pick orientation, MNE, dSPM and group analyses

Dear list,

I'm currently comparing group grand averages in a set of functional labels
which are derived from the PALS B12 Brodmann parcellation. These were then
used with subjects' stcs after morphing to fsaverage.
Now I'm really struck that with surface orientation AND mean flipping the
minima and maxima, even for dSPM shrink to values below 1 while the
expected temporal dynamics are preserved. In the 'wild', that is, *before*
averaging, the signed dSPMS are between -7 and 8, just as the
free-orientation dSPM maxima are around 8 --- *after* --- averaging.

I'm wondering whether this could be a result of the morphing, the
anatomical variability, or even the segmentation quality.

Any hint would be appreciated.

Denis
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140807/e08eed7c/attachment.html

Hi Denis,

This effect can be influenced by a lot of variables. I would say
anatomical variability is a huge one but there are tons of factors
which affect just that:

Did you decimate? (I guess you must have to morph, but how severely
did you decimate?)
Are you using --loose, or --loosevar
What parameters with those
Did you use cps?

I suspect the morphing will also influence this, but that is easy to
check (and wise to do see how the labels morph back on the
individual's surface?). Though as long as you have FreeSurfer quality
scans, I don't expect the segmentation to be an issue. What if any
smoothing did you do (at each stage)?

HTH,
D

Hi Dan,

Hi Denis,

This effect can be influenced by a lot of variables. I would say
anatomical variability is a huge one but there are tons of factors
which affect just that:

Did you decimate? (I guess you must have to morph, but how severely
did you decimate?)

I think I did not explicitlly decimate. Simply a 20 steps morpch from
subject to fsaverage.

Are you using --loose, or --loosevar
What parameters with those

our default, loose=0.2

Did you use cps?

I'm actually not sure whether Python takes the cps into account / where /
when /

I suspect the morphing will also influence this, but that is easy to
check (and wise to do see how the labels morph back on the
individual's surface?).

Yeah, or compute the grand ave time series based on time courses extracted
from unorphed stcs.

Though as long as you have FreeSurfer quality
scans, I don't expect the segmentation to be an issue. What if any
smoothing did you do (at each stage)?

see above

HTH,
D

more imporatanlty, does all this actually matter at all if the SNR seems to
be ok.

Denis

Hi Denis,
   With the orientation fixed, is it the across vertex (within a parcel for a give subject) or the across subject averaging that is reducing the dSPMs? Also, is the data some kind of event related response where you'd expect the peaks across subjects to along in time? I ask because the signed estimates when not aligned across subjects (in time) could go down quite a bit when you average across subjects...the "mis-alignment" could also simply come from the possibility the polarity is not the same across subjects, i.e., though the peaks are around the same time, for some subjects the peaks are negative and for some positive (both in vertex by vertex grand-averaging case ..or the label wise average for each subject separately case)

Hari

Hi Dan,

Hi Denis,

This effect can be influenced by a lot of variables. I would say
anatomical variability is a huge one but there are tons of factors
which affect just that:

Did you decimate? (I guess you must have to morph, but how severely
did you decimate?)

I think I did not explicitlly decimate. Simply a 20 steps morpch from
subject to fsaverage.

How did you map the full ~300,000 vertices from each participant to
fsaverage without decimating?

Hi Hari,

Hi Denis,
   With the orientation fixed,

should try with fixed ...

is it the across vertex (within a parcel for a give subject) or the across
subject averaging that is reducing the dSPMs?

to me it seem it's the cross-subject reduction, cross-vertex looks sane-ish
thanks to flipping.

Also, is the data some kind of event related response where you'd expect
the peaks across subjects to along in time?

yes, roughtly so. I cannot say too much here since I'm not looking at
standard components.

I ask because the signed estimates when not aligned across subjects (in
time) could go down quite a bit when you average across subjects...the
"mis-alignment" could also simply come from the possibility the polarity is
not the same across subjects,

I also thought about this.

i.e., though the peaks are around the same time, for some subjects the
peaks are negative and for some positive (both in vertex by vertex
grand-averaging case ..or the label wise average for each subject
separately case)

the question would then be how to deal with it.

Hari

Hi Dan,

Hi Denis,

This effect can be influenced by a lot of variables. I would say
anatomical variability is a huge one but there are tons of factors
which affect just that:

Did you decimate? (I guess you must have to morph, but how severely
did you decimate?)

I think I did not explicitlly decimate. Simply a 20 steps morpch from
subject to fsaverage.

