[FieldTrip] channelrepair issue

"Jörn M. Horschig" jm.horschig at donders.ru.nl
Sat Jul 9 03:33:18 CEST 2011


Hi,

SInce the new updates (which should be available in the new FT download 
version, see also the corresponding mail on the mailing list), I would 
suggest to use a user-defined neighbourstructure, verify this and then 
call channelrepair. E.g. try 'triangulation' as a method - it might 
result in a better neighbourhood representation.
Also, currently neighbourselection is using config settings first before 
searching the data structure for sensor position information. So if you 
define cfg.layout, it will definitely use the layout (so, 2D 
coordinates) instead of 3D coordinates defined in your data. This might 
be suboptimal as well.

Sorry for all the inconveniences. We will try to check why the problem 
you got causes an issue and hopefully find a more generic way to solve 
this.
Jim, if not already done, it would be great if you followed Jan-Mathijs' 
suggestion to provide further information such as example data so that 
we can reproduce your problem and try to solve it.
Don, thanks for helping out ;)

Best,
Jörn


On 7/9/2011 12:57 AM, Rojas, Don ( Employee MEG-SOM ) wrote:
> As a short follow up on this, the behavior of ft_neighbourplot is 
> suspect only when supplying a cfg.neighbours field and plotting the 
> result using the 2d layout file (as in the example I gave below), so 
> this issue may not be the same as for the channel repair function. 
> Using no layout file, the 3D plotted locations and labels appear to be 
> correct. I can verify that the lower level functions called by 
> ft_neighbourplot such as channelposition.m are returning the correct 
> position and label info, but haven't had time to track the 2d plotting 
> result down further. The neighbours returned by ft_neighbourselection 
> also appear to be okay given a specified input distance.
>
> On Jul 8, 2011, at 1:17 PM, Rojas, Don ( Employee MEG-SOM ) wrote:
>
>> Jörn, Jim et al.
>>
>> There is indeed a general problem, most likely with channel indexing, 
>> in gradiometer systems. I think the data are read correctly with the 
>> low level functions, but functions like ft_channelrepair, 
>> ft_neighbourselection and ft_neighbourplot are not accounting for the 
>> locations correctly. I have attached a picture that I think 
>> illustrates this issue reasonably well. The following code was used 
>> t<neighbors.jpg>o generate this plot:
>>
>> % read in data
>> cfg = [];
>> cfg.dataset = 'e,rfhp1.0Hz,COH';
>> coh = ft_preprocessing(cfg);
>>
>> % get neighbors and plot
>> cfg = [];
>> cfg.neighbourdist = .03;
>> cfg.neighbours = ft_neighbourselection(cfg,coh);
>> cfg.layout = '4D248.lay';
>> ft_neighbourplot(cfg,coh);
>>
>> You can see that the channel selected is labeled A216, which is 
>> incorrect - it should be A248. It's three neighbors are also 
>> incorrect (A208, A118 and A134 should probably be A195, A228 and A247).
>>
>> I'm guessing this is the same indexing issue for gradiometers. 
>> Jan-Mathijs mentioned a fixed version of ft_channelrepair appearing 
>> today, so perhaps it will be the same problem and the fix can be 
>> propagated to these other functions.
>>
>> Don
>>
>> On Jul 8, 2011, at 2:20 AM, Jörn M. Horschig wrote:
>>
>>> Dear Jim, Don and others,
>>>
>>> First of all, what Don suggested sounds indeed like the cause, if I 
>>> remember things correctly (with my limited experience).
>>>
>>> Anyhow, could one of you check ft_neighbourplot with the dataset and 
>>> the neighbourselection of choice? This allows you to see what 
