External Email - Use Caution
Dear MNE users,
can anyone enlighten me a little about the three 'default' projections
PCA-mag-v1, PCA-mag-v2 and PCA-mag-v3. Where do they come from and why are they
"discaded" by the maxfiltering process? My mags look pretty clean with them
applied, but after maxfiltering, they look much worse, similar to not applying
the PCA projections.
best,
Marijn.
External Email - Use Caution
First they are discarded as you project to the low rank coordinate system
of maxfilter. After this the SSPs are not valid any longer. Question : do
you take into account bad channels before computing maxfilter? It sounds as
if something is not right with its computation.
Question 2: if your data looks good, do you need maxfilter then in the
first place?
Denis
External Email - Use Caution
Hi Denis,
thanks for confirming that the PCA projections are indeed dropped.
Q1: I?m careful to take bad channels into account. However, I don?t have access to a crosstalk file or a fine calibration file. Maybe these will make all the difference.
W2: I guess you are right in that I don?t *need* maxfilter, since the data is pretty clean to begin with. I was just wondering about the interplay between the PCA projections and maxfilter, since the projections seem superior in this case.
best,
Marijn.
External Email - Use Caution
I?d say if your data is clean enough to do without maxfilter then don?t bother with it. The proj approach is less invasive and more conservative.
External Email - Use Caution
Hi Marijn,
Regarding your first, question:
can anyone enlighten me a little about the three 'default' projections
PCA-mag-v1, PCA-mag-v2 and PCA-mag-v3. Where do they come from
They are coming from the empty room recording.
Also,
However, I don?t have access to a crosstalk file or a fine calibration
file
According to Maxfilter manual, you can find them at:
\(NEUROMAG\_ROOT\)/databases/ctc/ct\_sparse\.fif (NEUROMAG_ROOT)/databases/sss/sss_cal.dat
Best,
Mainak
Best,
Mainak
External Email - Use Caution
Thanks, Mainak! I managed to find the files based on this. Sadly, using the
fine calibration files did not make the filter much better. So I'm going
with Denis' advice: drop the SSS for this particular subject.
External Email - Use Caution
Ah, one more thing. Is there a way to keep using the PCA projections, but
still make use of maxfilter to "move" the head position? What would happen
is I first applied the projections and then started doing maxfilter? Or is
there a way to re-compute the PCA projections from empty room data after
maxfiltering?
External Email - Use Caution
Is there a way to keep using the PCA projections, but still make use of
maxfilter to "move" the head position?
No not currently. You could maybe hack it together by creating a RawArray
with your give `meas_info` and data with three "time points" that are your
PCA vectors.
What would happen is I first applied the projections and then started
doing maxfilter?
maxwell_filter should complain if you try to do this. It would cause a
mismatch between the multipole moment forward model and data. In theory we
might be able to compensate for this (e.g., by applying the projectors to
the multipole moment model) but this has not been implemented or tested.
Or is there a way to re-compute the PCA projections from empty room data
after maxfiltering?
Typically what people tend to do is process the empty room data the same
way that the subject's data is processed (same translation, tSSS params,
etc.) for example for computing an empty-room noise covariance. In theory
you could instead (or additionally) compute the projection vectors from it.
Eric
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20190717/6cd57d92/attachment.html
External Email - Use Caution
Hi Marijn,
Empty room PCA vectors and the external SSS vectors that are based on a
physical model (vector spherical harmonics converging at the origin of the
coordinate system) typically span a very similar signal subspace.
Therefore, it sounds strange if your MaxFilter-processed data look very bad
while SSP-processed data with just three empty room vectors are good, and
potential bad channels have been excluded in the operations. I would like
to understand what's going on. You could try MaxFilter with a lower
external expansion order by setting -out 1 and then MaxFilter and SSP
should be even more compatible in your case. But I doubt that this solves
the problem. I would be happy to take a look at your problematic file, if
possible?
Best,
Samu