Are you using --loose, or --loosevar
What parameters with those

our default, loose=0.2

Did you use cps?

I'm actually not sure whether Python takes the cps into account / where /
when /

I suspect the morphing will also influence this, but that is easy to
check (and wise to do see how the labels morph back on the
individual's surface?).

Yeah, or compute the grand ave time series based on time courses extracted
from unorphed stcs.

Though as long as you have FreeSurfer quality
scans, I don't expect the segmentation to be an issue. What if any
smoothing did you do (at each stage)?

see above

HTH,
D

more imporatanlty, does all this actually matter at all if the SNR seems
to be ok.

Denis

> Dear list,
>
> I'm currently comparing group grand averages in a set of functional
labels
> which are derived from the PALS B12 Brodmann parcellation. These were
then
> used with subjects' stcs after morphing to fsaverage.
> Now I'm really struck that with surface orientation AND mean flipping
the
> minima and maxima, even for dSPM shrink to values below 1 while the
expected
> temporal dynamics are preserved. In the 'wild', that is, *before*
averaging,
> the signed dSPMS are between -7 and 8, just as the free-orientation dSPM
> maxima are around 8 --- *after* --- averaging.
>
> I'm wondering whether this could be a result of the morphing, the
anatomical
> variability, or even the segmentation quality.
>
> Any hint would be appreciated.
>
> Denis
>
> _______________________________________________
> Mne_analysis mailing list
> Mne_analysis at nmr.mgh.harvard.edu
> Mne_analysis Info Page
>
>
> The information in this e-mail is intended only for the person to whom
it is
> addressed. If you believe this e-mail was sent to you in error and the
> e-mail
> contains patient information, please contact the Partners Compliance
> HelpLine at
> MyComplianceReport.com: Compliance and Ethics Reporting . If the e-mail was sent to you
in
> error
> but does not contain patient information, please contact the sender and
> properly
> dispose of the e-mail.
>
_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
Mne_analysis Info Page

_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
Mne_analysis Info Page

_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
Mne_analysis Info Page

The information in this e-mail is intended only for the person to whom it
is
addressed. If you believe this e-mail was sent to you in error and the
e-mail
contains patient information, please contact the Partners Compliance
HelpLine at
MyComplianceReport.com: Compliance and Ethics Reporting . If the e-mail was sent to you in
error
but does not contain patient information, please contact the sender and
properly
dispose of the e-mail.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140808/bbfeb90e/attachment.html

> Hi Dan,
>
>
>>
>> Hi Denis,
>>
>> This effect can be influenced by a lot of variables. I would say
>> anatomical variability is a huge one but there are tons of factors
>> which affect just that:
>>
>> Did you decimate? (I guess you must have to morph, but how severely
>> did you decimate?)
>
>
> I think I did not explicitlly decimate. Simply a 20 steps morpch from
> subject to fsaverage.

How did you map the full ~300,000 vertices from each participant to
fsaverage without decimating?

As I said, not *explcitly*, it's the an oct 6 source space that we usually
recommend as default.

