Separating phase-locked and non-phase-locked power

[@munazaramzan603]

Love the way u teach...sir U r my inspiration....thanks a lot for being there... I have seen all of ur neural analysis videos.... Those became my life saving pathways....

[@mikexcohen1]

Thank you, Munaza. That's very kind of you to write.

[@carolinaogawa1332]

Some authors argue that computing non phase-locked power by subtracting the ERP before doing the time-frequency decomposition neglects the fact that the ERP latency/amplitude can jitter from trial to trial, leading to imprecise non phase-locked power. I would love to know your opinion about it, Mike :) Thank you!

[@mikexcohen1]

Hey Carolina. That's an important question without a clear answer. The thing is that the ERP is mathematically defined as the time- and phase-locked component of the single-trial EEG data. So jittered activity is not technically the ERP. I've tried a few times to fit the single-trial amplitude (thus assuming that the latency/waveform-shape is constant but amplitude varies per trial), and it hasn't changed the results in the datasets I looked into. I haven't tested single-trial latency jitter fitting, but at some point you need to worry about overfitting noise... Anyway, I take a very practical approach to this situation: The single-trial dynamics that can be quantified with time-frequency analyses are -- in my experience -- more statistically robust, reproducible, and correlated with behavior than the (single-trial or average) ERP. One example paper of this is Cohen and Donner, 2013.

[@carolinaogawa1332]

@@mikexcohen1 Very interesting, Mike! Thank you! :)

[@benjamin4190]

Thank you for your videos! I have a question. My data has a sinusoidal ERP and I want to know if there is also induced phase synchronization at the frequency of the ERP. If the phase-locked power and the non-phase locked power overlap -- does that mean that there was induced phase synchronization? What if they only slightly overlap? I would think that you can only infer something when the two powers aren't overlapping. I was trying to do ITC over the trials minus ERP, but you will get 0 coherence. Except if there is large amplitude ERP differences trial to trial you get a bimodal polar histogram. Hope you see this :)

[@mikexcohen1]

Yeah that's tough, because the ERP is the phase-locked activity but is not necessarily oscillatory (or can contain a combination of oscillations and non-oscillatory activity), but removing all the phase-locked signals will also remove any phase-locked oscillatory information. I'm not a huge fan of interpreting ERPs; I would focus on the phase-locked and non-phase-locked parts of the signal.

[@benjamin4190]

​@@mikexcohen1 Thank you for the reply! So would there be any way to use the phase-locked info to determine if the ERP is endogenous or an artifact from electrical stimulation? My hunch is leaning towards investigating the overlap between phase-locked and non phase-locked. Have you ever heard of anything on that? Easiest thing I can think of is just to see if there is still ITC after the stimulation period.

[@mikexcohen1]

Hmm, interesting thought. The artifact is obviously phase-locked, but some endogenous activity would also be phase-locked. If you have enough channels, you might be able to isolate a spatiotemporal filter for the artifact and project it out? I'd guess there some be some methods studies about this that might be helpful.

[@bokkieyeung504]

Hi Mike, thanks for the video. I have two questions about this course. 1) you construct non-phased-locked signal by substracting ERP from raw signal of a random single trial? Then if we have 99 trials, we will get 99 "versions" of non-phased-locked signal? 2) among your 5 example results, I'm confused about C) Phase-locked power and D) ERP power. I expect they are the same things. Just look back to the operationalizations: Phase-locked = Event-related potential, it seems to me they are the same things...

[@mikexcohen1]

Hi Bokkie. 1) It's probably more accurate to call them residual time series than versions, but yes, they're obtained by subtracting the ERP from each individual trial. 2) They're conceptually similar but defined in different ways. The phase-locked power in panel C comes from a subtraction of two nonlinear operations (power), whereas the ERP power comes from a nonlinear operation after a linear operation. In general, linear operations are the same regardless of order, whereas nonlinear operations change depending on the order.

[@tobybromfield3664]

Could you give an example of non phase-locked? I understand phase locked, as there's the same time and phase and so you can average it. So why would you ever do something that isn't phase-locked? I just don't quite get the induced activity

[@IsobelFrench]

Love your jokes :D

[@mikexcohen1]

Thank you, Isobel. Very kind of you to pretend to tolerate my jokes ;)

[@IsobelFrench]

Mike X Cohen 🤣🤣