[FieldTrip] Combineplanars and timelockstatistics

[Sara Aurtenetxe]

Dear Thomas, 

Thanks a lot for your answer.
It makes sense that the non-combined planars may not be ideally suited for sensor-level statistics if sources are not equally
measured by the same planar gradiometers across participants.
Similarly, I can see how dipoles with changing orientation could produce fields with opposite polarity, that is one with a negative peak and another one with a positive peak.
In the non-combined case, these would cancel out to zero in the average, whereas in the combined case both components would sum up to a non-zero average.

In fact, by doing some further research I found a relevant post by Eric Marris and Michael Wibral on this topic:
http://mailman.science.ru.nl/pipermail/fieldtrip/2010-March/002668.html

I now clearly see your point about the combined planar gradients providing a better basis for comparisons across participants, especially for the scenario where
the dipole would be shifting its orientation or position relative to the helmet across participants.

I also imagine that source analysis would provide important additional information here, as it would allow to test the hypothesis that the fields have similar sources across participants.

The question that remains to me, however, is whether it is acceptable to ignore the shifts in the dipole orientation from a neurophysiological perspective.
Is it legitimate to assume that although the dipoles presumably have the same source but different orientations, they reflect the same process if their orientations differ? 
Some biophysical insights on this would be very helpful I guess.

Looking forward to receive further advice on this.
Thanks!

Sara


Sara Aurtenetxe



----- Original Message -----
From: "Thomas Hartmann" <thomas.hartmann at th-ht.de>
To: fieldtrip at science.ru.nl
Sent: Wednesday, November 4, 2015 11:19:45 AM
Subject: Re: [FieldTrip] Combineplanars and timelockstatistics

dear sara,

Am 2015-11-04 um 09:51 schrieb Sara Aurtenetxe:
> When doing timelockstatistics (ERFs) at the gradiometers level:
>
> - Do the gradiometers need to be combined (ft_combineplanar) before the stats?
>    And/or do they need to be combined for visualization of effects?
>
> - Which is the (mathematical) explanation for the answer/s?
let me answer both questions at the same time: the two planar 
gradiometers that make up the set of two that you find at each sensor 
location point into orthogonal directions. you can imagine one pointing 
along the x-axis, the other pointing along the y-axis of a 2d coordinate 
system. so, the gradiometer pointing along the x-axis would pick up 100% 
percent of an activity that increases or decreases in that direction. if 
the activity increases or decreases along the y-axis, the x-axis 
gradiometer would not pick up anything. instead, the other gradiometer 
would pick up the whole energy.
if there is activity that increases or decreases in an angle 45° to both 
gradiometers, both would pick up half the energy. this works accordingly 
for any orientation of the underlying source. i.e. the two gradiometers 
pick up the x and the y part of the signal. this means, that the 
activity at the x-gradiometer is meaningless without the activity at the 
y-gradiometer. and as the orientations of the respective gradiometers 
with respect to the head are arbitrary, it does not make sense to 
compare, let's say the x-gradiometer at one spot with the x-gradiometer 
at another one. mathematically speaking: instead of looking at the 
coordinates of your vector, you want the length of it.

so, in order to get that, you need to do: sqrt(x^2 + y^2).

so, long story short: yes, you need to to ft_combineplanar before sensor 
level stats.

> - Does apply the same when doing source analysis with 'lcmv'?
no, this is a different story. it is quite the opposite: the activity at 
a voxel (or grid point) is the linear combination (i.e. a weighed sum) 
of the activity at all the sensors. the weighing coefficients are what 
you calculate when you first do your forward and the then the inverse 
model (e.g., bem modelling as forward model, LCMV for the inverse 
solution). the forward modelling takes into account the orientation of 
the sensors and also knows, where the head is, i.e. how the brain is 
oriented with respect to the sensor.

so, no need to do ft_combineplanar here.

best,
thomas
>
> Thanks a lot in advance,
> All the best,
>
> Sara
>   
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip

-- 
Dr. Thomas Hartmann

Centre for Cognitive Neuroscience
FB Psychologie
Universität Salzburg
Hellbrunnerstraße 34/II
5020 Salzburg

Tel: +43 662 8044 5109
Email:thomas.hartmann at th-ht.de

"I am a brain, Watson. The rest of me is a mere appendix. " (Arthur Conan Doyle)

_______________________________________________
fieldtrip mailing list
fieldtrip at donders.ru.nl
http://mailman.science.ru.nl/mailman/listinfo/fieldtrip