>
>>
>> Are you using --loose, or --loosevar
>> What parameters with those
>
>
> our default, loose=0.2
>
>>
>> Did you use cps?
>>
>
> I'm actually not sure whether Python takes the cps into account / where /
> when /
>
>>
>> I suspect the morphing will also influence this, but that is easy to
>> check (and wise to do see how the labels morph back on the
>> individual's surface?).
>
>
> Yeah, or compute the grand ave time series based on time courses
extracted
> from unorphed stcs.
>
>>
>> Though as long as you have FreeSurfer quality
>> scans, I don't expect the segmentation to be an issue. What if any
>> smoothing did you do (at each stage)?
>>
>
> see above
>
>>
>> HTH,
>> D
>>
>
> more imporatanlty, does all this actually matter at all if the SNR seems
to
> be ok.
>
> Denis
>
>>
>> > Dear list,
>> >
>> > I'm currently comparing group grand averages in a set of functional
>> > labels
>> > which are derived from the PALS B12 Brodmann parcellation. These were
>> > then
>> > used with subjects' stcs after morphing to fsaverage.
>> > Now I'm really struck that with surface orientation AND mean flipping
>> > the
>> > minima and maxima, even for dSPM shrink to values below 1 while the
>> > expected
>> > temporal dynamics are preserved. In the 'wild', that is, *before*
>> > averaging,
>> > the signed dSPMS are between -7 and 8, just as the free-orientation
dSPM
>> > maxima are around 8 --- *after* --- averaging.
>> >
>> > I'm wondering whether this could be a result of the morphing, the
>> > anatomical
>> > variability, or even the segmentation quality.
>> >
>> > Any hint would be appreciated.
>> >
>> > Denis
>> >
>> > _______________________________________________
>> > Mne_analysis mailing list
>> > Mne_analysis at nmr.mgh.harvard.edu
>> > Mne_analysis Info Page
>> >
>> >
>> > The information in this e-mail is intended only for the person to whom
>> > it is
>> > addressed. If you believe this e-mail was sent to you in error and the
>> > e-mail
>> > contains patient information, please contact the Partners Compliance
>> > HelpLine at
>> > MyComplianceReport.com: Compliance and Ethics Reporting . If the e-mail was sent to
you
>> > in
>> > error
>> > but does not contain patient information, please contact the sender
and
>> > properly
>> > dispose of the e-mail.
>> >
>> _______________________________________________
>> Mne_analysis mailing list
>> Mne_analysis at nmr.mgh.harvard.edu
>> Mne_analysis Info Page
>
>
>
> _______________________________________________
> Mne_analysis mailing list
> Mne_analysis at nmr.mgh.harvard.edu
> Mne_analysis Info Page
>
>
> The information in this e-mail is intended only for the person to whom
it is
> addressed. If you believe this e-mail was sent to you in error and the
> e-mail
> contains patient information, please contact the Partners Compliance
> HelpLine at
> MyComplianceReport.com: Compliance and Ethics Reporting . If the e-mail was sent to you
in
> error
> but does not contain patient information, please contact the sender and
> properly
> dispose of the e-mail.
>
_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
Mne_analysis Info Page

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140808/04e6b768/attachment.html

IMO, oct-6 is severely decimated (only 2.75% of the points in the
source space originally). I'm not sure that --loose 0.2 will
accurately capture the range of orientations (especially dangerous if
not using the cps). Plus by morphing on top of this coarse spacing
with 20 smooth steps that ends up smoothing the result a lot. If you
want to keep the analysis in the time domain, I would first explore
the timecourses of the labels within the individuals to see what sort
of pattern emerges (particularly with respect to their individual
anatomy). As I said before it could be anatomical variability, but you
may be able to correct for this if the whole label actually has a
consistent orientation across participants (the current decimation (in
combination with your other parameters), could be causing the sign
flipping you are seeing). In that case simply adding cps "should"
improve the situation.

Of course one "easy fix" is to just pick the peak frequency in the
region/time you are interested in and average the power in that band
(this will save you from the flipping issues, by removing phase).

HTH,
D

Aside:
What would be "explicit" decimation? To me reducing the source space
to oct-6 is decimation (how else could one decimate?).

Thanks Dan,

IMO, oct-6 is severely decimated (only 2.75% of the points in the
source space originally).

I agree. But what's the alternative? Do you compute inverse solutions on
the entire source space?

I'm not sure that --loose 0.2 will
accurately capture the range of orientations (especially dangerous if
not using the cps). Plus by morphing on top of this coarse spacing
with 20 smooth steps that ends up smoothing the result a lot. If you
want to keep the analysis in the time domain, I would first explore
the timecourses of the labels within the individuals to see what sort
of pattern emerges (particularly with respect to their individual
anatomy).

I was feeling that this is probably the way to go.

As I said before it could be anatomical variability, but you
may be able to correct for this if the whole label actually has a
consistent orientation across participants (the current decimation (in
combination with your other parameters), could be causing the sign
flipping you are seeing). In that case simply adding cps "should"
improve the situation.

I feel we should handle CPS a bit more verbosely, e.g. log whether it's
used or not when computing / applying inverse.

Of course one "easy fix" is to just pick the peak frequency in the
region/time you are interested in and average the power in that band
(this will save you from the flipping issues, by removing phase).

indeed.

HTH,
D

Aside:
What would be "explicit" decimation? To me reducing the source space
to oct-6 is decimation (how else could one decimate?).

forget about the wording. I simply expressed the fact that these are
parameters we normally don't touch.

Thanks Dan,

IMO, oct-6 is severely decimated (only 2.75% of the points in the
source space originally).

I agree. But what's the alternative? Do you compute inverse solutions on the
entire source space?

I often do, but you could always try with ico 5 (higher numbers than
that make more work, by breaking morphing), to keep it simple and
still have morphing.

> Thanks Dan,
>
>>
>> IMO, oct-6 is severely decimated (only 2.75% of the points in the
>> source space originally).
>
>
> I agree. But what's the alternative? Do you compute inverse solutions on
the
> entire source space?

