[Mne_analysis] LCMV analysis in MNE-python and in Fieldtrip

Виктория Манюхина mvo96 at inbox.ru
Fri Jun 28 10:35:52 EDT 2019
Search archives:

        External Email - Use Caution        

Dear MNE-python experts,

We try to localize induced visual gamma activity using LCMV beamformers by first localizing single epochs and then doing multitaper analysis on ‘virtual sensors’.
To do this we filter our raw data ( 25 -115 Hz), compute noise covariance matrix [-1 - 0] and data covariance matrix [0 - 1.2] (method=’shrinkage’, rank=None) on epochs, apply ‘make_lcmv’ function (reg=0.05, pick_ori=’max-power’, weight_norm=’nai’, rank=None) and then ‘apply_lcmv_epochs’ function with parameter max_ori_out=’signed’. Next we compute power spectral density with multitaper method (psd_array_multitaper: bandwidth=5) and average the result over epochs.
The result looks reasonable: there is a clear increase in gamma amplitude during stimulation relative to the prestimulus baseline. However, the amplitude of this gamma increase is approximately 2 times lower than that obtained using the similar analysis in Fieldtrip/LCMV.
We tried different approaches to calculation of the noise covariance matrix in the MNE-python, but it does not make much difference. Actually, in the Fieldtrip the noise covariance is not used at all. They use the ‘lambda’ parameter (e.g. 5%). Could it be the main reason for the differences in the results?
We still would like to use MNE-python for the LCMV analysis. What do you think could help us to increase the signal amplitude?

Best,
Viktoria
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20190628/f5ca2f0c/attachment.html 


More information about the Mne_analysis mailing list