Hi,
we collected meg and eeg data in a task with several runs. I am wondering
how to average the forward solution of these runs combining the meg and
eeg data? mne_average_forward_solutions only averages meg forward
solutions but not any eeg data.
thanks, Sarah
The development version of MNE contains mne_average_forward_solutions
which does both MEG and EEG.
- Matti
Hi
I have reservations about the concept of averaging MEG forward solutions.
You will effectively spatially blur data that does not need to be blurred.
You lose information about the brain this way.
The forward solution for MEG is based on head structural information and
position relative to the MEG sensors.
If you assume the subject did not move between sessions, then you should
gain nothing by averaging forward solutions.
On the other hand, if the subject's head DID move between runs, a more
precise solution would be to solve the forward solution per run, compute
source space inverse models of each run, THEN do averaging across runs,
because then you are in a common coordinate frame, and you are not spatially
blurring anything.
On the other hand, if the head moved DURING runs, you are required to at
least consider correcting for that continuously using the SSS technique,
which could achieve the same thing for you - i.e. getting all your data into
a common coordinate system.
In the EEG forward solution, head movement is not going to change your
forward solution at all. It is based entirely on head structural information
and EEG electrode position information, which does not change. The exception
to this is if you are accounting for changes in conductivities of the tissue
layers - and those may change dynamically. (typically we simply assume
constant values from the literature)
So actually, I don't think you would gain anything at all by averaging EEG
forward solutions. You also shouldn't lose anything, since across runs the
EEG forward solutions should be identical.
This is just my perspective. Matti (or anybody else!)- please correct me if
I am mistaken in any of the above.
Daniel
hi,
Hi
I have reservations about the concept of averaging MEG forward
solutions. You will effectively spatially blur data that does not
need to be blurred. You lose information about the brain this way.The forward solution for MEG is based on head structural
information and position relative to the MEG sensors.
If you assume the subject did not move between sessions, then you
should gain nothing by averaging forward solutions.
On the other hand, if the subject's head DID move between runs, a
more precise solution would be to solve the forward solution per
run, compute source space inverse models of each run, THEN do
averaging across runs, because then you are in a common coordinate
frame, and you are not spatially blurring anything.
On the other hand, if the head moved DURING runs, you are required
to at least consider correcting for that continuously using the SSS
technique, which could achieve the same thing for you - i.e.
getting all your data into a common coordinate system.
This is true but still averaging the forward solutions is a good
first-order approximation as discussed in
Uutela K, Taulu S, and Hamalainen M, Detecting and correcting for
head movements in neuromagnetic measurements. Neuroimage, 14:
1424-31, 2001.
Also,
Wehner DT, Hamalainen MS, Mody M, and Ahlfors SP, Head movements of
children in MEG: quantification, effects on source estimation, and
compensation. Neuroimage, 40: 541-50, 2008.
indicates that even the within-run movement might not be such a bad
problem as we are lead to believe.
In the EEG forward solution, head movement is not going to change
your forward solution at all. It is based entirely on head
structural information and EEG electrode position information,
which does not change. The exception to this is if you are
accounting for changes in conductivities of the tissue layers - and
those may change dynamically. (typically we simply assume constant
values from the literature)
So actually, I don't think you would gain anything at all by
averaging EEG forward solutions. You also shouldn't lose anything,
since across runs the EEG forward solutions should be identical.
This is exactly true if the runs were within the same session. Across
sessions, the electrode positions may be a little bit different.
This is just my perspective. Matti (or anybody else!)- please
correct me if I am mistaken in any of the above.
- Matti
I defer to Matti on the relative value of doing what I proposed.
I stand by the following more restricted statements:
Procedure A: average forward models, then average MEG data in sensor space
and use common forward/inverse model
Procedure B: use individual forward models, compute inverse for each
dataset, then average in brain source space
- It is less precise to do procedure A than procedure B between runs (or
sessions) provided that it is true that you have minimal movement within run
(or session). The level of precision difference will depend on many factors,
including differences in head position (in MEG) and differences in EEG
electrode placement (in EEG)
- If there is head movement within a run or session, averaging forward
models (only in MEG) between adjacent runs (not sessions) may have an
advantage of crudely correcting for the movement. SSS, if you trust the
method, would represent a finer correction because it accounts for head
position continuously.
Doing procedure B also is computationally slower.
D