<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;background-color:#FFFFFF;font-family:Calibri,Arial,Helvetica,sans-serif;">
<p>Hi Jenny,</p>
<p><br>
</p>
<p>Just a quick word to say that once you removed bad channel - you don't have to interpolate if you use the LIMO EEG plug-in. Group level statistics accounts for missing channels across subjects when computing the null distribution for cluster stats. That
means that if you have many subjects (say > 20) then you can be a little more stringent if you wish too.</p>
<p><br>
</p>
<p>Cyril</p>
<p><br>
</p>
<p><br>
</p>
<div id="Signature">
<div id="divtagdefaultwrapper" style="font-size:12pt; color:#000000; background-color:#FFFFFF; font-family:Calibri,Arial,Helvetica,sans-serif">
<span style="font-family:Garamond,Georgia,serif">--</span><br style="font-family:Garamond,Georgia,serif">
<div id="divtagdefaultwrapper" style="font-size:12pt; color:#000000; background-color:#FFFFFF; font-family:Calibri,Arial,Helvetica,sans-serif">
<span style="font-family:Garamond,Georgia,serif">Dr Cyril Pernet,<br>
Senior Academic Fellow, <span style="font-family:Garamond,Georgia,serif">Neuroimaging Sciences</span><br>
Centre for Clinical Brain Sciences (CCBS)<br>
<br>
The University of Edinburgh<br>
Chancellor's Building, Room GU426D<br>
49 Little France Crescent<br>
Edinburgh EH16 4SB<br>
</span><a id="LPNoLP" href="mailto:cyril.pernet@ed.ac.uk"><span style="font-family:Garamond,Georgia,serif">cyril.pernet@ed.ac.uk</span></a><br style="font-family:Garamond,Georgia,serif">
<span style="font-family:Garamond,Georgia,serif"></span><a id="LPNoLP" href="http://www.sbirc.ed.ac.uk/cyril"><span style="font-family:Garamond,Georgia,serif"></span></a><a id="LPNoLP" href="http://www.sbirc.ed.ac.uk/cyril"><span style="font-family:Garamond,Georgia,serif">http://www.sbirc.ed.ac.uk/cyril</span></a><br style="font-family:Garamond,Georgia,serif">
<span style="font-family:Garamond,Georgia,serif"></span><span style="font-family:Garamond,Georgia,serif"></span><a id="LPNoLP" href="http://www.ed.ac.uk/edinburgh-imaging "><span style="font-family:Garamond,Georgia,serif">http://www.ed.ac.uk/edinburgh-imaging
</span></a><br>
<br>
</div>
</div>
</div>
<br>
<br>
<div style="color: rgb(0, 0, 0);">
<div>
<hr tabindex="-1" style="display:inline-block; width:98%">
<div id="x_divRplyFwdMsg" dir="ltr"><font style="font-size:11pt" face="Calibri, sans-serif" color="#000000"><b>From:</b> Tarik S Bel-Bahar <tarikbelbahar@gmail.com><br>
<b>Sent:</b> 24 August 2016 22:16<br>
<b>To:</b> Jenny Bress<br>
<b>Cc:</b> eeglablist<br>
<b>Subject:</b> Re: [Eeglablist] Removing bad channels before average referencing - how stringent should I be?</font>
<div> </div>
</div>
</div>
<font size="2"><span style="font-size:10pt;">
<div class="PlainText">Hello Jenny, some tools, notes, and articles below, Best wishes.<br>
<br>
****************************************************<br>
Try/review/check plugins/documentation/settings/articles on the<br>
following tools.<br>
Each is googlable and has online or within-function documentation.<br>
<br>
*Within eeglab:<br>
clean_rawdata including it's bad channel detection function,<br>
trimOutlier<br>
PREP plugin that uses ASR andCleanline (see article on this),<br>
Cleanline and ASR plugins can also be used independently.<br>
Also see FASTER plugin and SASICA plugins/articles<br>
<br>
*outside eeglab: for example,<br>
See TAPEEG, standalone and matlab<br>
See SCADS method from Jung et al., implemented in Cartool<br>
See ERPLAB too which should have some form of principled bad channel detection<br>
<br>
<br>
<br>
<br>
*****************************extra notes for J<br>
Those are reasonable questions in your case, for which there are no<br>
established and robust standards across the field. Dense EEG has been<br>
around for nearly 20 years, so the existing base of published methods<br>
and "usual EEG laboratory methods" should be your first resource. Of<br>
course, researchers usually use their own preferred methods, the<br>
settings for which they only sometimes publish. Further, default<br>
plugins/settings don't always work for finding the bad channels.<br>
You're probably best off using some combination of visual detection by<br>
expert, relying on thresholds/algorithms usually used for bad-channel<br>
detection for your eeg system, reviewing published methods in recent<br>
high-quality articles using an eeg protocol similar to yours, and<br>
extra checking and reiterations on your end. The automated software<br>
for artifact-detection has been getting better, so it will also be<br>
helpful to examine the output of some automatic methods in eeglab and<br>
other tools (see tools above and some of the articles below). Last,<br>
ICA should be very useful for your purposes, as it can pick up<br>
spatially localized artifacts quite well.<br>
<br>
<br>
A good caveat is to not remove many contiguous channels, so you don't<br>
have a "large patch" with no channels. Another simple enough method<br>
is to use a human to visually detect the worst 5 to 10% of the<br>
channels. Checking the published methods/thresholds of some groups is<br>
useful, though stringency depends on the lab and researchers and can<br>
be variable. How many channels you can safely remove also depends on<br>
if you have 64, 128, or 256 channels.<br>
<br>
ICA is pretty good at picking up stereotyped artifacts in dense eeg,<br>
and often picks up ICs that reflect singel noisy channels or periods<br>
where single channels go bad briefly. Note that before ICA, it's<br>
better not to interpolate the bad channels.Also, ICA-cleaning often<br>
helps a lot with cleaning up channels that have continuous muscle or<br>
electrical noise. ICA-classification plugins, such as SASICA, can help<br>
you detect ICs that capture the activity of one bad channel that<br>
occurs during only some trials.<br>
<br>
<br>
****************************some example articles for J, each can be<br>
found through google scholar<br>
<br>
A practical guide to the selection of independent components of the<br>
electroencephalogram for artifact correctionJournal of neuroscience<br>
methods, 2015 - Elsevier<br>
Automatic artifacts and arousals detection in whole-night sleep EEG<br>
recordings Journal of neuroscience …, 2016 - Elsevier<br>
The PREP pipeline: standardized preprocessing for large-scale<br>
EEGanalysis Frontiers in …, 2015 - ncbi.nlm.nih.gov<br>
Reliability of fully automated versus visually controlled pre-and<br>
post-processing of resting-state EEGClinical …, 2015 - Elsevier<br>
Comparing the Performance of Popular MEG/EEG Artifact Correction<br>
Methods in an Evoked-Response StudyComputational …, 2016 - dl.acm.org<br>
Automated rejection and repair of bad trials in MEG/EEG6th<br>
International …, 2016 - hal.archives-ouvertes.fr<br>
Hybrid wavelet and EMD/ICA approach for artifact suppression in<br>
pervasive EEGJournal of neuroscience methods, 2016 - Elsevier<br>
<br>
</div>
</span></font></div>
</div>
</body>
</html>