[FieldTrip] ICA on elekta data

[Martin Luessi]

Hi Craig,

What you could do to handle the different units is whiten the data using 
a noise covariance. I.e., you collect empty-room noise, compute a noise 
covariance "C" and then pre-multiply the data with "C^{-1/2}".

This method is used in the MNE software to combine different sensor 
types (gradiometers, magnetometers, eeg) for ICA, minimum norm solutions 
etc. The ICA code is here:

https://github.com/mne-tools/mne-python/blob/master/mne/preprocessing/ica.py

The code is in Python, but it should be straight-forward to translate 
parts to Matlab. If you don't want to implement it yourself, you could 
also use MNE to clean the raw data, save it back as a  raw .fif file and 
then continue your analysis in FT. There is an example here:

http://martinos.org/mne/auto_examples/preprocessing/plot_ica_from_raw.html#example-preprocessing-plot-ica-from-raw-py

I hope this helps,

Martin

PS: Robert, I hope you don't mind that I advertise MNE on this list ;). 
I felt that it was appropriate since whitening is very useful when 
working with an Elekta system.


On 10/01/13 04:03, Robert Oostenveld wrote:
> Hi Craig,
>
> I am CCing the FT mailing list, as there is a lot of expertise there as well (as well as interest in the answer probably).
>
> You would for sure combine the gradiometers. They have their signals expressed in the same units (fieldtrength/distance). But you would probably also benefit from including the magnetometers, as they reflect the same sources that linearly mix on the channel level. However, the magnetometers are expressed in another unit (fieldstrength) which results in them being potentially weighted differently. E.g. imagine that you express distance in mm, then the planar signals are 1000 times smaller than if you express distance in meter, whereas the magnetometer signals (the 102 rows of the same data matrix) don't change.
>
> If you were to do PCA, it would make a large difference since the channel-level variances are hugely different. But I think that if you do ICA, the "sphering" step that is done prior to the unmixing matrix estimation takes care of the different scaling and it would not matter if you were to to it in meter or milimeter. As I don't have the hands-on experience, it is something I would explicitly check by doing the ICA decomposition twice on differently scaled gradiometer data (keeping the magnetometer rows in the data the same).
>
> best regards,
> Robert
>
>
>
>
>
> On 30 Sep 2013, at 15:33, Richter, Craig wrote:
>
>> Hi Robert,
>>
>> How would you approach ICA on data from the Elekta system?  I have the 102 chips with 1 magnetometer, and two orthogonal planar gradiometers, as you know.  It seems to me that the magnetometers should be done separately, but how would you handle the gradiometers?  Do these need to be combined?
>>
>> I wanted to see how you would proceed.
>>
>> Best,
>>
>> C.
>>
>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>
>


-- 
Martin Luessi, Ph.D.

Research Fellow

Department of Radiology
Athinoula A. Martinos Center for Biomedical Imaging
Massachusetts General Hospital
Harvard Medical School
149 13th Street
Charlestown, MA 02129

Fax: +1 617 726-7422


The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.