Dear MNE users,
A few weeks ago a strange problem in the analysis of EEG-only and MEG/
EEG data was brought to my attention. This problem manifested it in
the following ways:
1. The source estimates were "spotty" on the cortical surface and
changed rapidly as a function of time.
2. When you clicked on the surface to look at a source waveform it
clearly contained high temporal frequencies not present in the data.
I was finally track down the reason and, consequently, the
circumstances in which this happens. As one might expect, this is some
kind of an error magnification in the computation of the source
estimates. This is, indeed, true. This problem does *not* occur with
MEG-only data. With EEG present, it happens under one of the following
scenarios:
1. The data have been passed through Maxfilter and only EEG data are
used. Maxfilter does not, of course, have any effect on the EEGs but
mne_inverse_operator got confused when Maxfilter information was
present and MEG was completely omitted.
2. Combined MEG/EEG source estimates are computed and the data have
been processed with Maxfilter. This does not happen always, which made
the debugging a little bit more demanding.
To my knowledge, there is no problem with MEG/EEG estimates when
Maxfilter has *not* been used. EEG-only estimates have been always
correct if Maxfilter has not been applied to the data.
After these bad news, some good ones:
1. The reason for this trouble has been identified.
2. The problem is confined to mne_inverse_operator.
3. mne_inverse_operator source code has been fixed
4. A file containing the fixed mne_inverse_operator is now on the
download web site.
If you suspect that you are affected by this bug:
1. Download and install the fixed mne_inverse_operator
2. Say
mne_redo_file <affacted inv file>
Further good news are that MNE version 2.7.0 is on its way and will be
available for download as a Christmas present. It will, of course,
contain this fix and a few other improvements.
I am sorry for the trouble this bug may have caused.
Best,
Matti
PS. Both the standard and dev installations of MNE (2.6) at the
Martinos center have this bug fix installed.