Hi Raymundo,
I am wondering what exactly the noise covariance computation is used for in EEG.
Should noise covariance be computed for each recording within a time point (Pre, Post, Follow up), or should one noise covariance matrix should be used for each recording within a time point? We have four recordings per time point.
I've been computing noise covariance for each recording within a time point (making four matrices per time point) while using one head model and channel file per time point, applying these to all recording within a time point.
Please let me know what you recommend. Thank you.
Best,
Adam
The computation of the noise covariance has a goal to obtain an accurate model of the sensors (EEG, MEG) noise which is used to improve the performance of sources estimation methods such as minium-norm estimation.
Since empty-room recordings are not possible for EEG, there are different options to compute it. You may want to check this link:
https://neuroimage.usc.edu/brainstorm/Tutorials/NoiseCovariance#Variations_on_how_to_estimate_sample_noise_covariance
Once the registration of the electrodes and the anatomy is done, it makes sense to copy the channel file from the registered recording to the other three recordings in the same session (time point). Of course, if the cap was not moved.
A copy of the channel file is recommended as it could be possible that pre-processing was different for the different recordings in the same session (e.g., different ICA-derived components were removed for different recordings, a channel was bad for only on for different recordings, etc).
For the head model it could be copied for recordings in the same session if and only if:
- The electrode types did not change (they have the channels as EEG type, which should be your case)
- The positions of the electrodes in the channel file did not change. This is true if the channel files were copied after registering the sensors.
Raymundo,
Yes, it seems we are following the correct guidelines for channel files and head models; however, we're wondering about noise covariance and whether these should be copied to the other recordings within a session or not after copying channel file and head model to other recordings within a session.
I think that it may not be appropriate to copy the noise covariance matrix between recordings because noise may increase within a session between recordings (such as due to the gel drying over time, like the tutorial says). Therefore, this noise would only be accounted for if we computed noise covariance at an individual recording basis.
If this is the case, then I believe computing noise covariance on a per-recording basis is acceptable; and given that the electrodes are not explicitly altered within a session, it also seems acceptable that the channel files and head models may be copied to other recordings within a session.
Do you agree with this logic?
TLDR: Is it okay to localize electrodes to scalp with channel file from recording A (rA), create head model for rA from this channel file, copy rA head model & channel file to recordings B-D (all are of same session), compute one noise covariance matrix per recording for recordings A-D (4 noise matrices total). Then compute sources for recordings A-D per recording.
Thanks again,
Adam