<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<style type="text/css" style="display:none;"><!-- P {margin-top:0;margin-bottom:0;} --></style>
</head>
<body dir="ltr">
<div id="divtagdefaultwrapper" style="font-size:12pt;color:#000000;font-family:Calibri,Helvetica,sans-serif;" dir="ltr">
<p>Hello community,</p>
<p><br>
</p>
<p>As the title says, I am trying to run a statistical analysis between two conditions at the source level. I am performing a single-subject, between-trials statistical analysis using the Montecarlo method and ft_sourcestatistics, on localized sources using
the LCMV beamformer, which I would like to constraint over a specific ROI (the motor cortex).
<br>
</p>
<p><br>
</p>
<p>However, the tutorials are very limited in the scope of running statistics at the source level. Plus, it seems that the incorporation of prior knowledge in the form of ROIs is still not implemented in the function (?). Is there any way to bypass this and
test for significance over specific ROIs? I understand that cluster analysis can only be used to accept/reject the null-hypothesis, and the locations of the clusters cannot be used as spatial markers of the activity, something which is specified here: <a href="http://www.fieldtriptoolbox.org/faq/how_not_to_interpret_results_from_a_cluster-based_permutation_test/" class="OWAAutoLink" id="LPlnk615497" previewremoved="true" tabindex="-1" disabled="true">http://www.fieldtriptoolbox.org/faq/how_not_to_interpret_results_from_a_cluster-based_permutation_test/</a></p>
<p><br>
</p>
<p>At first, I thought I would create a mask over the ROIs I am interested in, and use that mask to "zero out" the power in all other source locations before running the statistical analysis. Is that the "correct" way to go about it? Should I normalize the
power of the remaining sources in some way, since I am removing power from other areas across the brain?</p>
<p><br>
</p>
<p>Obviously if any of you have any other suggestions, I would gladly read them.</p>
<p><br>
</p>
<p>Thank you all for your time!</p>
<p><br>
</p>
<p>P.S. I am attaching my code for the statistical analysis down below. This code produces the exact same results both w/ and w/o the atlas/roi cfg options.<br>
</p>
<p><br>
</p>
<p></p>
<div>statistics<br>
cfg = [];<br>
cfg.dim = sources_all.dim;<br>
cfg.method = 'montecarlo';<br>
cfg.statistic = 'ft_statfun_indepsamplesT'; %function to use when calculating the parametric t-values<br>
</div>
<div><br>
</div>
<div>cfg.parameter = 'pow';<br>
<br>
cfg.alpha = 0.05; % (type-I error rate) - note that this only implies single-sided testing<br>
cfg.tail = 0; % two-sided test<br>
cfg.correcttail = 'alpha'; % Bonferroni correction for the two tails, i.e., divide alpha by two<br>
% cfg.correcttail = 'prob'; % multiplying the p-values (in stat.prob, stat.posclusters.prob and stat.negclusters.prob) with a factor of two<br>
<br>
cfg.correctm = 'cluster'; % the correction to use<br>
cfg.numrandomization = 1000;<br>
<br>
cfg.design = eeg_all_curr.trialinfo;<br>
% cfg.uvar = 1; % row of design matrix that contains unit variable (in this case: trials)<br>
cfg.ivar = 1; % row of design matrix that contains independent variable (the conditions)</div>
<div><br>
</div>
<div>% prepare neighbours sctructure<br>
cfg_neigh = [];<br>
cfg_neigh.elec = elec;<br>
cfg_neigh.method = 'triangulation';<br>
cfg_neigh.feedback = 'no'; % visualizes the neighbors<br>
neighb = ft_prepare_neighbours(cfg_neigh);<br>
<br>
cfg.neighbours = neighb; % the spatial structire<br>
cfg.minnbchan = 2; % minimum number of channels required to form a cluster<br>
<br>
% load the atlas<br>
atl = ft_read_atlas([paths.templ '\atlas\brainnetome\BNA_MPM_thr25_1.25mm.nii']);<br>
<br>
% specify ROIs<br>
cfg.atlas = atl;<br>
cfg.roi = atl.tissuelabel(contains(atl.tissuelabel, 'precentral', 'IgnoreCase', true) & contains(atl.tissuelabel, 'Left', 'IgnoreCase', true));<br>
<br>
stats = ft_sourcestatistics(cfg, sources_all);<br>
<br>
save('stats.mat', 'stat');</div>
<p></p>
</div>
</body>
</html>