[FieldTrip] (no subject)
Kaelasha Tyler
ktyler at swin.edu.au
Thu Nov 6 06:20:31 CET 2014
Hi all,
Some great suggestions and resources here. I'll try a few of these suggestions out and get back to the post with results.
Cheers,
Kaelasha
________________________________
From: fieldtrip-bounces at science.ru.nl [fieldtrip-bounces at science.ru.nl] on behalf of Joram van Driel [joramvandriel at gmail.com]
Sent: Thursday, 6 November 2014 2:15 AM
To: FieldTrip discussion list
Subject: Re: [FieldTrip] (no subject)
Dear Kaelasha,
I recently came across the exact same problem in my Neuromag source analysis: mislocalization of motor beta over M1 (it was too posterior in my case). I used LCMV, so my analyses are quite similar to yours.
Do you use the Elekta Maxfilter routine of tSSS (a preprocessing step to remove non-brain noise)? There, I had the option -trans enabled, which transforms the sensor data of every subject to a common space, so that subjects become better comparable in sensor space. But this should not be used if you do source analysis, because the forward model uses the coregistered MRI, taking into account how that subject was positioned in the scanner. When I redid tSSS without the -trans option, and recomputed the forward model, motor beta suppression was right where it should be.
As for the coregistration, we used a polhemus point method of digitized 3D points, which allows for continuous tracking of head position in the Elekta scanner. With those points, there are built-in ways to coregister the MRI using the Elekta software (somebody else helped me with that, so I don't know the details how to do this).
With the coregistered MRIs, I just followed the fieldtrip tutorial up to ft_prepare_leadfield, and everything went fine.
Best,
Joram
Thursday, 6 November 2014 2:14 AM
Hi Kaelasha,
It's hard to evaluate why the localization in not exactly how you would expect it, but here are some idea you could try from the top of my head.
Note I never used LCMV beamformer myself, although I've done DICS.
Also, to me there seems to be not something obviously wrong in the code or pictures, so I'm just suggesting some thoughts here.
- Beamformer performance is VERY dependent on the data. Check your sensor level data, clean it more if necessary.
- Play around with including more or less data centered around the component you are localizing for both the common filter as the data you localize. Since I don't know the details of the data (nor do i have experience with LCMV) your choices here are hard to evaluate.
- Plot the scalp as well to get a better idea whether your co-registration is accurate. If you have additional headpoints e.g. by using a Polhemus, that would help.
- Play around with some regularization - in ft_sourceanalysis use e.g.: cfg.alpha = '10%';
- Evaluate your lcmv beamformer with a dipole fit.
- Check other subjects/recordings - other recordings might be better (or worse...)
Finally, if you haven't already, take a look at the scripts we used in our recent workshop using a Neuromag system (although as I said I see nothing wrong with your script):
http://fieldtrip.fcdonders.nl/workshop/stockholm2014q3
The beamformer lecture there given by Robert on Beamforming is also online: http://natmeg.se/wp/activities/natmeg-lectures/
Good luck!
Stephen
On Wed, Nov 5, 2014 at 2:50 PM, Kaelasha Tyler <ktyler at swin.edu.au<mailto:ktyler at swin.edu.au>> wrote:
Dear Field trippers,
And especially dear Neuromag users,
I have been going back over the analysis of my study, ironing out issues, and am finding that I am still having problems with source analysis coming up with some unexpected and probably inaccurate locations for clusters of significant effects.
I have decided to go back and look at a simple right handed button push, in one subject, to check if I was able to locate a realistic region in M1.
>From the attached jpeg, you will see that the analysis is not accurately locating left hemisphere motor regions. I know I can't expect absolute accuracy, but I would be hoping for more than this.
I have included my code (below) and if any one has any suggestions about how to remedy this and produce more accurate results, please let me know!!!
I am wondering if I have realigned my mri to the neuromag system correctly, and have attached another jpeg of my volume conduction model, plotted alongside the neuromag sensors.
Neuromag users, does this look accurate to you?
As a general fix, I was wondering if any neuromag users would be happy to supply their FT code for use of mri, including realignment of mri to the neuromag system, segmentation and the creation of the volume conduction model?
As always, any help is much appreciated!
