Combining different MEG sensortypes

[Lauri Parkkonen]

Hello Michael,

At least for MNE and beamforming, the most common approach is to use
both magnetometers and planar gradiometers together in a single (mixed)
forward solution; however, weighting must be applied because the units
and numerical scales of the two sensor types are different. This
weighting is typically done by estimating the noise covariance matrix
and then using the reciprocals of the diagonal elements for each
channel. If obtaining the full noise covariance matrix is too
troublesome for some reason, the obvious shortcut is to just compute the
baseline noise RMS, i.e., the diagonal of the matrix, which is what the
multi-dipole modelling software (by Neuromag) does.

There is no localization or amplitude bias if you source model
SSS'ed/MaxFiltered data directly whereas -- as you pointed out -- such
bias does exist for ICA or otherwise projected data. The important
difference is that SSS includes complete models for all possible signal
patterns originating from the volumes inside and outside the sensor
array, and these subspaces are linearly independent. On the contrary, a
component/signal vector determined from the data (by ICA or PCA, for
example) generally has a non-vanishing projection on the brain signal
subspace and without knowing that subspace, there is no way to "undo"
the distortion of that part of the signal but one has to carry the
information about the projection all the way to the lead field.

You certainly know the following but for those who may wonder: After
MaxFiltering, one may still need to take into account the reduced number
of degrees of freedom in the regularisation in MNE and beamforming. For
example, dropping the lower N of the eigenvalues may not have the
desired effect as some of the retained eigenvalues may already be zero
in MaxFiltered data (while they would be small but non-zero for
non-filtered data).

Best regards,
Lauri

22.11.2010 20:54, Michael Wibral kirjoitti:
> Dear Fieldtrip users (with a Neuromag system),
>
> I have a question on how to combine the Information from the planar gradiometers and the magnetometers of a 306 channel Neurmag system best for beamformer weight computation and source time course reconstruction. Do you compute a complete leadfield mixing both types of gradiometers (i.e. you do an unweighted analysis)? Do you somehow weight the sensors for their different noise levels? Do you compute two sets of timecourses (one from grads, one from megnetometers)?
> A related question: Do you update the leadfileds for projections that MAxfiltering does (like it should be done when using ICA)?
>
> I am asking because it has been shown that the more sensors are available the better the time course reconstruction (a recent paper by Matthew Brookes). Hence it would be a pity to have to throw some of the information away.
>
> Michael
>
> ---------------------------------------------------------------------------
> You are receiving this message because you are subscribed to
> the  FieldTrip list. The aim of this list is to facilitate the discussion
> between  users of the FieldTrip  toolbox, to share experiences
> and to discuss  new ideas for MEG and EEG analysis.
> See also http://listserv.surfnet.nl/archives/fieldtrip.html
> and http://www.ru.nl/neuroimaging/fieldtrip.
> ---------------------------------------------------------------------------

---------------------------------------------------------------------------
You are receiving this message because you are subscribed to
the  FieldTrip list. The aim of this list is to facilitate the discussion
between  users of the FieldTrip  toolbox, to share experiences
and to discuss  new ideas for MEG and EEG analysis.
See also http://listserv.surfnet.nl/archives/fieldtrip.html
and http://www.ru.nl/neuroimaging/fieldtrip.
---------------------------------------------------------------------------