I am performing functional connectivity analysis on resting state MEG data. I have performed source reconstruction using both MNI/Dipole model sLORETA and then reduced my data to the Desikan atlas. On calculating the AEC(N*N) connectivity, I get high correlation values for delta band. I am given to understand that the maximum connectivity should be found in the alpha-beta range, so I think the high connectivity in the delta range is some sort of artifact. What do you think I'm doing wrong here?
P.S: I check the orthogonalization box during AEC calculation.
If the time series are too short, the low-frequency AEC yields artificially high values. Make sure your epochs are at least 10 times longer than the longest cycle (1/slowest frequency) tested. Also, note that most AEC studies actually low-pass filter the Hilbert enveloppes very aggressively before computing AEC. See papers from M Brookes' group for instance. Otherwise the AEC scores in alpha-beta and higher are just too low. For this to be valid, you also need to look at very long epochs of data.
Thanks for your reply. I use resting state recordings of an average duration of more than 5 minutes, so I end up with enough number of cycles, not just for delta (2-4Hz) but also for any further low pass filtering(0.2 Hz in Deco et.al.)
Even though I haven't low passed the envelopes yet, I obtain decent correlations in the alpha-beta band. The problem is that the delta band has the highest correlations of all bands. Do you think that will change upon low passing the envelopes?
Since I am orthogonalizing the signal during the AEC step, spatial leakage should not be the cause of spurious correlations. Just out of curiosity, is this the same algorithm as the one proposed by Brookes?
I can share my entire brainstorm pipeline if that helps.
Another confounding factor are eye blinks and movements artifacts that can inflate low-frequency AEC. I suggest you replicate the approach from Brookes in terms with same analytical parameters, data length etc.
@Francois, @MartinC: this is something I have mentioned before - allowing filtering of amplitude envelopes from Hilbert or directly adding a filtering option to the AEC process. Can you looking into this, please?
Thanks.
I wanted to use MEG-ROI-nets pipeline(https://github.com/OHBA-analysis/MEG-ROI-nets) in order to obtain AECs. For this I require the source reconstructed files(already performed in Brainstorm using Desikan Killainy atlas) and "A set of ROIs or a spatial basis set, in the same space and resolution as the MEG data, saved as a nifti" . How do I obtain the latter using brainstrom ?
Can you be more specific of what needs to be implemented? Or maybe it would be easier if you discuss directly this with Martin.
"A set of ROIs or a spatial basis set, in the same space and resolution as the MEG data, saved as a nifti" . How do I obtain the latter using brainstrom ?
This is not something you would get from Brainstorm, but from FreeSurfer or any other T1 segmentation pipeline: FsTutorial/AnatomicalROI - Free Surfer Wiki
(the volume version aparc+aseg.mgz of the Deskian/Killiany atlas)