<html><head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
</head>
<body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;" class="">
Dear Manon,
<div class=""><br class="">
</div>
<div class="">Based on your information I believe things might have gone wrong when you interpolated your reconstructed coherence onto the MRI template.</div>
<div class=""><br class="">
</div>
<div class="">Both inputs (i.e. the source reconstructed coherence, and the template) need to be:</div>
<div class="">1) in the same coordinate system</div>
<div class="">(and just to be sure) 2) in the same metric units </div>
<div class=""><br class="">
</div>
<div class="">Regarding 1): if source.pos is according to the 4D neuroimaging ALS-convention, where the coordinate axes are based on the fiducials, and the template positions (as imposed by the mri.transform) are in ACPC-based RAS, the interpolation goes wrong</div>
<div class="">Regarding 2): if the source positions are in ‘cm’ , and the MR image in ‘m’ (or ‘mm’, ‘feet’, ‘yard’ or something else), FieldTrip typically tries to equate the units before interpolation, but it may go wrong there too.</div>
<div class=""><br class="">
</div>
<div class="">Another thing to look into perhaps is the fact that your cross-spectral density matrix is highly rank deficient, given the number of trials (and the fact that you used just a single ‘hanning’ taper for spectral decomposition). To get a stable
source reconstruction I’d recommend using regularization in ft_sourceanalysis (cfg.lambda), as well as multitapering for ft_freqanalysis (cfg.tapsmofrq in combination with cfg.taper = ‘dpss’).</div>
<div class=""><br class="">
</div>
<div class="">Final note: the coherence that is outputted by the classical DICS implementation is coherence^2 and needs to be square-rooted to be comparable in magnitude to the coherence that is the output of ft_connectivityanalysis.</div>
<div class=""><br class="">
</div>
<div class="">Best wishes,</div>
<div class="">Jan-Mathijs</div>
<div class=""><br class="">
</div>
<div class=""><br class="">
</div>
<div>
<blockquote type="cite" class="">
<div class="">On 20 May 2019, at 18:35, CHATEAUX Manon <<a href="mailto:manon.chateaux@ens-cachan.fr" class="">manon.chateaux@ens-cachan.fr</a>> wrote:</div>
<br class="Apple-interchange-newline">
<div class="">
<title class=""></title>
<div style="font-family:Arial;font-size:14px" class="">
<p class=""><br class="">
<span style="font-family:times new roman,times,serif;" class="">Dear FieldTrip community,<br class="">
<br class="">
I am encountering problems in doing source reconstruction on coherences, computed between signal recorded by my 248 magnetometers (4D system) and the signal of a given EMG electrode that I called ‘FDI’.<br class="">
<br class="">
<br class="">
<br class="">
After preprocessing and cleaning my MEG and EMG data, I had a look at the distribution of the coherence values at the channel level. Around the beta band, I have quite clean results (See Figure 1).<br class="">
Here is the script I used to obtain this figure, from my ‘data_CMC’ that is the clean data, ready to undergo CMC analysis :<br class="">
cfg = [];<br class="">
cfg.output = 'powandcsd'; %returns the power and the cross-spectra<br class="">
cfg.method = 'mtmfft';<br class="">
cfg.taper = 'hanning';<br class="">
cfg.foilim = [18 100];<br class="">
cfg.keeptrials = 'no';<br class="">
cfg.channel = {'MEG' 'OP' 'FDI' 'ECD' 'FCD' 'TB' 'BB'};<br class="">
cfg.channelcmb = {'MEG' 'MEG'; 'MEG' 'OP'; 'MEG' 'FDI'; 'MEG' 'ECD'; 'MEG' 'FCD'; 'MEG' 'TB'; 'MEG' 'BB'};<br class="">
cfg.pad = 'nextpow2';<br class="">
<br class="">
freq_cross_spctr.FG = ft_freqanalysis(cfg, data_CMC.FG);<br class="">
<br class="">
Remark : I have 150 trials and they are very short, some are as short as about 100 ms which lead me to adapt cfg.foilim in order that it does not go under about 18Hz. Indeed, I calculated it so that we can have 2 periods of the minimum frequency in my shortest
trial.<br class="">
<br class="">
cfg = [];<br class="">
cfg.xlim = [CMC_cross_spectr.FG.freq(1,1) CMC_cross_spectr.FG.freq(1,5)]; %Between 18 and 26Hz<br class="">
cfg.refchannel = 'FDI';<br class="">
cfg.layout = lay;<br class="">
cfg.colorbar = 'yes';<br class="">
cfg.parameter = 'cohspctrm';<br class="">
cfg.marker = 'on';<br class="">
cfg.dataname = ['FG_' int2str(floor(CMC_cross_spectr.FG.freq(1,1))) '_' int2str(floor(CMC_cross_spectr.FG.freq(1,5))) '_' cfg.refchannel];<br class="">
ft_topoplotER(cfg, CMC_cross_spectr.FG);<br class="">
<br class="">
<br class="">
<br class="">
<br class="">
