[FieldTrip] scaling signals for joint EOG-MEG ICA decomposition

[Eelke Spaak]

Dear Max,

Is there a particular reason you are including the EOG and ECG data in
the ICA decomposition? The pipeline that would make most sense to me
is to do the ICA decomposition on the MEG data alone, then check the
correlation of the EOG/ECG timecourses with individual ICA components
to classify which of those likely correspond to artifacts. Doing a
joint decomposition will lead to some mixing in of EOG/ECG data into
the reconstructed (cleaned) MEG data.

Cheers,
Eelke

On 12 February 2018 at 17:44, Maximilien Chaumon
<maximilien.chaumon at gmail.com> wrote:
> Dear eeglab and FieldTrip users,
>
> On our MEG system, we systematically record EOG and EKG along with the MEG
> data. Running an ICA on the data to remove blinks and EKG works fine most of
> the time, but every now and then, the output is turned to puzzling
> complex-valued numbers.
>
> I realized that the huge scale discrepancy between the EKG and the MEG data
> could be what causes the problem, so I just multiplied all my EOG & EKG
> signals by 1e-6 and ran the ICA again, which returns real-valued numbers.
> The topographies look fine but I'm stuck trying to find out a way to scale
> this output to recover my originally scaled data. Is it possible?
>
> Any comments are welcome, thanks!
> Best,
> Max
>
>
> _______________________________________________
> fieldtrip mailing list
> fieldtrip at donders.ru.nl
> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip