I have a question about using the inverse matrix computed using MEG
data in MNE from a visual evoked experiment. I have two scans, both
visual evoked experiments, and I want to take the inverse matrix (and
associated noise covariance matrix) computed during scan 1 and apply
it to scan 2. I don't think there is a way to do this using the
compiled MNE, but I have been trying to use the matlab toolbox code to
read in the matrices (forward solution, inverse matrix and noise
covariance matrix) I've dumped from MNE run on scan 1.
My question is, is there a way to apply the solution for scan 1 to
scan 2's dataset? Can I do this using the matlab toolbox or is there
a way to do this using MNE itself? Does anyone know of specific
matlab commands from the MNE toolbox that
would allow us to do this? I had thought mne_read_inverse_operator.m
was the right command to read it
in, but I don't know how to use the matrix once read in. ?Is there a
toolbox command for
applying it to the other dataset? Thanks,
Erik Beall
In this case, I believe you can use the inversion from the first run
on the second.
Simply, open the second run's average with the first runs inverse
operator ** in mne_analyze. This should work.
**You should be aware of possible mislocalizations introduced doing
this, because the person's head will not be in exactly the same
position. I would recommend at least comparing the head positions
between the two runs. If you have maxfilter available, I would
probably use this to transform the head position.
I don't actually see any reason you would want to use the same forward solution, just the same noise covariance matrix. So, I would use mne to make a new forward solution with the second run then just use the noise covariance matrix from the first run to create an inverse solution for the second run.
Well, as for which is "more wrong" I'm not sure, it would probably
depend on the particular specifics, however, personally, I would
prefer the noise covariance fit the forward model than the average
data, but both should provide you a solution.
Thanks Daniel. We looked at the motion between runs, and it was
minimal, although maybe we should try MaxMove anyway. However, I
don't see where this is an option in MNE, am I missing something? MNE
computes the inverse solution on the fly, and I dumped it to file, but
I dont' know how to load it in other than using the matlab toolbox.
The reason we want to do this is to look at the stability of the
solution from one scan to the next. So we could use the empty room
noise covariance matrix, a resting state scan noise covar matrix,
either of the two visual scans, or the average of the two visual
scans, right? For our first try, we wanted to just keep it simple and
take the inverse operator for scan1 to scan 2 and see how similar the
solution was. Does that make sense?
erik
Fair enough, the other thing that can be done is that you can average forward solutions (if I am not mistaken), but almost any of these solutions will lead to some mislocalizations.
I've been given a more thorough understanding of MEG by John Mosher
and Rey Ramirez, who both patiently explained how the inverse solution
works in MEG and that I was asking crazy things. Now that I know
better, I can ask slightly less crazy things next time that aren't
simple wastes of time (I certainly won't be doing the things I'd asked
for - they don't make any sense!). Thanks,
erik