But now I want to precisely localize these coherences, in order to be able to answer my scientific question. I used the DICS beamformer method as advised by the tutorials. However, the reconstruction gives a terrible topographical mapping of the coherences
that is not clean at all contrary to what I saw at the sensor level (Figure 2). Plus, it displays very low coherence values.<br class="">
<br class="">
For information, I don’t have any MRI of the participant. Thus, I used the fiducial positions, gathered just before the experiment started, and altered an MRI template so that it is aligned – as well as possible – to these recorded fiducials positions. (The
participant moved less than 1mm during the experiment). Then I computed the segmented MRI and the head and source model from this manual alignment.<br class="">
Here is the script that follows the computation of the forward model and that lead me to figure 2 :<br class="">
<br class="">
cfg = [];<br class="">
cfg.method = 'dics';<br class="">
cfg.refchan = 'FDI';<br class="">
cfg.frequency = [freq_cross_spctr.FG.freq(1,1) freq_cross_spctr.FG.freq(1,5)];<br class="">
cfg.headmodel = hdm_cm;<br class="">
cfg.sourcemodel = sourcemodel;<br class="">
cfg.grad = sensors_cm;<br class="">
cfg.dics.reducerank = 2; %default value for MEG. This is done to remove the weakest orientation.<br class="">
cfg.dics.normalize = 'yes'; %I use this because we don't have any baseline or second condition to compare to, for the moment<br class="">
source_coh_lft = ft_sourceanalysis(cfg, freq_cross_spctr.FG);<br class="">
<br class="">
source_coh_lft.pos = sourcemodel.pos;<br class="">
source_coh_lft.dim = sourcemodel.dim;<br class="">
<br class="">
cfg = [];<br class="">
cfg.parameter = 'coh';<br class="">
cfg.interpmethod = 'nearest';<br class="">
source_coh_int = ft_sourceinterpolate(cfg, source_coh_lft, mri_realigned_to_headshape_m);<br class="">
<br class="">
<br class="">
cfg = [];<br class="">
cfg.method = 'surface';<br class="">
cfg.surffile = 'surffile.mat';<br class="">
cfg.funparameter = 'coh';<br class="">
cfg.surfdownsample = 10; %gives more beautiful results<br class="">
ft_sourceplot(cfg, source_coh_int);<br class="">
colormap jet<br class="">
<br class="">
<br class="">
<br class="">
<br class="">
<strong class="">My questions are the following :</strong><br class="">
- Why are the two maps so different ? Why are there such low coherence values in Figure 2? What am I missing?<br class="">
- Is there any way to improve the results of Figure 2?<br class="">
- Can I compute the CMC first at the sensor level and then do a reconstruction?<br class="">
- I am not confident with the method I used to align the template MRI to my recorded fiducials. Is there a more rigorous way to do it?<br class="">
<br class="">
I am new to signal processing and especially to FieldTrip so any remark, suggestion or resource would be extremely helpful and highly appreciated!<br class="">
<br class="">
<br class="">
Thank you very much in advance for your help and for the time you have spared reading through this message.<br class="">
<br class="">
Manon Châteaux<br class="">
Neuroscience Master Student</span><br class="">
<br class="">
</p>
<div class=""><a href="http://www.ens-paris-saclay.fr/" rel="noopener noreferrer" style="color: #00778b; text-decoration: none" target="7mL6wUk9oqyI1TJNqtITpjN" class=""><img class="gfx" src="https://webmail.ens-paris-saclay.fr/imp/view.php?ctype=image%2F%2A&id=2.3&imp_img_view=data&actionID=view_attach&muid=%7B5%7DINBOX2278&view_token=RPdTvn1Sj9EF8RucbH-0qUG" style="border: 0; max-height: 100px; max-width: 230px; padding-left: 2px"></a>
<table align="left" style="border: 0; cellpadding: 0; cellspacing: 0; margin-bottom: 25px" width="400px" class="">
<tbody class="">
<tr class="">
<td align="left" style="color: #00778b; font-size: 9pt" class=""><span style="font-size: 12pt" class="">Manon CHATEAUX</span><br class="">
<br class="">
<i class="">Normalienne Etudiante en dernière année du département de Biologie</i><br class="">
<br class="">
<span class="nl" style="text-align: left;font-size: 9pt"><a href="mailto:manon.chateaux@ens-cachan.fr" class="">manon.chateaux@ens-cachan.fr</a></span><br class="">
</td>
</tr>
</tbody>
</table>
</div>
</div>
<span id="cid:37B04B25-DF56-41C8-B307-054618A0819F@home"><Figure 1.jpg></span><span id="cid:752B5C9F-4112-4C04-9AEA-D6F46E41CC61@home"><Figure 2.jpg></span>_______________________________________________<br class="">
fieldtrip mailing list<br class="">
<a href="https://mailman.science.ru.nl/mailman/listinfo/fieldtrip" class="">https://mailman.science.ru.nl/mailman/listinfo/fieldtrip</a><br class="">
https://doi.org/10.1371/journal.pcbi.1002202<br class="">
</div>
</blockquote>
</div>
<br class="">
</body>
</html>