mne_compute_raw_inverse gave perfectly correlated source current estimates

✉️ Mailing List Archive (read-only)
1

Hello,

I am trying to use mne_compute_raw_inverse to project single trial
epochs onto the surface. The command finishes without erros, but is
generating perfectly correlated source estimates from distant regions.
The scale and sign of current estimates differ, but the correlation is
either 1 or -1.
Below is the command:

#do forward solution
mne_do_forward_solution --overwrite \
--meas ${s}_prep_ave.fif --subject ${s} --fwd ${s}_prep_ave_fwd.fif --meg --eeg
    
#create inverse operator
mne_do_inverse_operator --fwd ${s}_prep_ave_fwd.fif --depth \
--loose 0.2 --meg --eeg --senscov ${s}_prep_cov.fif --subject ${s}

#project raw data, for each label
mne_compute_raw_inverse --in ${s}_${rn}_sss_ds_raw.fif \
--inv ${s}_prep_ave_fwd.fif-meg-eeg-inv.fif \
--picknormalcomp \
--align_z \
--labeldir ./ALL-Labels \
--orignames \
--out ${s}-{rn}-label-source

However, the same inverse operator worked fine with averaged evoked
data, results are reasonable when viewed in mne_analyze.
Any idea what could have gone wrong?

Thanks!
Kai

2

Hello Kai,

a possible explanation is the "nave" used to set the regularization of
the inverse problem.

if you work with raw data, nave is 1 while with 100 epochs averaged
it's 100 and the inverse
solution is then far less regularized. A solution is to set the --nave
option to a higher value in
mne_compute_raw_inverse but you might amplify noise.

hope this helps

Alex

3

Hi Alex, I changed --nave to 100 and still got perfect correlations.
Actually source estimates look identical between --nave 1 and --nave
100.
Any other suggestions?
Thanks
Kai

4

hi,

did you check that your labels are correct?

Alex

5

Hi Matti and Alex,
Yes MEG-only solution worked. Now source estimates are no longer
perfectly correlated. Any idea what went wrong regarding including EEG
data into the inverse solution?

I used labels created by freesurfer's a2009 parcelation, I double
checked and they are correct.

Thanks!
Kai

6

Hi Matti and Alex,
Yes MEG-only solution worked. Now source estimates are no longer
perfectly correlated. Any idea what went wrong regarding including EEG
data into the inverse solution?

I used labels created by freesurfer's a2009 parcelation, I double
checked and they are correct.

Are some of the EEG channels dysfunctional and not marked bad? Have you checked the EEG data by switching the average EEG reference off in mne_browse_raw?

- Matti

7

Hi Matti,
Bad channels were previously marked using mne_mark_bad_channels. I went through the raw data again by turning off EEG reference projection, and added a couple more to the bad list. I also added "eegReject 40e-6" to the covariance file. That worked, now source estimates are no longer perfectly correlated.
Thanks!
Kai