I'm currently trying to applying a a prior source constraints via the fMRI
weighting file. But according to MNE documentation:
"It turns out that the fMRI weighting has a strong influence on the MNE
but the noise-normalized estimates are much less affected by it."
(http://martinos.org/mne/stable/manual/mne.html#cbbdijhi)
Can anyone please explain to me why this is such a case, maybe in both
mathematical and intuitive sense. Thank you very much in advance.
dSPM / sLORETA use scaling to downweight locations
where the noise is strong / highly amplified. This amplification
factor is proportional to the fMRI weight. So if you normalize
by division you pretty much cancel the main effect of the fMRI weight.
Is there anyway to use the dSPM or sLORETA algorithm and still be able to
reliably apply the fMRI prior weighting constraints?
Because it seems to me that, with fMRI constraints, WMNE algorithm performs
well with reasonable, as expected results, but after the noise
normalization steps (dSPM/sLORETA), the results become unexpected, if not
inaccurate.
Thank you MNE users
Hi Thinh,
Could you share any plots/images of your unexpected/inaccurate results?
Also, could you motivate why you want to do a dSPM (on a whole brain
basis) given that you have fMRI priors?
As mentioned by Alex, the dSPM/sLORETA normalizations make it such that
everywhere on the brain, the baseline is of unit variance.. This would
indeed negate the fMRI prior weighting because the fMRI prior scales down
some vertices relative to the others (whereas dSPM tends to equalize them
by making them unity)..
In my mind, given that you have an fMRI prior, if you are still interested
in obtaining a z-score type metric, rather than currents, then perhaps one
way to do that would be to mask out the vertices that are not active in
the fMRI before converting to z-score..
Thank you for the reply, I guess it makes little sense trying to use both
fMRI prior and dSPM/sLORETA algorithm. I was just concern about the
accuracy of the results using WMNE alone, as it is accepted that
dSPM/sLORETA out perform WMNE. I don't have any reliable way to validate my
results so I was just trying things out. Maybe I don't fully understand the
concept, intuition behind the normalizations done in dSPM/sLORETA, if you
have any good literature/papers in mind that you could point me to, it
would be very much appreciated.
Hi Thinh,
The dSPM sLORETA outperform MNE is because they normalize noise at each voxels.
To validate your results, you can do enough simulations. You can put simulated sources at your expected locations and then using WMNE to reconstuct them under specific conditions. To some degree, this methods can
validate your results. In fact, for real MEG, no gold truth to proof your results or you simultaneously recording EcoG at each voxels.
2014-07-02
junpeng.zhang
???Thinh Nguyen <thinhnguyen0405 at gmail.com>
???2014-07-02 06:29
???Re: [Mne_analysis] Applying the fMRI weighted w file on the inverse operator
???"Discussion and support forum for the users of MNE Software"<mne_analysis at nmr.mgh.harvard.edu>
???
Hi Hari,
Thank you for the reply, I guess it makes little sense trying to use both fMRI prior and dSPM/sLORETA algorithm. I was just concern about the accuracy of the results using WMNE alone, as it is accepted that dSPM/sLORETA out perform WMNE. I don't have any reliable way to validate my results so I was just trying things out. Maybe I don't fully understand the concept, intuition behind the normalizations done in dSPM/sLORETA, if you have any good literature/papers in mind that you could point me to, it would be very much appreciated.
Best,
Thinh Nguyen
Hi Thinh,
Could you share any plots/images of your unexpected/inaccurate results?
Also, could you motivate why you want to do a dSPM (on a whole brain
basis) given that you have fMRI priors?
As mentioned by Alex, the dSPM/sLORETA normalizations make it such that
everywhere on the brain, the baseline is of unit variance.. This would
indeed negate the fMRI prior weighting because the fMRI prior scales down
some vertices relative to the others (whereas dSPM tends to equalize them
by making them unity)..
In my mind, given that you have an fMRI prior, if you are still interested
in obtaining a z-score type metric, rather than currents, then perhaps one
way to do that would be to mask out the vertices that are not active in
the fMRI before converting to z-score..