Here is the code as I have been currently using it:
%% Load MRI
mri = ft_read_mri('Subject1.nii');
%% Realigning
cfg=[];
cfg.coordsys = 'neuromag';
mri = ft_volumerealign(cfg, mri);
%% Segmentation.
cfg = [];
cfg.coordsys='neuromag';
cfg.units='mm';
seg = ft_volumesegment(cfg, mri);
%% Prepare volumne conduction model
cfg = [];
cfg.method = 'singleshell';
vol = ft_prepare_headmodel(cfg, seg);
%% Non warped grid for use in single subject comparison btw conditions
cfg = [];
cfg.grid.xgrid = -20:1:20;
cfg.grid.ygrid = -20:1:20;
cfg.grid.zgrid = -20:1:20;
cfg.grid.unit = 'cm';
cfg.grid.tight = 'yes';
cfg.vol = vol;
sourcemodel = ft_prepare_sourcemodel(cfg);
%% Calculates leadfields
cfg = [];
cfg.grid=sourcemodel;
cfg.vol = vol;
cfg.channel={'MEGGRAD'};% For Planar gradiometers only
cfg.grad=grad;%This needs to be edited to represent the MEG data set which has just been loaded
grid = ft_prepare_leadfield(cfg, Condition1);
%% Append individuals data from two conditions.
%This is so as to create a common filter, from the two conditions.
cfg=[];
cfg.appenddim = 'rpt';
combined = ft_appendtimelock(cfg, Condition1, Condition2);
%% Calculate covariance and timelock. Input data should NOT have been timelocked before this.
cfg = [];
cfg.keeptrials = 'yes';
cfg.covariance = 'yes';
cfg.covariancewindow = 'all';
cfg.vartrllength = 2;
cfg.channel='MEGGRAD';
tlckavgCond1 = ft_timelockanalysis(cfg, Condition1);
tlckavgCombined = ft_timelockanalysis(cfg, combined);
%% Create spatial filter using the lcmv beamformer
cfg = [];
cfg.grid = grid; % leadfield, which has the grid information
cfg.grad=grad;
cfg.vol = vol; % volume conduction model (headmodel)
cfg.keepfilter = 'yes';
cfg.lcmv.fixedori = 'yes'; % project on axis of most variance using SVD
cfg.channel='MEGGRAD';
sourceCombined = ft_sourceanalysis(cfg, tlckavgCombined);%
%%
cfg.grid.filter = sourceCombined.avg.filter;
cfg.rawtrial='yes';
sourceCond1=ft_sourceanalysis(cfg, tlckavgCond1);
%% The cluster stats at source space:
%%
cfg=[];
cfg.dim=sourceCond1.dim;
cfg.method = 'montecarlo';
cfg.statistic = 'depsamplesT';
cfg.parameter = 'pow';
cfg.correctm = 'cluster';
cfg.numrandomization = 1000;
cfg.alpha = 0.05;
cfg.tail = 0;
cfg.clusteralpha = 0.005;
cfg.minnbchan = 8;
cfg.correcttail = 'alpha';
cfg.clusterstatistic = 'maxsum';
Nsub = 84;
cfg.design(1,1:2*Nsub) = [ones(1,Nsub) 2*ones(1,Nsub)];
cfg.design(2,1:2*Nsub) = [1:Nsub 1:Nsub];
cfg.ivar = 1; % the 1st row in cfg.design contains the independent variable
cfg.uvar = 2; % the 2nd row in cfg.design contains the subject number
stat = ft_sourcestatistics(cfg, sourceCond1, sourceCond2);
%% interpolate the t maps to the structural MRI of the subject %
cfg = [];
cfg.parameter = 'mask';
statplot = ft_sourceinterpolate(cfg,stat, mri);
%% plot the t values on the MRI %
cfg = [];
cfg.method = 'slice';
cfg.funparameter = 'mask';
cfg.maskparameter = 'funparameter';
figure
ft_sourceplot(cfg, statplot);
%%
_______________________________________________
fieldtrip mailing list
fieldtrip at donders.ru.nl<mailto:fieldtrip at donders.ru.nl>
http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
--
Joram van Driel, MSc.
PhD student @ University of Amsterdam
Brain & Cognition @ Department of Psychology
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20141106/7c5f6e03/attachment-0002.html>
More information about the fieldtrip
mailing list