Analyzing MEG and EEG data from spontaneous
recordings of epileptic patients with BabySQUID1, I ran into 3 issues
with the dipole fit procedure.
1) Preprocessing - Noise covariance
estimate
From my understanding this matrix shall reflect the
interdependencies of different channels with respect to the recorded
noise / background activity. Is that correct? Based on this assumption I
would calculate the noise covariance matrix on a recording segment
without epileptic activity. Would that make sense? Would it be more
appropriate to derive the noise covariance matrix from the entire
recording?
2) EEG source localization
Which unit range does MNE expect
as EEG input? When I enable EEG data for source localization (current
density estimate, dipole fit), I receive unrealistically strong sources
in a very deep location. I assume this is due to a scaling issue.
3)
Bdip content
I would like to display the confidence ellipsoid to a
dipole fitted by MNE and saved in bdip format. From my understanding the
ellipsoid semi-axis parameters are stored in the error matrix. Reading
this into a 5x5 matrix in Matlab, I am not able to determine which
values correspond to the axis. Here an example of an error matrix:
E =
[-7.8358e-39 0 0 0 0; 0 1.9263e-38 0 3.3224e-31 7.2643e-42; 3.8519e-34
4.2782e+09 0 1.3004e-41 3.4159; 0 -1.1768e-19 2.2510e-41 1.4324e-16
-3.3895e+38; 3.8519e-34 4.2782e+09 0 1.4324e-16 -3.3895e+38]
I would
be very glad if you could provide any hints regarding one or more of
these problems.
Analyzing MEG and EEG data from spontaneous recordings of epileptic patients with BabySQUID1, I ran into 3 issues with the dipole fit procedure.
1) Preprocessing ? Noise covariance estimate
From my understanding this matrix shall reflect the interdependencies of different channels with respect to the recorded noise / background activity. Is that correct?
The noise covariance gives estimates of the variances of Gaussian noise of the measurement channels (on the diagonal) and the covariances between channels (in the off-diagonal elements).
Based on this assumption I would calculate the noise covariance matrix on a recording segment without epileptic activity. Would that make sense? Would it be more appropriate to derive the noise covariance matrix from the entire recording?
If the epileptic activity is infrequent, you can even use the entire recording. Be sure to include a long enough data segment to make the covariance matrix well behaved.
2) EEG source localization
Which unit range does MNE expect as EEG input? When I enable EEG data for source localization (current density estimate, dipole fit), I receive unrealistically strong sources in a very deep location. I assume this is due to a scaling issue.
The EEG data in the fif files should be in volts.
3) Bdip content
I would like to display the confidence ellipsoid to a dipole fitted by MNE and saved in bdip format. From my understanding the ellipsoid semi-axis parameters are stored in the error matrix. Reading this into a 5x5 matrix in Matlab, I am not able to determine which values correspond to the axis. Here an example of an error matrix:
E = [-7.8358e-39 0 0 0 0; 0 1.9263e-38 0 3.3224e-31 7.2643e-42; 3.8519e-34 4.2782e+09 0 1.3004e-41 3.4159; 0 -1.1768e-19 2.2510e-41 1.4324e-16 -3.3895e+38; 3.8519e-34 4.2782e+09 0 1.4324e-16 -3.3895e+38
MNE does not calculate the confidence intervals for dipoles. Therefore, the error matrix will contain invalid data. There is a field called "errors_computed" in the bdip structure, set to zero to indicate that the error estimates have not been computed.