Hi, guys.
I am recently using the MEG resting state data (collected on Elekta Neuromag Triux) to analyze the functional connectivity network.
I performed MaxFilter on the data. After preprocessing, epoching into 10s, downsampling from 1000 Hz to 50Hz. I use mne.minimum_norm.apply_inverse_epochs to get the source estimates. Using HCPMMP1 atlas, I extracted the time course for each ROI with mne.extract_label_time_course. Then with the help of mne_connectivity.spectral_connectivity_epochs, I calculated the coherence between each pair of ROIs in common frequency bands.
However, I plot the seed based results (eg, the coherence between vmPFC and other ROIs). I found that there’s no networks like DMN. More precisely, it seems that an ROI only has relatively good functional connections with the surrounding parts, while other parts look like the background.
Did I do something wrong?
More precisely, it seems that an ROI only has relatively good functional connections with the surrounding parts, while other parts look like the background.
This sounds like there could be a lot of mixing in your reconstructed sources. No inverse model is perfect, so source mixing/volume conduction artefacts which exist in sensor space can also persist following source reconstruction.
For this reason, it’s often recommended that you use connectivity metrics which are robust to these instantaneous interactions (e.g. imaginary part of coherency) even when working in source space. I would recommend to try a measure like this and see if the results look more like you expect.
A separate note: MNE v0.24.0 is over 3 years old. Just generally I would recommend upgrading to the latest version (v1.6.1; would also require you getting the standalone MNE-Connectivity package if you aren’t using this already).
Thanks!
I will try that. And my MNE version is now 1.6.0, I forgot to change the version number which is automatically generated. Anyway, thanks for your advice.