I often do, but you could always try with ico 5 (higher numbers than
that make more work, by breaking morphing), to keep it simple and
still have morphing.

If only all this wouldn't be that impractical wrt to resources consumption
...

>
>>
>> I'm not sure that --loose 0.2 will
>> accurately capture the range of orientations (especially dangerous if
>> not using the cps). Plus by morphing on top of this coarse spacing
>> with 20 smooth steps that ends up smoothing the result a lot. If you
>> want to keep the analysis in the time domain, I would first explore
>> the timecourses of the labels within the individuals to see what sort
>> of pattern emerges (particularly with respect to their individual
>> anatomy).
>
>
> I was feeling that this is probably the way to go.
>
>>
>> As I said before it could be anatomical variability, but you
>> may be able to correct for this if the whole label actually has a
>> consistent orientation across participants (the current decimation (in
>> combination with your other parameters), could be causing the sign
>> flipping you are seeing). In that case simply adding cps "should"
>> improve the situation.
>
>
> I feel we should handle CPS a bit more verbosely, e.g. log whether it's
used
> or not when computing / applying inverse.
>
>>
>> Of course one "easy fix" is to just pick the peak frequency in the
>> region/time you are interested in and average the power in that band
>> (this will save you from the flipping issues, by removing phase).
>>
>
> indeed.
>
>>
>> HTH,
>> D
>>
>> Aside:
>> What would be "explicit" decimation? To me reducing the source space
>> to oct-6 is decimation (how else could one decimate?).
>>
>
> forget about the wording. I simply expressed the fact that these are
> parameters we normally don't touch.
>
>
>>
>> >
>> >
>> >
>> >>
>> >> > Hi Dan,
>> >> >
>> >> >
>> >> >>
>> >> >> Hi Denis,
>> >> >>
>> >> >> This effect can be influenced by a lot of variables. I would say
>> >> >> anatomical variability is a huge one but there are tons of factors
>> >> >> which affect just that:
>> >> >>
>> >> >> Did you decimate? (I guess you must have to morph, but how
severely
>> >> >> did you decimate?)
>> >> >
>> >> >
>> >> > I think I did not explicitlly decimate. Simply a 20 steps morpch
from
>> >> > subject to fsaverage.
>> >>
>> >> How did you map the full ~300,000 vertices from each participant to
>> >> fsaverage without decimating?
>> >>
>> >
>> > As I said, not *explcitly*, it's the an oct 6 source space that we
>> > usually
>> > recommend as default.
>> >
>> >>
>> >>
>> >> >
>> >> >>
>> >> >> Are you using --loose, or --loosevar
>> >> >> What parameters with those
>> >> >
>> >> >
>> >> > our default, loose=0.2
>> >> >
>> >> >>
>> >> >> Did you use cps?
>> >> >>
>> >> >
>> >> > I'm actually not sure whether Python takes the cps into account /
>> >> > where
>> >> > /
>> >> > when /
>> >> >
>> >> >>
>> >> >> I suspect the morphing will also influence this, but that is easy
to
>> >> >> check (and wise to do see how the labels morph back on the
>> >> >> individual's surface?).
>> >> >
>> >> >
>> >> > Yeah, or compute the grand ave time series based on time courses
>> >> > extracted
>> >> > from unorphed stcs.
>> >> >
>> >> >>
>> >> >> Though as long as you have FreeSurfer quality
>> >> >> scans, I don't expect the segmentation to be an issue. What if any
>> >> >> smoothing did you do (at each stage)?
>> >> >>
>> >> >
>> >> > see above
>> >> >
>> >> >>
>> >> >> HTH,
>> >> >> D
>> >> >>
>> >> >
>> >> > more imporatanlty, does all this actually matter at all if the SNR
>> >> > seems
>> >> > to
>> >> > be ok.
>> >> >
>> >> > Denis
>> >> >
>> >> >>
>> >> >> > Dear list,
>> >> >> >
>> >> >> > I'm currently comparing group grand averages in a set of
>> >> >> > functional
>> >> >> > labels
>> >> >> > which are derived from the PALS B12 Brodmann parcellation. These
>> >> >> > were
>> >> >> > then
>> >> >> > used with subjects' stcs after morphing to fsaverage.
>> >> >> > Now I'm really struck that with surface orientation AND mean
>> >> >> > flipping
>> >> >> > the
>> >> >> > minima and maxima, even for dSPM shrink to values below 1 while
>> >> >> > the
>> >> >> > expected
>> >> >> > temporal dynamics are preserved. In the 'wild', that is,
*before*
