[FieldTrip] combining magnetometers and planad gradiometers for analysis

Mon Jun 13 16:31:35 CEST 2011

Dear Michael, 
Thank you for your explanations.
As I understand, there is no routine way of handling combined gradiometers and magnetometers in Fieldtrip. I feel that the easiest way will be to average, as you have suggested, or to take a subset of the sensors (GRA or MAG).

Elena

________________________________________
From: fieldtrip-bounces at donders.ru.nl [fieldtrip-bounces at donders.ru.nl] on behalf of Michael Wibral [michael.wibral at web.de]
Sent: Monday, June 13, 2011 2:20 PM
To: Email discussion list for the FieldTrip project
Subject: Re: [FieldTrip] combining magnetometers and planad gradiometers for analysis

Dear Elena,

we do not use neuromag data. So I cannot say anything wrt to neuromag data. On theoretical grounds, the problem is that both sensor types have highly varying SNR from brain location to brain location. While SNR gets worse with depth for both sensor types, this happens more rapidly with the gradiometers. For example for a deep source that means:
Gradiometers have little true signal but standard noise, since their leadfields are small, the inverse of their leadfield is big and noise contributions to your voxels are high, for magnetometers its sligtly different. So if one naively uses both types in one inversion noise will often be dominated by gradiometers and signal by magnetometers. For shallow sources its reverse , the noise will be dominated by the high noise of the magnetometers and the signal comes from the gradiometers, so again things will be matched unfortunately. i guess this creates the relatively homogenic power distribution you observe (unless there's a bug in the code or the analysis setup - see below).
For some background you might want to have a look at:

Commonalities and differences among vectorized beamformers in electromagnetic source imaging.
Huang MX et al, Brain Topogr.

Do handle this two things are necessary:
1. Compute the beamformer filters for data that have baseline and task combined. searate filter computation will almost certainly create problems.
2. To localize activity use the statistics function not some simple difference measure or relative power picture.

For the reasons above, this might still not work. In this case the best workaround would be a weighted average with different, leadfield dependend weights for each voxel and modality. A simpler possibility might be simple averaging (this might work for your case, since source look very similar for the two modalities).

Michael

------------------------------------------------------------
Von: "Elena Orekhova" <Elena.Orekhova at neuro.gu.se>
Gesendet: 09.06.2011 22:48:34
An: "Email discussion list for the FieldTrip project" <fieldtrip at donders.ru.nl>
Betreff: Re: [FieldTrip] combining magnetometers and planad gradiometers for analysis

Dear Michael,
I still have not resolved this problem and do not know whether  is specific for my data(or program bag)
or the other Neuromag users also had it.
Do you use Neuromag?

Elena

------------------------------------------------------------

From: fieldtrip-bounces at donders.ru.nl [fieldtrip-bounces at donders.ru.nl] on behalf of Michael Wibral [michael.wibral at web.de]
Sent: Thursday, June 09, 2011 5:06 PM
To: Email discussion list for the FieldTrip project
Subject: Re: [FieldTrip] combining magnetometers and planad gradiometers for analysis

Hi Elena,

disregard my last email, I overlooked the att.

Michael

------------------------------------------------------------
Von: "Elena Orekhova" <Elena.Orekhova at neuro.gu.se>
Gesendet: Jun 2, 2011 8:03:16 PM
An: "Email discussion list for the FieldTrip project" <fieldtrip at donders.ru.nl>
Betreff: Re: [FieldTrip] combining magnetometers and planad gradiometers for analysis

@font-face { font-family: "\FF2D \FF33 \660E \671D "; }@font-face { font-family: "Verdana"; }@font-face { font-family: "Cambria Math"; }@font-face { font-family: "Calibri"; }@font-face { font-family: "Cambria"; }p.MsoNormal, li.MsoNormal, div.MsoNormal { margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: Cambria; }.MsoChpDefault { font-family: Cambria; }div.WordSection1 { page: WordSection1; }
Dear Michael,

I did normalize the data
(sourcepst.avg.nai=sourcepst.avg.pow./sourcepre.avg.pow)
as it was suggested in the tutorial.
I used the same scales in all plots for comparison purposes. If  automatic scaling is used the result for MAG+GRA does not look any better (see attachment).

