LORETA to Fieldtrip

[Robert Oostenveld]

On 27 Mar 2006, at 16:51, Vladimir Litvak wrote:
> I thought about this somewhat more and there are some things I
> realized and
> some new questions that came up:
>
> 1) Why do I need the Collin brain if I don't plot the results in
> fieldtrip?
> A much simpler way would be to create a grid according to the
> maximal values
> of coordinates in LORETA (there is a list). Then I don't need any
> transforms
> - I just fill this grid directly using the MNI coordinates (plus some
> constant). Then after the analysis I can simply export the result
> back to
> LORETA.

You can use the collin brain to learn how volumes are represented in
FT (and in SPM for that matter): with the transformation matrix.
Furthermore, you can map the LORETA data onto the collin brain to
check whether you have interpreted the LORETA 3D array correctly. It
could be that in LORETA part of the brain has been cut off, which
would mess up the matching, that is something which you need to be
aware of.

Spatial clustering in FT requires that you specify the shape of the
3D array correctly. The best way of knowing that it is correct is by
plotting it in FT. FT sourceplotting also has options for combining
anatomy+functional+statistics in one display (using grey+color
+tranparency).

Why not ask around whether someone has read it into Matlab before. I
suspect that there is a mailing list, or you can ask Roberto Pasqual-
Marqui directly.

> 2) Does sourcestatistics handle multiple timeframes i.e. does
> clustering
> both in time and in space the same way as with scalpmaps? If yes,
> how do I
> specify the input for this? I haven't tried yet but I suspect that
> would be
> a pretty heavy operation to run.

No, sofar I have not implemented Xvox-Yvox-Zvox-freq-time clustering,
only Xvox-Yvox-Zvox clustering. We typically do not look at multiple
frequencies or latencies, since on the channel level we can already
decide (channel statistics) which time-frequency "tile" is
significant. Subsequently we use multitapering on exactly that time-
frequency tile to capture the complete effect in one TF-bin (with the
desired width in both dimensions) and beam that single bin. So the
standard way to deal with source data is that it is 3-D only.
Actually it would not be too hard to implement the clustering in 4-D
or 5-D, although I have no idea on the computational demands. I
suspect it to be computationally feasible, since it uses the Matlab
image processing toolbox, which is very efficient. For your LORETA
data, the 4-D clustering does seem to be interesting so I could look
at implementing it for N-D. The channel clustering is much more
difficult, since the channel locations do not "code" a nice natural
dimension in the data but  they have a weird neighbourhood structure.

Robert