0
$\begingroup$

My colleague and I started exploring EEG processing and, especially, functional connectivity. We decided to try the Phase Slope Index (PSI) and conducted a simple experiment as follows: it consists of three steps, research participant should raise his left hand (1), then his right one (2), then both (3). The description is simplified a bit but shows what the experiment was. We expected to see the information flow between the pre-motor cortex and primary motor cortex but we seemed to fail. The data was pre-processed (filtered at [0.25; 40] Hz, re-referenced to the average electrode, segmented into three pieces according to three stages of experiment; also blink artefacts were removed, so were motion ones). Then we computed the all-to-all PSI matrix and what we have had is depicted on the pictures (if needed and more appropriate, I could add the graph representation). All in all, we haven't seemed to reach what we aimed and now we're trying to figure out whether the fail refers to the interpretation, or experiment design, or both, or something else. My questions are:

  1. Can one guess what stage of experiment every matrix refers to? If so, how? I mean, do they at least look like the result of the experiment described? If they do, how can one interpret them?
  2. What could be an experiment that would show the noticeable PSI for two areas given beforehand? I mean, we had two areas given, pre-motor cortex and primary motor cortex, and we supposed that we would find a synchronization between them in such a motion experiment; something's gone wrong, though. If the problem is in the design of experiment, what design would be better?
$\endgroup$
  • $\begingroup$ I doubt that you will find anyone here who can tell you how your matrices should look like, neither help you interpreting them. These questions address physiology more than signal processing. I highly suggest you to find a paper which applied this method and try to reproduce their experiment and their results (I would look at something between hemisphere like both M1 especially with EEG). And finally, why do you think that areas so close to each other should be synchronized. Why should the brain be so redundant at that distance? Check your hypothesis and understanding of the brain/behavior. $\endgroup$ – Irreducible Feb 21 '18 at 13:01
  • $\begingroup$ Can I please ask if this was resolved? $\endgroup$ – A_A Feb 27 '18 at 10:59
0
$\begingroup$
  1. Can one guess what stage of experiment every matrix refers to? If so, how? I mean, do they at least look like the result of the experiment described? If they do, how can one interpret them?

I do not think that I can. But here is a guess that might give you some insight: The most "populated" matrices are those for a single hand up in the air (whether left or right).

Why?

Functional connectivity metrics (of which there are so many by now) assess some of "dependence" or "correlation". There is a higher chance for any two pairs of waveforms to have more correlated signals when the diversity of the generated signals is low. With two hands being raised, the brain generates a large set of impulses to drive all the necessary muscles. Consequently, the chance of them having a high correlation is low.

If you notice, a key characteristic of functional connectivity estimation in related papers is some form of averaging, whether temporal, spatial or across and within subjects.

Whether it is Magnetoencephalography (higher spatial resolution than Electroencephalography), or of course, fMRI or EEG, there is an intermediate stage that "summarises" observations.

In fMRI you sum voxel traces around anatomical areas indicated by an atlas, in MEG you sum contributions of elements around broad areas (simplest method) OR you do some form of source localisation and then look at the connectivity of the localised sources. Similar workflow applies in EEG too (e.g. LORETTA as above).

This is a required step to increase the Signal to Noise Ratio (SNR) and look at (hopefully) robust correlations. Along this idea are also methods that examine functional connectivity in specific bands of brain activity

Therefore, I doubt that you will get anything meaningful by looking at each individual channel without some form of spatial or temporal filtering / averaging.

This however, is half of the story. The other thing you can do is, of course, to change the task.

  1. What could be an experiment that would show the noticeable PSI for two areas given beforehand? I mean, we had two areas given, pre-motor cortex and primary motor cortex, and we supposed that we would find a synchronization between them in such a motion experiment; something's gone wrong, though. If the problem is in the design of experiment, what design would be better?

You will definitely see differences in the default mode network between two radically different tasks. For example, at rest and at "work".

To reduce your post-processing workload, try to use tasks that are not likely to generate a lot of artifacts. For example: Resting state and "solve a mathematical problem".

This design will give you something to compare "activity" against and help in reducing spurious correlations further.

Finally, be aware of filtering. The more your filter a waveform, the more it "looks like the filter" and therefore, the more chances there are to have more increased correlations.

Hope this helps.

$\endgroup$
  • $\begingroup$ I disagree that moving both hands/arms will generate a lot of pulses, the brain will sent more likely one impulse, that is processed on the spinal cord level, that activates a certain synergy of muscle activities (there are some groups doing locomotion/eeg/connectivity, a lot of muscles and no problems).A bigger problem are the artefacts as you mentioned. However, to reduce that I would suggest a design involving finger movements, as you have a better task were you can control what the subject is doing (or not) and correlated it with your physiology, or do some cortico muscular connectivity... $\endgroup$ – Irreducible Feb 21 '18 at 14:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.