Hello there, I am looking for suggestions on how to better address the source estimation in my dataset.
Briefly, I have collected data from 30 participants in three conditions: resting, passive viewing (participants were only asked to watch a 10-min video) and "active viewing" (participant show another 10-min excerpt of the same movie, and had to press a button whenever something appeared). I used an High-Density EEG (256 channels) and I have collected the structural MRI of each participant.
I preprocessed the data and created a single epoch from each block (each 10 min long), to analyze the differences occurring in time between the passive and the active one. Before doing so, I wanted to create the source model but I am not sure about the best strategy to apply for the noise covariance.
I read in the tutorials that the best options for my type of data may be
A. to calculate noise covariance from Resting
B. to use the identity matrix
C. to calculate the noise within each epoch, and then use only the diagonal part.
Option A seems not the best to me, as Resting was a different recording, with different neural activity and different preprocessing. Which solution would you recommend?
Any suggestion would be appreciated
Were the resting recordings performed on the same day, immediately before or after recording the viewing conditions, without removing the EEG cap in-between?
If this is the case, your best choice is probably to use a segment of resting recordings (eyes open, or at least with not too much alpha) to compute the noise covariance matrix.
The resting recordings must be preprocessed exactly in the same way as the viewing recordings. Note that for ICA cleaning, it requires to process the files with Process2.
If you need to pre-process differently the resting recordings for studying different effects, this is not a problem to have them twice in your database.
If you think there might be too much overlap between activity of the brain at rest and the active conditions, then use only the diagonal values (as explained in the tutorial: https://neuroimage.usc.edu/brainstorm/Tutorials/NoiseCovariance#Variations_on_how_to_estimate_sample_noise_covariance)
If you are not sure of this solution: process completely one pilot study with the two options A and B, and compare the results.
Dear Francois,
thank you for your suggestion.
Yes, the resting recording (eyes open) were performed right before the other blocks.
I think I will try both indications, thank you again for the suggestions!
P.S. I think I will use sLoreta for source estimation, does this change anything regarding the method?
Hi Francois, I am using the resting-state recording to compute the noise covariance matrix as suggested. I still have a doubt that I wanted to share with you and the community.
My resting-state recordings are 5 minutes long. Despite this, some of them have some epochs marked as BAD. In the case the BAD epochs are present, let's say, at 90 seconds and at 180 seconds, how should I calculate the noise covariance matrix? From the whole recording (leaving the epochs marked as bad) or from the longest time-window (in this example, between 180 and 300 seconds)? Thank you!