>> >> >> > averaging,
>> >> >> > the signed dSPMS are between -7 and 8, just as the
>> >> >> > free-orientation
>> >> >> > dSPM
>> >> >> > maxima are around 8 --- *after* --- averaging.
>> >> >> >
>> >> >> > I'm wondering whether this could be a result of the morphing,
the
>> >> >> > anatomical
>> >> >> > variability, or even the segmentation quality.
>> >> >> >
>> >> >> > Any hint would be appreciated.
>> >> >> >
>> >> >> > Denis
>> >> >> >
>> >> >> > _______________________________________________
>> >> >> > Mne_analysis mailing list
>> >> >> > Mne_analysis at nmr.mgh.harvard.edu
>> >> >> > Mne_analysis Info Page
>> >> >> >
>> >> >> >
>> >> >> > The information in this e-mail is intended only for the person
to
>> >> >> > whom
>> >> >> > it is
>> >> >> > addressed. If you believe this e-mail was sent to you in error
and
>> >> >> > the
>> >> >> > e-mail
>> >> >> > contains patient information, please contact the Partners
>> >> >> > Compliance
>> >> >> > HelpLine at
>> >> >> > MyComplianceReport.com: Compliance and Ethics Reporting . If the e-mail was
sent to
>> >> >> > you
>> >> >> > in
>> >> >> > error
>> >> >> > but does not contain patient information, please contact the
>> >> >> > sender
>> >> >> > and
>> >> >> > properly
>> >> >> > dispose of the e-mail.
>> >> >> >
>> >> >> _______________________________________________
>> >> >> Mne_analysis mailing list
>> >> >> Mne_analysis at nmr.mgh.harvard.edu
>> >> >> Mne_analysis Info Page
>> >> >
>> >> >
>> >> >
>> >> > _______________________________________________
>> >> > Mne_analysis mailing list
>> >> > Mne_analysis at nmr.mgh.harvard.edu
>> >> > Mne_analysis Info Page
>> >> >
>> >> >
>> >> > The information in this e-mail is intended only for the person to
>> >> > whom
>> >> > it is
>> >> > addressed. If you believe this e-mail was sent to you in error and
>> >> > the
>> >> > e-mail
>> >> > contains patient information, please contact the Partners
Compliance
>> >> > HelpLine at
>> >> > MyComplianceReport.com: Compliance and Ethics Reporting . If the e-mail was sent to
>> >> > you
>> >> > in
>> >> > error
>> >> > but does not contain patient information, please contact the sender
>> >> > and
>> >> > properly
>> >> > dispose of the e-mail.
>> >> >
>> >> _______________________________________________
>> >> Mne_analysis mailing list
>> >> Mne_analysis at nmr.mgh.harvard.edu
>> >> Mne_analysis Info Page
>> >
>> >
>> >
>> > _______________________________________________
>> > Mne_analysis mailing list
>> > Mne_analysis at nmr.mgh.harvard.edu
>> > Mne_analysis Info Page
>> >
>> >
>> > The information in this e-mail is intended only for the person to whom
>> > it is
>> > addressed. If you believe this e-mail was sent to you in error and the
>> > e-mail
>> > contains patient information, please contact the Partners Compliance
>> > HelpLine at
>> > MyComplianceReport.com: Compliance and Ethics Reporting . If the e-mail was sent to
you
>> > in
>> > error
>> > but does not contain patient information, please contact the sender
and
>> > properly
>> > dispose of the e-mail.
>> >
>> _______________________________________________
>> Mne_analysis mailing list
>> Mne_analysis at nmr.mgh.harvard.edu
>> Mne_analysis Info Page
>
>
>
> _______________________________________________
> Mne_analysis mailing list
> Mne_analysis at nmr.mgh.harvard.edu
> Mne_analysis Info Page
>
>
> The information in this e-mail is intended only for the person to whom
it is
> addressed. If you believe this e-mail was sent to you in error and the
> e-mail
> contains patient information, please contact the Partners Compliance
> HelpLine at
> MyComplianceReport.com: Compliance and Ethics Reporting . If the e-mail was sent to you
in
> error
> but does not contain patient information, please contact the sender and
> properly
> dispose of the e-mail.
>
_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
Mne_analysis Info Page

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140808/86c211ce/attachment.html

Hi Denis,
   If you are interested in the grand average (across subjects) in a
label, then perhaps you could get a label average for each subject
separately and use those time-courses (1 per subejct) to then do a
"mean flip" like procedure when averaging across subjects?

