<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<style type="text/css" style="display:none"><!-- p { margin-top: 0px; margin-bottom: 0px; }--></style>
</head>
<body dir="ltr" style="font-size:12pt;color:#000000;background-color:#FFFFFF;font-family:Calibri,Arial,Helvetica,sans-serif;">
<p></p>
<div>Dear Stefan and all other Fieldtrip-users, </div>
<div><br>
</div>
<div>Many thanks for your input!</div>
<div><br>
</div>
<div>We have indeed also thought about this segmentation approach that you described and explored both alternatives: </div>
<div>(1) using 5 minutes to compute the covariance</div>
<div>(2) segmenting the 5 minutes into segments of 1 sec., computing the covariance in each segment separately), and concatenating the segments afterward. </div>
<div><br>
</div>
<div>Subsequently, we parcellated the sources using an atlas to get the mean timecourse within each ROI. </div>
<div><br>
</div>
<div>When plotting these mean timecourses, we observe that in case (1), the pattern of source activity within each ROI does not fluctuate much across time, i.e., sources that are “off” during the first few seconds of the recording, will be “off” during the
whole 5 minutes. The same thing applies to the “on” sources. In case (2), however, the pattern of “on/off” sources varies across time. </div>
<div><br>
</div>
<div>Since we are planning to use these reconstructed source time-series for HsMM-MAR modeling (which is a means to assess recurring patterns of functional connectivity signatures and their dynamics on a sub-second time-scale), we are concerned that using the
entire 5-minute recording for covariance computation and the generation of the spatial filter may introduce stationarity into the source time-series (due to the underlying assumption of stationarity). On the other hand, the segmentation prior to source analysis
into 2 or 1-second segments would be arbitrary and there is no guarantee that this time window will capture a “wide sense stationary” data segment. </div>
<div><br>
</div>
<div>Apart from the potential issue explained above, it is generally advised to use as much data as possible for a robust estimation of the covariance matrix, which argues in favor of approach (1).</div>
<div>Papers that advise this and/or follow this approach are for instance</div>
<div>- Brookes et al. (2007) [https://doi.org/10.1016/j.neuroimage.2007.09.050]<br>
</div>
<div>- Van Veen et al. (1997) [https://doi.org/10.1109/10.623056]<br>
</div>
<div>- Knyazev et al. (2008) [https://doi.org/10.1016/j.neuroimage.2007.09.050]<br>
</div>
<div>- Brookes et al. (2011) [https://doi.org/10.1073/pnas.1112685108]<br>
</div>
<div><br>
</div>
<div>We would highly appreciate any further input/opinions/suggestions on this!<br>
</div>
<p><br>
</p>
<p>Best, <br>
</p>
<p>Hanna<br>
</p>
<p><br>
</p>
<div id="Signature">
<div name="divtagdefaultwrapper" style="font-family:Calibri,Arial,Helvetica,sans-serif; font-size:; margin:0">
<div><font color="#333333" face="Cambria, serif"><b style="">Hanna Honcamp</b></font></div>
<font color="#808080" face="Cambria, serif">PhD Candidate | BAND Lab</font>
<div>
<div><font color="#808080" face="Cambria, serif">Faculty of Psychology and Neuroscience</font></div>
<div><font color="#808080" face="Cambria, serif">Dept. NP&PP | Maastricht University</font></div>
</div>
</div>
</div>
</body>
</html>