Eric Maris e.maris at DONDERS.RU.NL
Fri Mar 5 21:52:49 CET 2010

```Dear FT-colleagues,

> The fundamental underlying problem is that you go from a scalar field

> and massively paralellel univariate statistics to a vector field and,

> hence, truly multivariate statistics (e.g. because the local gradient

> vectors will rotate across the course of an ERP/ERF for example).

> Coming up with a p-value there sounds difficult to me. I always

> wondered how people with planar gradiometer devices solve that issue

> when doing time-domain stuff? Maybe there are solutions out there

Following up on the proposal by Jan-Mathijs, this is how you can do it:

·        Use the following test statistic at the level of the (_dV,_dH) channel pairs: calculate the difference between the experimental conditions of the lengths of the trial-averaged vector-valued (_dV,_dH)-signals (J-M calls this “to Pythagoras”). You do this for all channels and all time points

·        Find the channel- and time-point-specific permutation distributions of these statistics

·        Use these channel- and time-point-specific permutation distributions to find sensible thresholds for the statistics

·        Do cluster-based permutation testing using these thresholded test statistics that are subsequently combined in a maximum cluster statistic to control the false alarm

If you write the statfun for the channel-specific test statistic, then the rest can be performed using existing Fieldtrip code.

Best,

Eric Maris

dr. Eric Maris
Donders Institute for Brain, Cognition and Behavior

Center for Cognition and F.C. Donders Center for Cognitive Neuroimaging

P.O. Box 9104
6500 HE Nijmegen
The Netherlands
T:+31 24 3612651
Mobile: 06 39584581

F:+31 24 3616066
E: e. <mailto:e.maris at donders.ru.nl> maris at donders.ru.nl

>

> Michael

>

> -----Ursprüngliche Nachricht-----

> Von: jan-mathijs schoffelen <jan.schoffelen at DONDERS.RU.NL>

> Gesendet: Mar 5, 2010 11:12:03 AM

> An: FIELDTRIP at NIC.SURFNET.NL

> Betreff: Re: [FIELDTRIP] quiestions about planar transformation

>

> >Dear Mark,

> >

> >There may be some confusion here, relating to whether we are talking

> >about frequency domain data (=power, always a positive value), or

> >whether we are talking about time domain data (=amplitude, can be

> >positive and negative).

> >

> >(or magnetometer) data is not the best thing to do for time domain

> >stuff.

> >Reason: combination takes place by applying Pythagoras' rule to the

> _dV

> >and _dH pairs, and when the individual _dH and _dV components are

> >noisy, the noise is also squared, added, and squar-rooted, which leads

> >to an 'amplification' of noise. Not good. It works after trial-

> >averaging because you squeeze out the noise first and Pythagoras the

> >reduced-noise signals.

> >

> >For frequency domain data this is not an issue.

> >Reason: combination usually takes place by just adding the _dV and _dH

> >pairs (power is a squared value already). This is just a linear step

> >(just like averaging) and the order of performing the linear steps

> does

> >not matter for the result. So single trial combination prior to

> >averaging or vice versa should not make a difference here.

> >

> >In general I would advise against trying to do statistics on the

> >combined planar gradient and use it for visualization purposes only.

> >Alternatively, one could come up with a non-parametric statistical

> test

> >(permutation), in which I could think of a way of extracting a p-

> value

> >from a single _dH/_dV pair combined. But this would be a different

> >story...

> >

> >Best,

> >

> >Jan-Mathijs

> >

> >

> >> Hi all

> >>

> >> I am currently a little confused as to how to properly use megplanar

> >> and combineplanar for my 4D 148 axial gadiometer data. I would

> >> ideally like to use planar gradients as it will make subsequent ERF

> >> and frequency analysis easier to interpret. My question is when

> >> should I combine the v and h components?

> >>

> >> Looking  at some previous posts it is suggested that you should

> >> combine the gradients on an individual basis, but when I do this i

> >> get an almost flat amplitude across all sensors which doesn't look

> >> right (see attached image, 'bad_planar'). When i use combineplanar

> >> after averaging, the data the data looks ok (good_planar). I get the

> >> same result doing both megplanar and combineplanar on the averaged

> >> data.

> >>

> >> I just want to double check that the 'bad' planar fields i am

> getting

> >> are correct and not due to a bug in field trip. Intuitively, I would

> >> expect combining the components on the individual level world be

> more

> >> accurate as it would prevent fields from opposeingly orientated

> >> dipoles cancelling out when averaged.

> >>

> >> If it is the case that you should use combineplanar after trial

> >> averaging, how do you go about combining after statistical tests?

> >> i.e. when you have a p-value or a test statistic for the v and h

> >> components for each sensor, how would you combine these together?

> >>

> >> Any help or advice would be appreciated.

> >>

> >> Many thanks

> >>

> >> Mark

> >

> >Dr. J.M. (Jan-Mathijs) Schoffelen

> >Donders Institute for Brain, Cognition and Behaviour, Centre for

> >Cognitive Neuroimaging, Radboud University Nijmegen, The Netherlands

> >J.Schoffelen at donders.ru.nl

> >Telephone: 0031-24-3668063

> >

> >----------------------------------

> >The aim of this list is to facilitate the discussion between users of

> the FieldTrip  toolbox, to share experiences and to discuss new ideas

> http://listserv.surfnet.nl/archives/fieldtrip.html and

> http://www.ru.nl/neuroimaging/fieldtrip.

>

> ----------------------------------

> The aim of this list is to facilitate the discussion between users of

> the FieldTrip  toolbox, to share experiences and to discuss new ideas

> http://listserv.surfnet.nl/archives/fieldtrip.html and

> http://www.ru.nl/neuroimaging/fieldtrip.

----------------------------------
The aim of this list is to facilitate the discussion between users of the FieldTrip  toolbox, to share experiences and to discuss new ideas for MEG and EEG analysis. See also http://listserv.surfnet.nl/archives/fieldtrip.html and http://www.ru.nl/neuroimaging/fieldtrip.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20100305/e58f0cdc/attachment-0002.html>
```