This was the procedure I used for one of my studies and it worked
reasonably well.. Of course, if you did the loose and used unsigned
values, you don't run into the polarity issue..

Hari

Hari,

good you mention it. I had something similar in mind.

Denis

Hi Denis,
   If you are interested in the grand average (across subjects) in a
label, then perhaps you could get a label average for each subject
separately and use those time-courses (1 per subejct) to then do a
"mean flip" like procedure when averaging across subjects?

This was the procedure I used for one of my studies and it worked
reasonably well.. Of course, if you did the loose and used unsigned
values, you don't run into the polarity issue..

Indeed, but the you also loose comparibility with sensor space results and
for example TFR analyses are prohibitive with free orientation estimates,
etc., and so on.

Denis

Hari

> Hi Hari,
>
>
>
>> Hi Denis,
>> With the orientation fixed,
>>
>
> should try with fixed ...
>
>
>> is it the across vertex (within a parcel for a give subject) or the
>> across
>> subject averaging that is reducing the dSPMs?
>>
>
> to me it seem it's the cross-subject reduction, cross-vertex looks
> sane-ish
> thanks to flipping.
>
>
>> Also, is the data some kind of event related response where you'd expect
>> the peaks across subjects to along in time?
>>
>
> yes, roughtly so. I cannot say too much here since I'm not looking at
> standard components.
>
>
>> I ask because the signed estimates when not aligned across subjects (in
>> time) could go down quite a bit when you average across subjects...the
>> "mis-alignment" could also simply come from the possibility the polarity
>> is
>> not the same across subjects,
>>
>
> I also thought about this.
>
>
>> i.e., though the peaks are around the same time, for some subjects the
>> peaks are negative and for some positive (both in vertex by vertex
>> grand-averaging case ..or the label wise average for each subject
>> separately case)
>>
>>
> the question would then be how to deal with it.
>
>
>> Hari
>>
>>
>> Hi Dan,
>>
>>
>>
>>> Hi Denis,
>>>
>>> This effect can be influenced by a lot of variables. I would say
>>> anatomical variability is a huge one but there are tons of factors
>>> which affect just that:
>>>
>>> Did you decimate? (I guess you must have to morph, but how severely
>>> did you decimate?)
>>>
>>
>> I think I did not explicitlly decimate. Simply a 20 steps morpch from
>> subject to fsaverage.
>>
>>
>>> Are you using --loose, or --loosevar
>>> What parameters with those
>>>
>>
>> our default, loose=0.2
>>
>>
>>> Did you use cps?
>>>
>>>
>> I'm actually not sure whether Python takes the cps into account / where
>> /
>> when /
>>
>>
>>> I suspect the morphing will also influence this, but that is easy to
>>> check (and wise to do see how the labels morph back on the
>>> individual's surface?).
>>
>>
>> Yeah, or compute the grand ave time series based on time courses
>> extracted
>> from unorphed stcs.
>>
>>
>>> Though as long as you have FreeSurfer quality
>>> scans, I don't expect the segmentation to be an issue. What if any
>>> smoothing did you do (at each stage)?
>>>
>>>
>> see above
>>
>>
>>> HTH,
>>> D
>>>
>>>
>> more imporatanlty, does all this actually matter at all if the SNR seems
>> to be ok.
>>
>> Denis
>>
>>
>>> > Dear list,
>>> >
>>> > I'm currently comparing group grand averages in a set of functional
>>> labels
>>> > which are derived from the PALS B12 Brodmann parcellation. These were
>>> then
>>> > used with subjects' stcs after morphing to fsaverage.
>>> > Now I'm really struck that with surface orientation AND mean flipping
>>> the
>>> > minima and maxima, even for dSPM shrink to values below 1 while the
>>> expected
>>> > temporal dynamics are preserved. In the 'wild', that is, *before*
>>> averaging,
>>> > the signed dSPMS are between -7 and 8, just as the free-orientation
>>> dSPM
>>> > maxima are around 8 --- *after* --- averaging.
>>> >
>>> > I'm wondering whether this could be a result of the morphing, the
>>> anatomical
>>> > variability, or even the segmentation quality.
>>> >
>>> > Any hint would be appreciated.
>>> >
>>> > Denis
>>> >
>>> > _______________________________________________
>>> > Mne_analysis mailing list
>>> > Mne_analysis at nmr.mgh.harvard.edu
>>> > Mne_analysis Info Page
>>> >
>>> >
>>> > The information in this e-mail is intended only for the person to
>>> whom
>>> it is
>>> > addressed. If you believe this e-mail was sent to you in error and
>>> the
>>> > e-mail
>>> > contains patient information, please contact the Partners Compliance
>>> > HelpLine at
>>> > MyComplianceReport.com: Compliance and Ethics Reporting . If the e-mail was sent to
>>> you
>>> in
>>> > error
>>> > but does not contain patient information, please contact the sender
>>> and
>>> > properly
>>> > dispose of the e-mail.
>>> >
>>> _______________________________________________
>>> Mne_analysis mailing list
>>> Mne_analysis at nmr.mgh.harvard.edu
>>> Mne_analysis Info Page
>>>
>>
>> _______________________________________________
>> Mne_analysis mailing list
>> Mne_analysis at nmr.mgh.harvard.edu
>> Mne_analysis Info Page
>>
>>
>> _______________________________________________
>> Mne_analysis mailing list
>> Mne_analysis at nmr.mgh.harvard.edu
>> Mne_analysis Info Page
>>
>>
>> The information in this e-mail is intended only for the person to whom
>> it
>> is
>> addressed. If you believe this e-mail was sent to you in error and the
>> e-mail
>> contains patient information, please contact the Partners Compliance
>> HelpLine at
>> MyComplianceReport.com: Compliance and Ethics Reporting . If the e-mail was sent to you
>> in
>> error
>> but does not contain patient information, please contact the sender and
>> properly
>> dispose of the e-mail.
>>
>>
> _______________________________________________
> Mne_analysis mailing list
> Mne_analysis at nmr.mgh.harvard.edu
> Mne_analysis Info Page

