In order to compute intra-cortical coherence during a task I am considering the following methods:
- Performing surface Laplacian prior to computing intra cortical coherence by subtracting task based coherence.
- Performing intra cortical coherence by subtracting task based coherence.
- performing multivariate orthogonalization then computing intra cortical coherence by subtracting task based coherence.
If i perform surface Laplacian prior to computing coherence, should I be aware of any phase changes?
Hi Priya. The Laplacian will definitely change the phase values, more at some electrodes (where the signal is dominated by distant sources) and less at other electrodes (where the signal is dominated by local sources). It’s not really possible to predict a priori how the phases will change at which electrodes. So then it’s a matter of how to interpret those differences. In my opinion, it’s an advantageous difference, because volume-conducted effects are attenuated.
Thank you Mike. So the phase difference will affect my coherence values. Would computing imaginary coherence instead of MSC post Laplacian eliminate this problem?
I wouldn’t call it a problem. It’s a difference. When you filter the data, the data will change. Analogously, phase values change when you apply a temporal filter. I wouldn’t call that a problem either.
In general, volume conduction is a significant confound for EEG. There are two kinds of approaches for dealing with volume conduction: (1) spatial filters, (2) volume-conduction-robust analyses. Laplacian is in the first category and imaginary coherence is in the second. Neither is a perfect approach, though.
If you are concerned about this, then my general advice for any kind of data analysis is to try multiple analysis approaches and trust the patterns of findings that are qualitatively consistent across the approaches.
Thank you for that great explanation. I will apply multiple approaches to identify patterns that are consistent. Would it be possible to use spatial filters followed by imaginary coherence or spatial filter followed by task based coherence? Does it make sense to use multiple techniques that minimize volume conduction confound?
The Laplacian is a good spatial filter anyway; it’s not only for coherence analyses. So it’s not the situation that you can or should only use one approach.