which because of the short baseline, give initial raw GRAD numbers that appear roughly 100 times greater than their MAG counterparts.

A naive attempt to do simultaneous source localization with both MAGS and GRADS would simply ignore the MAGS, since they are so “tiny” that they don’t contribute to the model. However, in both the XFIT software that came with your Elekta machine, and in Brainstorm, and in many other software packages, we can explicitly take into account samples of the baseline (or noise) data. In Brainstorm, we ask you first to create a “noise covariance” matrix from the recordings. We “pre-whiten” the data by pre-scaling the channels by the observed std. deviations seen in the noise data (we can use either the diagonal values of the noise covariance matrix, or the full matrix, which includes rotations into virtual combinations before scaling). This brings GRADS and MAGS into the same basic range of units. The same whitener is applied to the head modeling, to keep units and scales consistent. Source estimation then proceeds normally with a combined array of 306 pre-whitened sensors and pre-whitened data.

Does this answer address your question?

– John

Alexandre · March 6, 2015, 1:19am

Hi,

Sorry for the plug but you might want to better understand this by looking at:

ncbi.nlm.nih.gov

Automated model selection in covariance estimation and spatial whitening of MEG and EEG signals.

DA Engemann and A Gramfort, NeuroImage, Mar 2015

Magnetoencephalography and electroencephalography (M/EEG) measure non-invasively the weak electromagnetic fields induced by post-synaptic neural currents. The estimation of the spatial covariance of the signals recorded on M/EEG sensors is a building block of modern data analysis pipelines. Such covariance estimates are used in brain-computer interfaces (BCI) systems, in nearly all source localization methods for spatial whitening as well as for data covariance estimation in beamformers. The rationale for such models is that the signals can be modeled by a zero mean Gaussian distribution. While maximizing the Gaussian likelihood seems natural, it leads to a covariance estimate known as empirical covariance (EC). It turns out that the EC is a poor estimate of the true covariance when the number of samples is small. To address this issue the estimation needs to be regularized. The most common approach downweights off-diagonal coefficients, while more advanced regularization methods are based on shrinkage techniques or generative models with low rank assumptions: probabilistic PCA (PPCA) and factor analysis (FA). Using cross-validation all of these models can be tuned and compared based on Gaussian likelihood computed on unseen data. We investigated these models on simulations, one electroencephalography (EEG) dataset as well as magnetoencephalography (MEG) datasets from the most common MEG systems. First, our results demonstrate that different models can be the best, depending on the number of samples, heterogeneity of sensor types and noise properties. Second, we show that the models tuned by cross-validation are superior to models with hand-selected regularization. Hence, we propose an automated solution to the often overlooked problem of covariance estimation of M/EEG signals. The relevance of the procedure is demonstrated here for spatial whitening and source localization of MEG signals.

Alex

John_Mosher · March 6, 2015, 7:49am

Nice paper, Alex, and thanks for the Acknowledgment in the paper! Indeed, a good paper for the more advanced user who wants to understand the complexities of pre-whitening. I was already working on putting the method of Ledoit and Wolf into our standard Brainstorm workflow.

For the archives, here’s a bit more detail:

Neuroimage. 2015 Mar;108:328-42. doi: 10.1016/j.neuroimage.2014.12.040. Epub 2014 Dec 23.
[B]Automated model selection in covariance estimation and spatial whitening of MEG and EEG signals.
[/B]Engemann DA, Gramfort A.

John

Mattia_Pagnotta · March 9, 2015, 10:29am

Dr Mosher thank you very much for your fast response and explication.
That’s what I was asking.
Dr Gramfort I read your paper on pre-whitening, it’s very interesting. It helped in understanding.

Thanks to both of you, your responses are very helpful.

Best regards,
Mattia

MarcoBuiatti · May 14, 2015, 9:35am

Dear John and all,

I have an old question related to what you mention here. I am also working with Neuromag MEG data, and my understanding is that magnetometers and gradiometers provide different, complementary information about the underlying brain activity depending on their different coil geometry.

The XFIT software manual that comes with the Neuromag Elekta system
(http://www.neurospin-wiki.org/pmwiki/Main/UserManuals?action=download&upname=XFit-5.5.pdf)
claims that XFIT does integrate the sensor-specific coil geometry information for source modeling. However, you are only mentioning a scaling procedure when combining the different sensors together.

My question: is the sensor-specific coil geometry information used in Brainstorm source reconstruction or not? And if not, aren’t we underexploiting the richness of Neuromag sensors?

Thanks,

Marco

[QUOTE=John Mosher;8026]Hello Mattia,

I’m assuming you’re asking how we combine these two MEG arrays, when the units are apparently so different? The magnetometers in the Elekta instrument are stored in units of T (as are the other MEG machines), but the planar gradiometers are stored in units of T/m, which because of the short baseline, give initial raw GRAD numbers that appear roughly 100 times greater than their MAG counterparts.

A naive attempt to do simultaneous source localization with both MAGS and GRADS would simply ignore the MAGS, since they are so “tiny” that they don’t contribute to the model. However, in both the XFIT software that came with your Elekta machine, and in Brainstorm, and in many other software packages, we can explicitly take into account samples of the baseline (or noise) data. In Brainstorm, we ask you first to create a “noise covariance” matrix from the recordings. We “pre-whiten” the data by pre-scaling the channels by the observed std. deviations seen in the noise data (we can use either the diagonal values of the noise covariance matrix, or the full matrix, which includes rotations into virtual combinations before scaling). This brings GRADS and MAGS into the same basic range of units. The same whitener is applied to the head modeling, to keep units and scales consistent. Source estimation then proceeds normally with a combined array of 306 pre-whitened sensors and pre-whitened data.

Does this answer address your question?

– John[/QUOTE]

Francois · May 14, 2015, 9:52am

Hi Marco,

Yes, the geometry of the sensor is correctly taken care of in Brainstorm and MNE. The sensor types we support are referenced in the file brainstorm3/toolbox/io/private/coil_def.dat.
To see how this translates in your Brainstorm database, right-click on the channel file > Edit channel file.
The Neuromag gradiometers/magnetometers are described with 4 points (columns Loc), each point is associated with a weight (column Weight) describing its contribution to the output of the sensor.

Cheers,
Francois

MarcoBuiatti · May 15, 2015, 12:23am

Thanks Francois, that’s reassuring.

All the best,

Marco

YangFei · February 4, 2021, 6:54am

Hello all,
I’m working on MEG recordings acquired using a Neuromag 306 Channel MEG.
There are three options for “sensors type”: MAG works on the 102 magnetometers, GRAD works on the 204 planar gradiometers, and ALL.

I have a question: what is the difference between these three options in the process of computing sources?
I think maybe use MAG+GRAD choise to compute sources can get the accurate result of sources. Do I understand correctly?

Thanks for your help.
Best regards,
Yang Fei

Francois · February 4, 2021, 11:00am

Indeed, with Elekta MEG, we do recommend using both types of sensors simultaneously for the source estimation:
https://neuroimage.usc.edu/brainstorm/Tutorials/TutMindNeuromag#Inverse_model
https://neuroimage.usc.edu/brainstorm/Tutorials/VisualSingle#Inverse_model:_Minimum_norm_estimates

The technical justification is discussed at the beginning this thread.

YangFei · February 4, 2021, 12:47pm

Thanks Francois