<div dir="ltr">Hi Tatu,<div><br></div><div>let me respond to you inline,<br><div class="gmail_extra"><br><div class="gmail_quote">On Thu, Dec 3, 2015 at 4:56 PM, Huovilainen, Tatu M <span dir="ltr"><<a href="mailto:tatu.huovilainen@helsinki.fi" target="_blank">tatu.huovilainen@helsinki.fi</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">
<div dir="ltr">
<div style="font-size:12pt;color:rgb(0,0,0);font-family:Calibri,Arial,Helvetica,sans-serif;background-color:rgb(255,255,255)">
<p><span></span>Hi all,<br>
<br>
I'm working with combined M/EEG dataset measured with Neuromag and I'm wondering about the whitening step before ICA. Right now the MEG part of the data is tSSS'd and movement corrected (with cHPI to 'default head position') so the rank seems to end up being
around 70. </p></div></div></blockquote><div>That's ok just use <= this value for n_components.</div><div><br></div><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div style="font-size:12pt;color:rgb(0,0,0);font-family:Calibri,Arial,Helvetica,sans-serif;background-color:rgb(255,255,255)"><p>Engemann and Gramfort (2015, below) point out that with combined M/EEG a FA model should be used for the estimation as the noise levels between sensor types are heteroscedastic, </p></div></div></blockquote><div>we went beyond that, the 'shrunk' estimator that you have as an option in MNE uses different regularizations for the sensor types. And the idea is still that you cover different potential scenarios by</div><div>picking the best covariance estimator as measured by the negative loglikelihood on unseen data.</div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div style="font-size:12pt;color:rgb(0,0,0);font-family:Calibri,Arial,Helvetica,sans-serif;background-color:rgb(255,255,255)"><p>but also recommend not to use FA model after the dimensionality has
been reduced. How do you recommend I find the noise covariance matrix in my case?</p></div></div></blockquote><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div style="font-size:12pt;color:rgb(0,0,0);font-family:Calibri,Arial,Helvetica,sans-serif;background-color:rgb(255,255,255)"><p>
Engemann and Gramfort recommend computing the FA model before the SSS and then applying dimensionality reducing operators to both the data and the covariance estimator. How would this work?<span style="font-family:arial,sans-serif;font-size:small;color:rgb(34,34,34)"> </span></p></div></div></blockquote><div>you can try FA it's mostly a numerical problem, I think we have improved it up to a point where it can work even on SSSed data. </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div style="font-size:12pt;color:rgb(0,0,0);font-family:Calibri,Arial,Helvetica,sans-serif;background-color:rgb(255,255,255)"><p>
I'm not sure about tSSS, but I have to use at least the movement correction as I'm aiming for ICA decomposition and further analyses in the IC domain. I also have combined M/EEG "empty room" measurement, so a participant not doing anything for a few minutes.</p></div></div></blockquote><div>On event-related data with a noise covariance from uninteresting data I usually first apply ICA to make sure the rank reduction is consistent. But we meanwhile improved our down-stream code that handles the numerical rank in the source-localization, it will probably be just fine.</div><div><br></div><div>Hope that helps,</div><div>Denis</div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr"><div style="background-color:rgb(255,255,255)"><p><font color="#000000" face="Calibri, Arial, Helvetica, sans-serif"><span style="font-size:12pt">
Regards,</span></font><br><font color="#000000" face="Calibri, Arial, Helvetica, sans-serif"><span style="font-size:12pt">
Tatu</span></font><br>
<br><font color="#000000" face="Calibri, Arial, Helvetica, sans-serif"><span style="font-size:12pt">
Engemann, D. A., & Gramfort, A. (2015). Automated model selection in covariance estimation and spatial whitening of MEG and EEG signals. NeuroImage, 108, 328-342.</span><span style="font-size:12pt"></span></font><br>
</p>
</div>
</div>
<br>_______________________________________________<br>
Mne_analysis mailing list<br>
<a href="mailto:Mne_analysis@nmr.mgh.harvard.edu">Mne_analysis@nmr.mgh.harvard.edu</a><br>
<a href="https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis" rel="noreferrer" target="_blank">https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis</a><br>
<br>
<br>
The information in this e-mail is intended only for the person to whom it is<br>
addressed. If you believe this e-mail was sent to you in error and the e-mail<br>
contains patient information, please contact the Partners Compliance HelpLine at<br>
<a href="http://www.partners.org/complianceline" rel="noreferrer" target="_blank">http://www.partners.org/complianceline</a> . If the e-mail was sent to you in error<br>
but does not contain patient information, please contact the sender and properly<br>
dispose of the e-mail.<br>
<br></blockquote></div><br></div></div></div>