>What you could do as a workaround is to average the separate results with a weighting per voxel and  that >is corresponding to the squared norms of the leadfields for the respective modalities for a given voxel , this >would guarantee equal amounts of backprojected noise from both modalities (if I'm not mistaken).

Thank you for your suggestion.  I may use it as a last resort.  Hopefully, there is still a way to get more optimal solution using both GRA and MAG in the same analysis...

Elena

------------------------------------------------------------

From: fieldtrip-bounces at donders.ru.nl [fieldtrip-bounces at donders.ru.nl] on behalf of Elena Orekhova [Elena.Orekhova at neuro.gu.se]
Sent: Thursday, June 02, 2011 6:22 PM
To: Email discussion list for the FieldTrip project
Subject: Re: [FieldTrip] combining magnetometers and planad gradiometers for analysis

BODY {direction: ltr;font-family: Tahoma;color: #000000;font-size: 10pt;}
Dear Michael,

I have tried to multiply the leadfield by -1 as you suggested:

for i = 1 : size (grid.leadfield, 2)
    grid.leadfield{i}(3:3:306, :) = -1*grid.leadfield{i}(3:3:306, :);
end

This had no effect on the 'lcmv' output. I attached the pictures for ‘GRA only, ’MAG only’ and ‘GRA + MAG’

In this experiment I measured evoked field in response to  the unilateral (left) click.
The source is expected to be in the right superior temporal cortex.  This is the case  for  ‘GRA only' and ’MAG only’ datasets. The combined sensors give meaningless result.

Elena

------------------------------------------------------------

From: fieldtrip-bounces at donders.ru.nl [fieldtrip-bounces at donders.ru.nl] on behalf of Michael Wibral [michael.wibral at web.de]
Sent: Wednesday, June 01, 2011 4:04 PM
To: Email discussion list for the FieldTrip project
Subject: Re: [FieldTrip] combining magnetometers and planad gradiometers for analysis

Dear Elena,

could you give the following a try: invert (*-1) the leadfileds for one of the two sensor types. Let me know what happens.
I would also be interested in taking a look at the results - maybe you could sent images off-list: Michael.Wibral <at> web.de.

Michael

------------------------------------------------------------
Von: "Elena Orekhova" <Elena.Orekhova at neuro.gu.se>
Gesendet: Jun 1, 2011 12:23:29 PM
An: "fieldtrip at donders.ru.nl" <fieldtrip at donders.ru.nl>
Betreff: [FieldTrip] combining magnetometers and planad gradiometers for analysis

@font-face { font-family: "Arial"; }@font-face { font-family: "Times"; }@font-face { font-family: "\FF2D \FF33 \660E \671D "; }@font-face { font-family: "Cambria Math"; }@font-face { font-family: "Cambria"; }p.MsoNormal, li.MsoNormal, div.MsoNormal { margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: Cambria; }p { margin-right: 0cm; margin-left: 0cm; font-size: 10pt; font-family: Times; }.MsoChpDefault { font-family: Cambria; }div.WordSection1 { page: WordSection1; }
Dear  fieldtrippers,
This message is mainly for Neuromag users.

When I do  'lcmv' beamforming analysis separately on planar gradiometers or magnetometers, I get quite meaningful results.
If I combine the two types of sensors without weighting, the result is meaningless.
Apparently, the algorithm does not take care of different scales and units of the GRA and MAG measurements.  Does anybody know how to deal with this problem?

Elena

_______________________________________________
fieldtrip mailing list
fieldtrip at donders.ru.nl
http://mailman.science.ru.nl/mailman/listinfo/fieldtrip