>>> neighbours are selected for which sensors and where they lie in 
>>> space. If this looks fine, than the problem lies indeed in how 
>>> sensor position is deduced from the data structure.
>>>
>>> Best,
>>> Jörn
>>>
>>>
>>> On 7/8/2011 9:11 AM, jan-mathijs schoffelen wrote:
>>>> Dear Don,
>>>>
>>>> This sounds like a very likely cause of the problem. When we wrote 
>>>> the 4D reading code at the time, we had the 248-channel 
>>>> magnetometer system in mind. It could be that there are some flaws 
>>>> related to the handling of gradiometers. This should not only lead 
>>>> to (erratic) issues at the level of channelrepair, but may also 
>>>> cause mayhem when doing planar gradient transformation, or source 
>>>> reconstruction.
>>>> Would you happen to
>>>> 1) feel like tackling this one with me, and improve FieldTrip to 
>>>> deal with gradiometer systems correctly?
>>>>
>>>> or
>>>>
>>>> 2) have a small gradiometer dataset which you can send to me (e.g. 
>>>> through DropBox), so that I can look into it?
>>>>
>>>> We can further discuss this off the discussion list.
>>>>
>>>> Best wishes,
>>>>
>>>> Jan-Mathijs
>>>>
>>>>
>>>>
>>>> On Jul 8, 2011, at 3:36 AM, Rojas, Don ( Employee MEG-SOM ) wrote:
>>>>
>>>>> Jim et al.
>>>>>
>>>>> In thinking about this further, it seems to me that it is the 
>>>>> indexing that is incorrect. The problem I think is specific to 
>>>>> gradiometer systems, which have two (or more) locations per 
>>>>> channel. The function appears to assume that the sensor index 
>>>>> based on the sens.label field (only 1 per channel) will correspond 
>>>>> 1 to 1 with the sens.grad field (2 locations per channel), which 
>>>>> is not correct. So, my guess is that for first-order gradiometer 
>>>>> systems, the correct index for the sensor location is 2x the index 
>>>>> in the sens.label field.
>>>>>
>>>>> So, in an example I have handy at home, the label index for 
>>>>> channel A234 is 107. Since I think Fieldtrip stores gradiometer 
>>>>> locations (by default) as:
>>>>>
>>>>> Chn 1 lower coil x, y z
>>>>> Chn 1 upper coil x, y z
>>>>> Chn 2 lower coil x, y z
>>>>> Chn 2 upper coil x, y z
>>>>> .
>>>>> .
>>>>> .
>>>>> Chn N
>>>>>
>>>>> I think the correct position info for channel A234 would be found 
>>>>> in sens.grad(214,:), not sens.grad(107,:), which is what the 
>>>>> ft_channelrepair function chooses. So I think this is a bug in the 
>>>>> function specific to gradiometer systems.
>>>>>
>>>>> Best,
>>>>>
>>>>> Don
>>>>>
>>>>> On Jul 7, 2011, at 5:38 PM, Jim Li wrote:
>>>>>
>>>>>> Thank you very much for confirming the problem, Don, I appreciate 
>>>>>> it, :)
>>>>>> Jim
>>>>>> On Thu, Jul 7, 2011 at 4:30 PM, Rojas, Don 
>>>>>> <Don.Rojas at ucdenver.edu <mailto:Don.Rojas at ucdenver.edu>> wrote:
>>>>>>
>>>>>>     Jim,
>>>>>>
>>>>>>     All I can do is confirm that this issue exists for our 4D
>>>>>>     data as well, although we don't use the ft_channelrepair
>>>>>>     function. The same 4 quite distant channels (one had been
>>>>>>     deleted in the dataset I had handy) were chosen for repair
>>>>>>     when I ran this on one of our datasets (also a 248 channel 4D
>>>>>>     system). I agree the choice doesn't make any sense given the
>>>>>>     3 cm distance you define. A quick look at the critical
>>>>>>     calculation in ft_channelrepair, line 109 distance = ... is
>>>>>>     correct given the inputs. So I wonder if the issue is with
>>>>>>     the actual channel indices used rather than the calculation,
>>>>>>     but I don't have time to track that down. All I can say is
