[FieldTrip] Advice on using LCMV beamformer

[Stephen Whitmarsh]

Hi everyone,

I hope there is someone out there who has some time to help me with using LCMV beamformer. I have no experience with it yet, although a little with DICS, and understand they are at least conceptually very similar. However, I am running into some questions. Before I go on, my data on sensor level looks good and it works out with dipolefitting as well (attached).

I would like to beamform an early (unilaterally stimulated) sensory evoked field. There is no conditional difference, I would just like to extract the timecourse of some maximum voxels. In other words:
1) get a contralateral cluster or a number of voxels with maximum amplitude.
2) extract the timecourses of those clusters, then average them.

For step 1, it seem to have several options, either:
1a) calculate the filter on the whole timecourse (a common filter as would be used in DICS)
1b) use that common filter to calculate the amplitude of the early component
1c) do the same for the baseline period
1d) subtract the baseline activation from the active period (or take a ratio)

Alternatively, I could:
2a) beamform the active period
2b) beamform the baseline period
2c) subtract the baseline activation from the active period (or take a ratio)

Or it seems, I might even:
3a) beamform the active period
3b) divide the active period by the projected noise

Do I get this correctly? Does anyone have an argument for one of these option?

Anyway, first practical problem is that when I plot either the whole, the baseline or the activation period I get the same results (attached). A difference therefore doesn't work, obviously. So this raises the question how the selection of latency works in ft_sourceanalysis? I trust cfg.latency works, but it doesn't seem to make a difference for me. At the same time, using selectdata to select a time period doesn't make a difference either. I'm obviously doing something wrong, but what?

Anyway, when I get this figured out, I still need to extract the timecourses. How would I proceed doing this? I imagine the fastest way would not be to loop ft_sourceanalysis for each timepoint (using a common filter I would think). Should I 'manually' multiply the data with the filter?

It would be great to have some feedback on the general idea, and perhaps someone can identify what I am doing wrong (script below). I would be happy to finish the LCMV beamformer tutorial on the FT website once I get it figured out, as I can imagine more people would have these questions.

All the best and thanks,
Stephen


% timelock average + covariance

cfg = [];
cfg.covariance = 'yes';
cfg.covariancewindow = 'all';
cfg.trials = find(clean_data.trialinfo(:,2) == 1);
ERF_single{isubject} = ft_timelockanalysis(cfg,clean_data);
cfg.trials = find(clean_data.trialinfo(:,2) == 2);
ERF_double{isubject} = ft_timelockanalysis(cfg,clean_data);

% combine gradiometers (single stim)
cfg = [];
ERF_single_cmb{isubject} = ft_combineplanar(cfg,ERF_single{isubject});
ERF_double_cmb{isubject} = ft_combineplanar(cfg,ERF_double{isubject});

% load headmodel MEG
temp = load(['/home/stephen/analysis/metacognition/sourcemodel/pp_' num2str(isubject) '_headmodel_singleshell.mat']);
headmodel_meg{isubject} = temp.headmodel_meg;

% headmodel_meg = ft_convert_units(headmodel_meg, 'cm');
gridtemp = load(['/home/stephen/analysis/metacognition/sourcemodel/pp_' num2str(isubject) '_grid.mat']);

hdr             = ft_read_header(flist{1});
cfg             = [];
cfg.grad        = hdr.grad;
cfg.vol         = headmodel_meg{isubject};
cfg.grid        = gridtemp.grid;
cfg.channel     = {'MEGGRAD'};
cfg.normalize   = 'yes';
leadfield       = ft_prepare_leadfield(cfg);

% get the filter
cfg = [];
cfg.method              = 'lcmv';
cfg.grid                = leadfield;
cfg.vol                 = headmodel_meg{isubject};
cfg.keepfilter          = 'yes';
cfg.senstype            = 'meg';
cfg.channel             = 'meggrad';
cfg.lcmv.fixedori       = 'yes'; % use only right gradiometers
cfg.reducerank          = 2;
source_lcmv_single{isubject}   = ft_sourceanalysis(cfg, ERF_single{isubject});
source_lcmv_double{isubject}   = ft_sourceanalysis(cfg, ERF_double{isubject});

% project post stim latency with previous common filter

cfg = [];
cfg.latency = latency;
ERF_single_latency{isubject} = ft_selectdata(cfg,ERF_single{isubject});

cfg = [];
cfg.method              = 'lcmv';
cfg.grid                = leadfield;
cfg.vol                 = headmodel_meg{isubject};
cfg.grid.filter         = source_lcmv_single{isubject}.avg.filter;
cfg.channel             = 'meggrad';
cfg.senstype            = 'meg';
% cfg.latency             = latency;
cfg.lcmv.fixedori       = 'yes';
cfg.reducerank          = 2;
source_lcmv_single_latency{isubject} = ft_sourceanalysis(cfg, ERF_single_latency{isubject});

% baseline

cfg = [];
cfg.latency = [-0.2 0];
ERF_single_baseline{isubject} = ft_selectdata(cfg,ERF_single{isubject});

cfg = [];
cfg.method              = 'lcmv';
cfg.grid                = leadfield;
cfg.vol                 = headmodel_meg{isubject};
cfg.grid.filter         = source_lcmv_single{isubject}.avg.filter;
cfg.channel             = 'meggrad';
cfg.senstype            = 'meg';
% cfg.latency             = latency;
cfg.lcmv.fixedori       = 'yes';
cfg.reducerank          = 2;
source_lcmv_single_baseline{isubject} = ft_sourceanalysis(cfg, ERF_single_baseline{isubject});

% get difference between baseline and component
source_lcmv_single_diff{isubject} = source_lcmv_single_latency{isubject};
source_lcmv_single_diff{isubject}.avg.pow = source_lcmv_single_latency{isubject}.avg.pow - source_lcmv_single_baseline{isubject}.avg.pow;

template_mri = ft_read_mri(['/opt/fieldtrip/external/spm8/templates/T1.nii']);

cfg = [];
cfg.voxelcoord = 'no';
cfg.parameter = 'avg.pow';
cfg.interpmethod = 'nearest';
source_lcmv_single_int{isubject} = ft_sourceinterpolate(cfg, source_lcmv_single{isubject}, template_mri);
source_lcmv_single_latency_int{isubject} = ft_sourceinterpolate(cfg, source_lcmv_single_latency{isubject}, template_mri);
source_lcmv_single_diff_int{isubject} = ft_sourceinterpolate(cfg, source_lcmv_single_diff{isubject}, template_mri);

cfg = [];
cfg.method = 'slice';
cfg.funparameter = 'avg.pow';
cfg.maskparameter = cfg.funparameter;
cfg.funcolorlim = [3e-22 5e-22];
% ft_sourceplot(cfg,source_lcmv_single_latency_int{isubject});
% ft_sourceplot(cfg,source_lcmv_single_int{isubject});
ft_sourceplot(cfg,source_lcmv_single_latency_int{isubject});

cfg = [];
cfg.method = 'slice';
cfg.funparameter = 'pow';
cfg.funcolorlim = [-1e-23 1e-23];
ft_sourceplot(cfg,source_lcmv_single_diff_int{isubject});




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20150716/6f6439ba/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: SEF_lcmv.jpg
Type: image/jpeg
Size: 25224 bytes
Desc: SEF_lcmv.jpg
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20150716/6f6439ba/attachment-0001.jpg>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: SVD_pp_6_0.025-0.055.png
Type: image/png
Size: 107491 bytes
Desc: SVD_pp_6_0.025-0.055.png
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20150716/6f6439ba/attachment-0001.png>