Glad to greet you again.
I have performed time-frequency analysis using Brainstorm from data simultaneously recorded from the subthalamic nucleus (STN) and some scalp EEG electrodes during a cognitive task performance.
I wonder in your opinion what would be the best type of analysis to detect functional connectivity or coupling between the time-frequency results from the STN and scalp EEG activity.
Thank you in advance,
Sounds like an amazing dataset!
The choice of connectivity measure depends on your scientific question.
Technically, you could consider assessing connectivity b/w the STN electrode and scalp sensors. If you have 32 electrodes or more, you could consider modeling the sources of scalp of EEG on the cortex and assess cortico-STN connectivity with greater anatomical specificity.
Thanks Sylvian, Francois:
We simultaneously captured LFP recordings from the STN and scalp EEG recordings. LFP were collected by electrode extension cables some days after surgery. However, due to the wounds on the scalp produced by the neurosurgical intervention, scalp EEG data were only recorded from electrodes placed over the frontal and occipital areas (approximately 10 electrodes each). So, I think it is not possible modeling the sources of scalp of EEG on the cortex and assess cortico-STN connectivity with greater anatomical specificity.
We have already obtained using Brainstorm time-frequency statistical results for each experimental condition of our cognitive task, for both STN and scalp EEG electrodes. For example, comparing condition A and B, we have obtained a statistical significant increase of beta band power (15-25 Hz) around 700-900 ms in the LFP STN, and a statistical significant increase of theta or alpha band power earlier (300-500 ms) or later (1000-1200 ms) over frontal and occipital scalp electrodes. In this respect, we have some doubts what would be the best or most useful procedure or connectivity approach to understand if these different LFP/EEG frequency bands in the STN or surface EEG are in any way “connected”.
So it looks like you are after a possible effect of cross-frequency coupling.
You could run a phase transfer entropy estimation is these frequency bands b/w the each STN leads and scalp electrodes. It would provide an assessment of possible directionality b/w regions within each band. The other thing you could use but which would require customization of BST's code is phase amplitude coupling (PAC) b/w STN beta's amplitude possibly modulated by the phase of slower cortical oscillations.
In relation to the two doubts that you have provided me:
- Due to different methodological issues, and as far as we understand this approach, I think this approach, a phase transfer entropy estimation it would not be the most appropriate method at this stage of our study.
- In this respect, we decided to collapse/average all STN bipolar contacts (STN channel) and the surface EEG electrodes both over frontal and occipital areas, independently. Therefore, we have a STN channel, a frontal channel, and an occipital channel. After that, as I mentioned, we performed time-frequency analysis and we obtained different statistical (cluster-based) results by comparing our A and B conditions. Thus, we seek to explore if the significant increase in alpha power range observed on frontal regions between 200-1000 ms is, in any way, associated or “connected” to the significant increase in beta power range on the STN between 800-1000 ms.
- Therefore, I think we should perform a phase amplitude coupling (PAC) analysis. We believe (please tell me if we would be wrong in any way) this approach will allow us, by measuring of cross frequency phase amplitude coupling, to explore if low frequency oscillations on frontal regions modulate the local STN activity of higher frequency oscillation in this particular experimental condition.
- Considering the above, do you have any recommendation before performing tPAC? What exactly do you mean by customization of BST's code?
Thanks in advance,
I you are looking at possible interactions between the respective power of alpha and beta, then this would not be a measure of PAC, but rather of amplitude/amplitude coupling.
You can extract the respective (Hilbert) envelopes of alpha and beta activity in each electrode cluster and then compute the correlation b/w the resulting time series.
To minimize the problem in which each subject has a different number of electrodes for each regions of interest (STN, frontal, and occipital) as a consequence of the of surgical intervention, we separately collapse each region of interest in a different electrode cluster, resulting in a single electrode cluster for each region (i.e., STN, frontal, occipital) but with a different number of trials each. In this respect, would there be a problem to compute the suggested connectivity measure if, for each subject, the number of trials is not the same for the different electrode cluster and a same condition?
Regarding your recommendation to test amplitude/amplitude coupling, we tried to run the process Connectivity > HCoh NxN, however, we did not find it explicitly. We assume that it now has another name on Brainstorm, so we run the process Connectivity > Envelope correlation NxN (2020). We have noticed that this option allows to set different options by setting time-windows of interest, choosing between different time-frequency transformation methods and connectivity measures. In our view, this option calculates power envelope of the time-series and then calculate correlation coefficients across these envelopes.
Nevertheless, we have doubt about how, with our current dataset, we should put our different data in the process1 or process2 tabs to extract the respective (Hilbert) envelopes of alpha and beta activity (or any other frequency of interest) in each electrode cluster and then compute the correlation b/w the resulting time series.
Should we then put in the process1 and compute separately for each subject, for example, the individual trials dataset, for experimental condition A, and for the three different electrode clusters (LFP, frontal, occipital), and then to perform this for each subject and condition? How could we them average after?
Thank you, we really appreciate your help,
Adding @hossein27en to the discussion.
It is preferable to use all your data available, even if this means different number of trials across subjects. The variance might be high across the resulting sample of connectivity values but this is accounted for in subsequent statistical testing.
From what I understand about your plan, you first need to extract the Hilbert envelopes of the data in the frequency bands of interest. This is done using the process1 box. Then you'll need to compute the correlation between the resulting envelopes at two different frequencies. To do this, you'll need to drop them into either box of the process2 panel.
I hope it makes sense and it works for you.
I'm not sure I understand the pipeline you describe here.
The connectivity functions (e.g. correlation) are not equipped to handle time-frequency files (e.g. Hilbert process results) in input. This is not a easy feature to add.
All the connectivity measures that use the Hilbert transform do it internally, see examples below:
- Amplitude Envelope Correlation: https://github.com/brainstorm-tools/brainstorm3/blob/master/toolbox/connectivity/bst_connectivity.m#L437
- Phase Locking Value: https://github.com/brainstorm-tools/brainstorm3/blob/master/toolbox/connectivity/bst_connectivity.m#L497
- PLV (time): https://github.com/brainstorm-tools/brainstorm3/blob/master/toolbox/connectivity/bst_connectivity.m#L539
- Phase Transfer Entropy: https://github.com/brainstorm-tools/brainstorm3/blob/master/external/fraschini/PhaseTE_MF.m#L67