It's been a few weeks, and I am still having trouble with this.
There seems to be an issue with the inverse operator from the nightly
build version of MNE. The activation shown on the brain looks unusal. I
am using the nightly build because in the previous versions working with
SSP for EEG affected the MEG channels and it was fixed in the nightly
build. Now, I seem to be running into some trouble when using an inverse
operator created from both MEG and EEG data, the MEG based inverse
operator works/looks fine. Anyone have any work-arounds or solutions?
I had a problem with the EEG projection vectors affecting MEG channels
as well. However, it only happened on my linux x86_64 install, not on
a linux i686 install or osx installs. As you note, this channel
corruption on x86_64 was fixed in a recent nightly. I assume, then,
that you're running the linux x86_64 nightly?
If so, one thing you could try is generating your inverses using all
four combinations of 1) the last stable release and 2) the latest
nightly combined with 1) the x86_64 linux version and 2) the i686
linux version. (The i686 linux version should run just fine on 64-bit
linux installs.) If you can identify which of the four combinations
suffers from this problem, then it could help narrow down the cause of
the problem.
We tried the stable release and the nightly build. The stable release
inverse operator seems to work fine but than the issue with the projection
vectors exists. With the nightly build, it's the opposite, the inverse
operator gives me the issue and there seems to be no problem with the EEG
and MEG projection vectors.
Hi Ricky,
One possibly useful test to see if the issue is related to machine
precision/round-off related bugs is to apply the inverse operator in
MATLAB at double precision using mne_ex_compute_inverse().
If that works fine, then it probably confirms a bug with mne_make_movie.
If that also looks weird, then it could potentially point to something
more upstream such as the SSP (that's used on the data) not being applied
to the lead field/noise covariance correctly etc..
Are the MEG data maxfiltered? Sometimes the inverse operator calculation with MEG + EEG does not work correctly if maxfilter is involved. Can you send the screen output of the inverse operator calculation?
mne_inverse_operator version 2.32 compiled at Jun 7 2011 04:07:51
Compute the MNE inverse operator decomposition
Forward solution :
averages_covariances/CPLEX2_09_merged_avg-7-fwd.fif
Noise covariance matrix :
averages_covariances/CPLEX2_09_merged_cov.fif
Source covariance matrix : identity matrix
Destination for the inverse operator data :
averages_covariances/CPLEX2_09_merged_avg-7-meg-eeg-inv.fif
Include MEG and EEG data.
1 bad channels from averages_covariances/CPLEX2_09_merged_cov.fif:
<MEG 1431>
Reading the forward solution....
Read data for 306 MEG channels and 5906 sources
Read data for 70 EEG channels and 5906 sources
Free source orientations.
The forward computation was performed in head coordinates.
1 bad channels read from the forward solution file.
305 MEG channels remained after excluding bad ones
No bad channels removed from EEG
Read 2 source spaces with a total of 5906 source locations
Source spaces are now in head coordinates.
Channel description list matched with the composite forward
solution matrix.
Global Cartesian head coordinate system forward matrix will be
employed.
Loaded projection from averages_covariances/CPLEX2_09_merged_cov.fif:
# 1 : planar-77--0.200-0.200-PCA-01 : 1 vecs : 204 chs MEG active
# 2 : planar-77--0.200-0.200-PCA-02 : 1 vecs : 204 chs MEG active
# 3 : planar-77--0.200-0.200-PCA-03 : 1 vecs : 204 chs MEG active
# 4 : axial-77--0.200-0.200-PCA-01 : 1 vecs : 101 chs MEG active
# 5 : axial-77--0.200-0.200-PCA-02 : 1 vecs : 101 chs MEG active
# 6 : axial-77--0.200-0.200-PCA-03 : 1 vecs : 101 chs MEG active
# 7 : eeg-77--0.200-0.200-PCA-01 : 1 vecs : 70 chs EEG active
# 8 : eeg-77--0.200-0.200-PCA-02 : 1 vecs : 70 chs EEG active
# 9 : eeg-77--0.200-0.200-PCA-03 : 1 vecs : 70 chs EEG active
# 10 : axial-PCA-01 : 1 vecs : 102 chs MEG active
# 11 : axial-PCA-02 : 1 vecs : 102 chs MEG active
# 12 : axial-PCA-03 : 1 vecs : 102 chs MEG active
# 13 : axial-PCA-04 : 1 vecs : 102 chs MEG active
# 14 : Average EEG reference : 1 vecs : 70 chs EEG active
Read a full noise covariance matrix from
averages_covariances/CPLEX2_09_merged_cov.fif
Picked appropriate channels from the sensor noise covariance matrix.
MEG/EEG correlations omitted.
No regularization applied to the noise-covariance matrix
Projection applied to the covariance matrix.
Decomposing the noise covariance...
Estimated covariance matrix rank = 361 (0.000178757)
293 MEG and 68 EEG-like channels in the whitened data
done.
Creating the source covariance matrix:
done
Applying linear projection to the forward solution...done
Whitening the forward solution...done
Scaling the source covariance...done
Decomposing...
Applying a priori source weighting to the forward solution...done
Transpose...done
SVD...done
largest singular value = 0.987327
scaling factor to adjust the trace = 19 (nchan = 375 nzero = 14)
done
Writing the solution to
averages_covariances/CPLEX2_09_merged_avg-7-meg-eeg-inv.fif...done
Attaching the environment to
averages_covariances/CPLEX2_09_merged_avg-7-meg-eeg-inv.fif...done
Inverse operator file
averages_covariances/CPLEX2_09_merged_avg-7-meg-eeg-inv.fif ready. Thank
you for waiting.