[FieldTrip] Seizure localization problems

[Dr.-Ing. Matthias Dümpelmann]

Dear colleagues,

we are trying to evaluate systematically the localization  of rhythmic activity at the onset of epileptic seizures in patients, who are subsequently evaluated with intracranial recordings.

We are trying to use DICS but quite often receive results, which are obviously wrong. In the attached screenshots you can see the peace of EEG, which is evaluated and the localization result with a maximum in the cerebellum.

This a typical localization for the cases with obvious miss localizations. A further seizure with a similar EEG pattern was localized in the right temporal lobe, which matches the EEG pattern and also the intracranial findings.

Head model is generated with ASA, also the electrodes file is in the ASA format. The geometrical match of source points, head model and electrodes is checked using ft_plot_.... methods.

Here is the code sniplet relevant for the localization algorithm:

% Analysis frequency
SourceFrequency = 4.0

% read eeg, electrodes and volume conductor model
%[dat] = ft_read_data(filename);
hdr = ft_read_header(filename);
elec =  ft_read_sens(elecfile,  'fileformat','asa_elc');
headmodel = ft_read_vol(volumefilename,'fileformat','asa_vol');
mri = ft_read_mri(mrifilename, 'fileformat','asa_mri');

% eeg data preprocessing
cfgeeg.dataset = filename;      
cfgeeg.continuous   = 'yes'
cfg.covariance       = 'yes';
cfg.covariancewindow = 'all';
cfg.demean = 'yes'

[data] = ft_preprocessing(cfgeeg);
 
% Freqanalysis for beamformer
cfg = [];
cfg.channel      = {'*'};
cfg.method       = 'mtmfft';
cfg.foilim     = [1 100];
cfg.tapsmofrq  = 4;
cfg.taper    = 'hanning';
cfg.output       = 'powandcsd';

 
% for common filter over conditions
powcsd_all      = ft_freqanalysis(cfg, data);

% Create leadfield grid
cfg                 = [];
cfg.channel         = elec.label;
cfg.elec            = elec;           
cfg.headmodel       = headmodel;
cfg.dics.reducerank = 3; % default for MEG is 2, for EEG is 3
cfg.grid.resolution = 0.5;   % use a 3-D grid with a 0.5 cm resolution
cfg.grid.unit       = 'cm';
cfg.grid.tight      = 'yes';
[grid] = ft_prepare_leadfield(cfg);

% source analysis
cfg              = []; 
cfg.channel      = elec.label; 
cfg.method       = 'dics';
cfg.frequency    = SourceFrequency;  
cfg.grid         = grid; 
cfg.headmodel    = headmodel;
cfg.elec         = elec;     
cfg.senstype     = 'EEG'; 
cfg.dics.keepfilter   = 'yes'; % We wish to use the calculated filter later on
cfg.dics.projectnoise = 'yes';
cfg.dics.lambda  = '0.3%';
source_raw = ft_sourceanalysis(cfg, powcsd_all);


I played with a couple of parameters like cfg.dics.lambda, but without success. 

Do you have ideas how to improve our localization results ?

Thank you,

Matthias



UNIVERSITÄTSKLINIKUM FREIBURG
Epileptologie
Breisacher Str. 64 · 79106 Freiburg
Telefon: +49 761 270-52410
matthias.duempelmann at uniklinik-freiburg.de
 
www.uniklinik-freiburg.de

-------------- next part --------------
A non-text attachment was scrubbed...
Name: analysisinterval_Seizure2.png
Type: image/png
Size: 110002 bytes
Desc: analysisinterval_Seizure2.png
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20160623/c9cfdcdb/attachment-0002.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: localizationresults.png
Type: image/png
Size: 589001 bytes
Desc: localizationresults.png
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20160623/c9cfdcdb/attachment-0003.png>