>>>>>>     that for the distance you specify, channels A235, A216, A199,
>>>>>>     A215 would be much closer to A234 than those that are
>>>>>>     selected via the channel repair function. Channel A8 is in
>>>>>>     fact nearly 21 cm away from channel A234 in the array, based
>>>>>>     on examination of the data read in from Eugene Kronberg's
>>>>>>     pdf4D matlab object code, which I think was originally the
>>>>>>     basis for the FieldTrip 4D reader function.
>>>>>>
>>>>>>     Best,
>>>>>>
>>>>>>     Don Rojas
>>>>>>
>>>>>>
>>>>>>     On Jul 7, 2011, at 1:26 PM, Jim Li wrote:
>>>>>>
>>>>>>>     Dear Fieldtrip developer,
>>>>>>>     Can anybody help me with this channel-repair issue I've been
>>>>>>>     having? I can provide the raw data if you need. I really
>>>>>>>     want to solve the problem, you know.
>>>>>>>     Thanks a lot,
>>>>>>>     Jim
>>>>>>>
>>>>>>>     On Thu, Jul 7, 2011 at 2:22 PM, Jim Li <megjim1 at gmail.com
>>>>>>>     <mailto:megjim1 at gmail.com>> wrote:
>>>>>>>
>>>>>>>         Hi Odelia,
>>>>>>>         Yes, I downloaded the latest version of Fieldtrip and
>>>>>>>         tried your suggestion "cfg.neighbourdist = 0.03",   but
>>>>>>>         it still weirdly uses the far-away channels like A121 to
>>>>>>>         do the channel-repair for A234 (these two channels are
>>>>>>>         24.1172cm apart!).  Here is the detail:
>>>>>>>         ---------------------------
>>>>>>>         cfg                         = [];
>>>>>>>         cfg.dataset                 = 'e,rfhp0.1Hz';
>>>>>>>         cfg.trialdef.eventtype      = 'TRIGGER';
>>>>>>>         cfg.trialdef.eventvalue     = 320;
>>>>>>>         cfg.trialdef.prestim        = 1;
>>>>>>>         cfg.trialdef.poststim       = 1;
>>>>>>>         cfg                         = ft_definetrial(cfg);
>>>>>>>         cfg.blc                     =
>>>>>>>         'yes';                              % do baseline
>>>>>>>         correction with the complete trial
>>>>>>>
>>>>>>>         cfg.blcwindow               = [-1 0];
>>>>>>>         cfg.channel                 = {'MEG'};
>>>>>>>         raw_DATA                    = ft_preprocessing(cfg);
>>>>>>>         cfg = [];
>>>>>>>         cfg.badchannel = {'A234'};
>>>>>>>         cfg.neighbourdist =  0.03;
>>>>>>>
>>>>>>>         [raw_DATA] = ft_channelrepair(cfg, raw_DATA)
>>>>>>>         ---------------------------------------------------------------------
>>>>>>>         And here is the outcome: -----------------------------
>>>>>>>
>>>>>>>         the input is raw data with 248 channels and 201 trials
>>>>>>>         repairing channel A234
>>>>>>>           using neighbour A121
>>>>>>>           using neighbour A143
>>>>>>>           using neighbour A173
>>>>>>>           using neighbour A8
>>>>>>>
>>>>>>>         repairing bad channels for trial 1
>>>>>>>         repairing bad channels for trial 2
>>>>>>>         repairing bad channels for trial 3
>>>>>>>         repairing bad channels for trial 4
>>>>>>>         repairing bad channels for trial 5
>>>>>>>         repairing bad channels for trial 6
>>>>>>>         repairing bad channels for trial 7
>>>>>>>         repairing bad channels for trial 8
>>>>>>>         repairing bad channels for trial 9
>>>>>>>         repairing bad channels for trial 10
>>>>>>>         ....
>>>>>>>         -------------------------
>>>>>>>         Thank you.
>>>>>>>         Jim
>>>>>>>
>>>>>>>     _______________________________________________
>>>>>>>
>>>>>>>     fieldtrip mailing list
>>>>>>>     fieldtrip at donders.ru.nl <mailto:fieldtrip at donders.ru.nl>
