[FieldTrip] why does ft_denoise_synthetic inserts NaNs in grad fields

[Alexander Backus]

Thanks for your answer Haiteng.

To share our findings on this topic:

After some more debugging, we realized that it could indeed be the ICA
function that inserts NaNs into the grad info.

It seems ft_denoise_synthetic only changes the grad info for some reference
channels...

We now use Haiteng's trick of replacing the post-ICA grad info, with the
post-denoise grad info, however, it feels somewhat dangerous to perform
such "dirty" fixes, since that grad info is probably removed for a certain
reason. We are oblivious to its impact subsequent steps like planar
transform and source reconstruction.

In addition, we dug up the following page on the FieldTrip wiki which
explains the denoising.
http://fieldtrip.fcdonders.nl/faq/how_does_the_ctf_higher-order_gradiometer_work

Thanks!

Cheers,
Alexander


On Sun, Jul 7, 2013 at 8:17 PM, Haiteng Jiang <haiteng.jiang at gmail.com>wrote:

> Hi Alexander,
>       I am not clear about what's  ft_denoise_synthetic doing. However , I
> don't have  such problem when I  use  . The reason you get
> 'mismatching number of channels'  probably because you specify cfg.channel
> ='MEG' instead  of including all channels. My FT code is something like
> this:
>
>            % reading
>           cfg = [];
>           cfg.dataset                 = dataset;
>           cfg.trialdef.eventtype      = '';
>           cfg.trialdef.eventvalue  =  ;
>           cfg.continuous = 'yes';
>           cfg.channel= 'all';
>           cfg.trialdef.prestim = ;
>           cfg.trialdef.poststim = ;
>           cfg = ft_definetrial(cfg);
>           trl =   cfg.trl;
>           cfg.detrend ='yes';
>           cfg.demean = 'yes';
>           data= ft_preprocessing(cfg);
>           cfg.gradient = 'G3BR';
>           data= ft_denoise_synthetic(cfg,data);
>
> In my pipeline , the grad  changes into NAN only after I run I  ICA to
> remove artifacts .   I replace this grad with previous good gradient
> information and assume it is fine for the latter source reconstruction.
> Hope this helps.
>                                                      Best,
>                                                      Haiteng
>
>
>>
>>
>> Message: 2
>> Date: Fri, 5 Jul 2013 09:38:55 +0200
>> From: Alexander Backus <a.backus at donders.ru.nl>
>> To: fieldtrip at science.ru.nl
>> Subject: [FieldTrip] why does ft_denoise_synthetic inserts NaNs in
>>         grad    fields
>> Message-ID:
>>         <
>> CAEfPLbcMKmSkhak0cqQHNp-Lye8xhLJKjauN4GWupBCdVBA6cQ at mail.gmail.com>
>> Content-Type: text/plain; charset="iso-8859-1"
>>
>>
>> Hi,
>>
>> When I use ft_denoise_synthetic on my preprocessed CTF275 MEG data to
>> covert to 3rd order gradient, it replaces the values in data.grad.chanpos
>> and data.grad.chanori with NaNs.
>>
>> I assume this is done because they are no longer valid.
>> When digging into the low-level code I found something about mismatching
>> number of channels.
>>
>> Anyway, when I want to subsequently covert to planar gradient (does this
>> actually makes sense?) or create a source model, I need these grad values.
>>
>> I have two questions:
>>
>> 1) What does ft_denoise_synthetic exactly do (in non-math terms) and why
>> are NaNs inserted in the grad subfields?
>>
>> 2) Can I still use the data for planar gradient conversion or source
>> modeling by using the earlier stored old gradient info (pre
>> ft_denoise_synthetic) or is this ill-advised and is there a more wise
>> course of action?
>>
>> Thanks in advance,
>> Best,
>> Alexander Backus
>>
>
>
>
> --
> Haiteng Jiang
> PhD candidate
> Neuronal Oscillations Group
> Donders Institute for Brain, Cognition and Behaviour
> Centre for Cognitive Neuroimaging
> Radboud University Nijmegen
>
> Visiting address
> Room 2.32
> Donders Centre for Cognitive Neuroimaging
> Kapittelweg 29
> 6525 EN  Nijmegen
> the Netherlands
>
>  Tel.: +31 (0)243668291
>  Web:  https://sites.google.com/site/haitengjiang/
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>



-- 
Alexander R. Backus, MSc
PhD Candidate at the Donders Institute for Brain, Cognition and Behaviour
Centre for Cognitive Neuroimaging
Radboud University Nijmegen
Memory and Space Research Group
www.doellerlab.com

P.O. Box 9101, NL-6500 HB Nijmegen , The Netherlands
Telephone: +31(0)24 36 10754
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20130709/fe0022a4/attachment-0002.html>