planar gradients compared

[Robert Oostenveld]

Hi MAsao,

On 8 Nov 2004, at 22:33, Matsuhashi, Masao (NIH/NINDS) wrote:

> I also copied the axial2planar.m and modified the number of
> gradiometers
> from 151 to 274, and now it works.

Could you send me the updated version, so that I can include it in the
next release?

> The result in our condition is
>
> ================
> method 1: 0.715570
> method 2: 0.923964
> method 3: 0.929785
> method 4: 0.977297
> method 5: 0.715570
> ================
>
> Actually, the result of 'orig' method is flower-like and so different
> from
> the original.

These numbers are the correlation coefficient, and method1 or 'orig',
which is the default(!) in fieldtrip is clearly not the best. Btw.
'orig' means that it was the original implementation of the planar
gradient computation by Ole Jensen, and does not mean that it is the
original axial data. If you plot the amplitude of the planar gradients
computed with 'orig', these should _not_ look like the original
topoplot. Planar gradient amplitudes are conceptually different from
axial gradients (e.g. amplitude is always a positive number, whereas
the original data is both positive and negative).

Did you check this for different dipole locations? And for different
noise levels? The idea was not just to run the script that I sent you,
but also to explore it to see how these parameters affect the planar
gradient estimation from the axial data.

> While doing this, I got the following warning message. Is this ok?
>
> ================
> Warning: higher order synthetic gradiometer configuration(?)
>> In fieldtrip\private\prepare_vol_sens at 145
>   In dipolesimulation at 35
> ================

It indicates a potential problem. Fieldtrip is not yet 100% capable of
handling the CTF higher order synthetic gradients correctly. At the
Donders, we useually do not use the higher order gradients and
therefore they have not been implemented completely yet. The problem
that I face is that I cannot read the coefficients in a decent fashion
(i.e. from the res4 files). In this case (computation of planar
gradients) I guess that it does not matter, but for forward and inverse
computations the warning is more serious.

> Another question: what is the unit of cfg.dip.mom to give to
> dipolesimulation(). With the unit vector [1,0,0] as in the above test
> script, the output (axial.trial) is in the order of 1e-9, where the
> actual
> recorded signal is usually less than 1e-11. Therefore it is not likely
> nAm.
> I am thinking of mixing the simulated data to the real brain background
> activity.

It is arbitrary units. It depends on the units of your other objects,
i.e. whether the volume conductor model is specified in cm or in m,
etc.

Robert