dWPLI and its interpretation

Hi Mike,

Hope you are fine under this long-lasting pandemic.

I encountered some problems about phase-based connectivity, specifically dWPLI. Although I have gone through related teachings in your ANTS book and videos on Udemy, I still could not confirm some points when back to my data. Here are problems:

(1) How to understand negative values of dWPLI? Before baseline correction, I found that raw dWPLI values calculated from my data involved negative values. Because I focus on prestimulus session, I am not going to conduct baseline correction. Then are there any meanings for those negative values? How to deal with them?

(2) If I analyzed dWPLI, do I need to consider the confound of volume conduction? In my data, I found that in the subject level, when the dWPLI was low, the power in that time-frequency window was also low. It seemed to have a positive correlation. According to descriptions in Chapter 25.10 in ANTS book, that positive correlation may imply volume conduction. However, according to what you said in the 171st video on Udemy, phase lag index has no confounder of volume conduction. Back to my case, I first filtered data using surface Laplacian and then used insensitive measure, i.e. dWPLI. So even though there were positive correlations between dWPLI and power, I suppose it very probably reflected the true connectivity. Am I correct?

(3) How to interpret the results of phase-based connectivity? There were many reviews or studies related to it now. As a beginner, I still could not grasp the “boundary” of its interpretation. In other words, how to avoid over-interpretation or what is the safe area in which phase-based connectivity could be interpreted?

Looking forward to your reply.

Hi Jinwen.

  1. The debiasing term (“d” in dwPLI) involves subtracting off a small quantity. So if the wPLI is already close to zero, then the dwPLI can be negative. For this reason, you cannot interpret the dwPLI in absolute terms between 0 and 1 the way you would interpret PLI or phase clustering. So you also don’t need to do anything in particular with them.

  2. Nope, the whole idea of phase-lag-based measures is that they are blind to volume conduction. But when power is low, then SNR is potentially also low. A positive correlation between power and connectivity doesn’t necessarily indicate a volume conduction artifact; it is simply one indicator of the possibility. So if you’re using the PLI with Laplacian, then you can sleep soundly knowing that your findings are not due to volume conduction :wink:

  3. The most conservative interpretation is that connectivity is simply a feature of the data that you investigated. But I think a closer link to the brain is warranted. Something like evidence for large-scale coordination across cortical areas. Some people try to link the electrode-level findings to brain regions. I think that’s appropriate in some cases (e.g., writing “frontal-parietal network” if the connectivity is between Fz and POz) and inappropriately precise in other cases (e.g., writing “ACC-extrastriate”). The level of anatomical interpretation that is justified also depends on the theory and prior data (e.g., if the EEG study is a replication of an fMRI study that showed specific localization).

Thank you so much. Your answers are so clear and helpful. Thanks again!