--
Hari Bharadwaj
Post-doctoral Associate,
Center for Computational Neuroscience
  and Neural Technology (CompNet),
Boston University
677 Beacon St.,
Boston, MA 02215

Martinos Center for Biomedical Imaging,
Massachusetts General Hospital
149 Thirteenth Street,
Charlestown, MA 02129

hari at nmr.mgh.harvard.edu
Ph: 734-883-5954
www.haribharadwaj.com

_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
Mne_analysis Info Page

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140808/0436650d/attachment.html

Hari,

good you mention it. I had something similar in mind.

Denis

Hi Denis,
   If you are interested in the grand average (across subjects) in a
label, then perhaps you could get a label average for each subject
separately and use those time-courses (1 per subejct) to then do a
"mean flip" like procedure when averaging across subjects?

This was the procedure I used for one of my studies and it worked
reasonably well.. Of course, if you did the loose and used unsigned
values, you don't run into the polarity issue..

Indeed, but the you also loose comparibility with sensor space results and for example TFR analyses are prohibitive with free orientation estimates, etc., and so on.

I agree..
In addition, using the surface fixed orientation constraint is indeed incorporating more prior information and does help improve localization (assuming you have individual MRIs and good registration with the sensors etc.)... It is indeed my preferred approach to constrain the orientations when possible.