>>>>>>>     http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>>>>
>>>>>>     -----------------------
>>>>>>     Don Rojas, Ph.D.
>>>>>>     Associate Professor of Psychiatry
>>>>>>     U. of Colorado Denver Anschutz Medical Campus
>>>>>>     Director, UCD Magnetoencephalography Lab
>>>>>>     13001 E. 17th Pl F546
>>>>>>     Aurora, CO 80045
>>>>>>     303-724-4994 <tel:303-724-4994>
>>>>>>
>>>>>>
>>>>>>     _______________________________________________
>>>>>>     fieldtrip mailing list
>>>>>>     fieldtrip at donders.ru.nl <mailto:fieldtrip at donders.ru.nl>
>>>>>>     http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> fieldtrip mailing list
>>>>>> fieldtrip at donders.ru.nl <mailto:fieldtrip at donders.ru.nl>
>>>>>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>>>
>>>>> -----------------------
>>>>> Don Rojas, Ph.D.
>>>>> Associate Professor of Psychiatry
>>>>> U. of Colorado Denver Anschutz Medical Campus
>>>>> Director, UCD Magnetoencephalography Lab
>>>>> 13001 E. 17th Pl F546
>>>>> Aurora, CO 80045
>>>>> 303-724-4994
>>>>>
>>>>> _______________________________________________
>>>>> fieldtrip mailing list
>>>>> fieldtrip at donders.ru.nl <mailto:fieldtrip at donders.ru.nl>
>>>>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>>
>>>> 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 <mailto:J.Schoffelen at donders.ru.nl>
>>>> Telephone: 0031-24-3614793
>>>>
>>>>
>>>> _______________________________________________
>>>> fieldtrip mailing list
>>>> fieldtrip at donders.ru.nl
>>>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>>
>>>
>>> -- 
>>> Jörn M. Horschig
>>> PhD Student
>>> Donders Institute for Brain, Cognition and Behaviour
>>> Centre for Cognitive Neuroimaging
>>> Radboud University Nijmegen
>>> Neuronal Oscillations Group
>>>
>>> P.O. Box 9101
>>> NL-6500 HB Nijmegen
>>> The Netherlands
>>>
>>> Contact:
>>> E-Mail:jm.horschig at donders.ru.nl
>>> Tel:    +31-(0)24-36-68493
>>> Web:http://www.ru.nl/donders
>>>
>>> Visiting address:
>>> Trigon, room 2.30
>>> Kapittelweg 29
>>> NL-6525 EN Nijmegen
>>> The Netherlands
>>> _______________________________________________
>>> fieldtrip mailing list
>>> fieldtrip at donders.ru.nl <mailto:fieldtrip at donders.ru.nl>
>>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>
>> -----------------------
>> Don Rojas, Ph.D.
>> Associate Professor of Psychiatry
>> U. of Colorado Denver Anschutz Medical Campus
>> Director, UCD Magnetoencephalography Lab
>> 13001 E. 17th Pl F546
>> Aurora, CO 80045
>> 303-724-4994
>>
>> _______________________________________________
>> fieldtrip mailing list
>> fieldtrip at donders.ru.nl <mailto:fieldtrip at donders.ru.nl>
>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
> -----------------------
> Don Rojas, Ph.D.
> Associate Professor of Psychiatry
> U. of Colorado Denver Anschutz Medical Campus
> Director, UCD Magnetoencephalography Lab
> 13001 E. 17th Pl F546
> Aurora, CO 80045
> 303-724-4994
>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip


-- 
Jörn M. Horschig
PhD Student
Donders Institute for Brain, Cognition and Behaviour
Centre for Cognitive Neuroimaging
Radboud University Nijmegen
Neuronal Oscillations Group

P.O. Box 9101
NL-6500 HB Nijmegen
The Netherlands

Contact:
E-Mail: jm.horschig at donders.ru.nl
Tel:    +31-(0)24-36-68493
Web: http://www.ru.nl/donders

Visiting address:
Trigon, room 2.30
Kapittelweg 29
NL-6525 EN Nijmegen
The Netherlands

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20110709/ba2d8c9d/attachment.html>


More information about the fieldtrip mailing list