Is connectivity more accurate with more epochs?

💬 Support & Discussions
,
1
  • MNE-Python version: 0.22.1
  • operating system: Windows 10

I’m calculating connectivity (using the coherence measurement) in the sensor space for 5 minutes of continuous meditation data. I am okay with epoching the data, and have done this to make the function (mne.connectivity.spectral_connectivity) work.

Am working on finding an ideal window for epochs, and have settled on 50s for the time being. I’m wondering, though, would finer-grain epoching make my connectivity measures more accurate since there are more chunks of data to statistically compare? I don’t know too much about this measure, but am hoping to learn as much as I can to incorporate it into our lab’s analysis.

Thank you!

-nv

2

Hi Neurovella,

the short answer is “yes”. In fact, too little trials can overestimate your actual connectivity. However, if the epochs become too short, you can lose frequency resolution.

Let me suggest the following paper for you to learn more about connectivity measures:
Bastos, A.M., and Schoffelen, J.-M. (2016). A Tutorial Review of Functional Connectivity Analysis Methods and Their Interpretational Pitfalls. Front. Syst. Neurosci. 9 .
DOI: https://doi.org/10.3389/fnsys.2015.00175

which even features a figure that looks at the interaction between sample size and connectivity measures (Figure 10).
This paper will also focus on a topic I want to briefly caution about here as well: many connectivity measures are flawed in sensor space because of volume conduction, so be aware of that!

Hope this helps!
Britta

1 Like
3

Britta,

Thank you, this is very helpful! The paper you sent over is an informative one, especially considering the methods that we’re using. I didn’t think about volume conduction, so I appreciate the warning. Hopefully we will soon implement some strong source signal separation techniques in the future!

Best,
-nv

4

Hi @britta-wstnr,

IIUC, @Neurovella aim to investigate the connectivity over the 5 minutes time. The 5min signal was bin into 50s of 6 epochs. The spectral_connectivity output for each of the 6 epochs then will be averaged.

But, what is the difference, from physiological point of view, if, say, one to calculate the connectivity directly over the 5minute span. Such that, no epoching was conducted.

Also, may I know whether it is common practice to epoch a continuous signal when one investigate the connectivity during resting state. At least in the OP case, epoching was conducted in order to respect the spectral_connectivity requirement

1 Like