Denis

Hari

> Hi Hari,
>
>
>
>> Hi Denis,
>> With the orientation fixed,
>>
>
> should try with fixed ...
>
>
>> is it the across vertex (within a parcel for a give subject) or the
>> across
>> subject averaging that is reducing the dSPMs?
>>
>
> to me it seem it's the cross-subject reduction, cross-vertex looks
> sane-ish
> thanks to flipping.
>
>
>> Also, is the data some kind of event related response where you'd expect
>> the peaks across subjects to along in time?
>>
>
> yes, roughtly so. I cannot say too much here since I'm not looking at
> standard components.
>
>
>> I ask because the signed estimates when not aligned across subjects (in
>> time) could go down quite a bit when you average across subjects...the
>> "mis-alignment" could also simply come from the possibility the polarity
>> is
>> not the same across subjects,
>>
>
> I also thought about this.
>
>
>> i.e., though the peaks are around the same time, for some subjects the
>> peaks are negative and for some positive (both in vertex by vertex
>> grand-averaging case ..or the label wise average for each subject
>> separately case)
>>
>>
> the question would then be how to deal with it.
>
>
>> Hari
>>
>>
>> Hi Dan,
>>
>>
>>
>>> Hi Denis,
>>>
>>> This effect can be influenced by a lot of variables. I would say
>>> anatomical variability is a huge one but there are tons of factors
>>> which affect just that:
>>>
>>> Did you decimate? (I guess you must have to morph, but how severely
>>> did you decimate?)
>>>
>>
>> I think I did not explicitlly decimate. Simply a 20 steps morpch from
>> subject to fsaverage.
>>
>>
>>> Are you using --loose, or --loosevar
>>> What parameters with those
>>>
>>
>> our default, loose=0.2
>>
>>
>>> Did you use cps?
>>>
>>>
>> I'm actually not sure whether Python takes the cps into account / where
>> /
>> when /
>>
>>
>>> I suspect the morphing will also influence this, but that is easy to
>>> check (and wise to do see how the labels morph back on the
>>> individual's surface?).
>>
>>
>> Yeah, or compute the grand ave time series based on time courses
>> extracted
>> from unorphed stcs.
>>
>>
>>> Though as long as you have FreeSurfer quality
>>> scans, I don't expect the segmentation to be an issue. What if any
>>> smoothing did you do (at each stage)?
>>>
>>>
>> see above
>>
>>
>>> HTH,
>>> D
>>>
>>>
>> more imporatanlty, does all this actually matter at all if the SNR seems
>> to be ok.
>>
>> Denis
>>
>>
>>> > Dear list,
>>> >
>>> > I'm currently comparing group grand averages in a set of functional
>>> labels
>>> > which are derived from the PALS B12 Brodmann parcellation. These were
>>> then
>>> > used with subjects' stcs after morphing to fsaverage.
>>> > Now I'm really struck that with surface orientation AND mean flipping
>>> the
>>> > minima and maxima, even for dSPM shrink to values below 1 while the
>>> expected
>>> > temporal dynamics are preserved. In the 'wild', that is, *before*
>>> averaging,
>>> > the signed dSPMS are between -7 and 8, just as the free-orientation
>>> dSPM
>>> > maxima are around 8 --- *after* --- averaging.
>>> >
>>> > I'm wondering whether this could be a result of the morphing, the
>>> anatomical
>>> > variability, or even the segmentation quality.
>>> >
>>> > Any hint would be appreciated.
>>> >
>>> > Denis
>>> >
>>> > _______________________________________________
>>> > Mne_analysis mailing list
>>> > Mne_analysis at nmr.mgh.harvard.edu
>>> > Mne_analysis Info Page
>>> >
>>> >
>>> > The information in this e-mail is intended only for the person to
>>> whom
>>> it is
>>> > addressed. If you believe this e-mail was sent to you in error and
>>> the
>>> > e-mail
>>> > contains patient information, please contact the Partners Compliance
>>> > HelpLine at
>>> > MyComplianceReport.com: Compliance and Ethics Reporting . If the e-mail was sent to
>>> you
>>> in
>>> > error
>>> > but does not contain patient information, please contact the sender
>>> and
>>> > properly
>>> > dispose of the e-mail.
>>> >
>>> _______________________________________________
>>> Mne_analysis mailing list
>>> Mne_analysis at nmr.mgh.harvard.edu
>>> Mne_analysis Info Page
>>>
>>
>> _______________________________________________
>> Mne_analysis mailing list
>> Mne_analysis at nmr.mgh.harvard.edu
>> Mne_analysis Info Page
>>
>>
>> _______________________________________________
>> Mne_analysis mailing list
>> Mne_analysis at nmr.mgh.harvard.edu
>> Mne_analysis Info Page
>>
>>
>> The information in this e-mail is intended only for the person to whom
>> it
>> is
>> addressed. If you believe this e-mail was sent to you in error and the
>> e-mail
>> contains patient information, please contact the Partners Compliance
>> HelpLine at
>> MyComplianceReport.com: Compliance and Ethics Reporting . If the e-mail was sent to you
>> in
>> error
>> but does not contain patient information, please contact the sender and
>> properly
>> dispose of the e-mail.
>>
>>
> _______________________________________________
> Mne_analysis mailing list
> Mne_analysis at nmr.mgh.harvard.edu
> Mne_analysis Info Page

--
Hari Bharadwaj
Post-doctoral Associate,
Center for Computational Neuroscience
  and Neural Technology (CompNet),
Boston University
677 Beacon St.,
Boston, MA 02215

Martinos Center for Biomedical Imaging,
Massachusetts General Hospital
149 Thirteenth Street,
Charlestown, MA 02129

hari at nmr.mgh.harvard.edu
Ph: 734-883-5954
www.haribharadwaj.com

_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
Mne_analysis Info Page

_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu
Mne_analysis Info Page

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140808/d72a049c/attachment.html