channel location file to import planar to EEGLAB

[Robert Oostenveld]

Hi Theresa

On 22 Jun 2006, at 1:27, Montez, T. wrote:
> I am using EEGLAB to compute ICA of CTF 151 channels data.
>
> I used to read the data in EEGLAB directly from the CTF format,
> but some components are affected by the bad channels.
The bad channels do not reflect a linear mixture of underlying
sources, therefore you should remove them prior to ICA. You should
also remove all trials containing anomalous artifacts.
> Since removing the bad channels will reduce the number of components
> and will give problems later at the planar transformation;
> I tried using megrepair first and then importing the matlab array
> of data.trial{1}. I use the channel locations of a previous
> dataset opened from the CTF format.
MEGREPAIR uses a nearest-neighbour approach for fixing the missing
channels. The "repaired" channels are linear combinations (i.e. a
linear mixture) of the neighbours and hence do not add any
information (i.e. components) to the data, doing ICA on the repaired
dataset does not give you all 151 components.
> - I still get some components reflecting the bad channels,
> should I increase the cfg.neighbourdist?
> (I was using the default of 4 cm).
>
> - To import the data after the planar transformation I need
> a channel location file for the dH and dV pairs,
> can you help me with that?
The planar transformation in MEGPLANAR also results in a linear
estimation of the horizontal and vertical gradients, i.e. if you
start with 151 axial channels and you do MEGPLANAR, you will have 302
planar channels, but the dimensionality of that data will still be
151. That means that ICA on the 302-channel planar data will result
in a maximum of 151 components.

I suggest to do the ICA on the axial data with the bad channels
removed. Subsequently you can switch from EEGLAB to FieldTrip and
compute the planar representation of the ICA components using the
MEGPLANAR function (I am not sure whether megplanar knows how to deal
with ICA data, but it is easy to convert the ICA data into something
that resembles an average ERF or alternatively to add support for the
ICA data in megplanar).

best regards,
Robert