Hi Dorrit,
On Tue, May 29, 2012 at 9:24 AM, Dorrit Inbar
<[log in to unmask]> wrote:
> I have a list of questions regarding the notes I wrote during
> the course, which I would like to ask you before I present in a lab meeting
> practical insights from the course. I apologize if the questions are too
> basic, especially for a course graduate like me...
>
> 1. Number of modes in DCM:
> Are the modes like ICA? How do they actually differ from ICA?
The modes are produced by PCA of the lead fields. They are not based
on the EEG/MEG data and the idea is just to reduce the number of
channels while preserving most of topographical information, as
opposed to finding independent sources.
> And what are the relations between the modes and the lead-filed? In p. 136
> in the manual it is written: "The number of modes specifies how many
> components from the lead filed are present in channel data". Especially-
> what is the difference between the modes in Induced-responses models and
> other models? (If I recall correctly, I was told that modes have different
> meaning in IND. Is it just because they represent the time-frequency
> components of the data?)
Yes, in DCM-IND modes are something different. They are obtained by
PCA of the time-frequency data and the idea is also to reduce the data
size for computational efficiency.
>
> 2. Robust averaging:
> During the lectures I wrote a note that it is possible the reject PARTS of
> the trial instead of rejecting a whole trial. This might be useful when only
> during a certain time during a specific trials the activity is an outlier.
> How can such comparison of parts of the trials be compared to find outliers?
> (I did not manage to find it when running the "robust averaging" option).
Robust averaging rejects only parts of trials that are actually
outlying. You don't need to do anything special for this to happen,
but you can save the weights and look at them to see what was actually
rejected.
>
> 3. RFT on data with a small number of electrodes:
> Since it is not recommended to perform RFT on data with a small number of
> electrodes (I assume that 19 is a small number) since the filed isn't
> sufficiently sampled, should I first perform 2-D-reconstruction? Or is
> there a different way to perform a spatial smoothing?
I think you should try producing images and seeing if they look like
they capture the topographies that interest you. 19 electrodes stretch
the assumptions of the method but may still work. The alternative is
to do 1D tests for each electrode to find effects in time and then
correct with Bonferroni across electrodes. I posted an example script
for creating 1D images on the list recently and it will be included in
next SPM8 update. In SPM12 I hope to integrate this functionality in
the GUI. Doing 3D source reconstructions and then stats is also
possible but not sure how well it'll work with 19 channels. Depends on
your data.
>
> 4. Source names and locations in the DCM:
> Is it possible to control the size of the specified sources around the
> coordinate?
This only makes sense for imaging DCM and there is a fixed patch size
for all sources set by DCM.M.dipfit.radius (default 16 mm). I
wouldn't worry too much about size. The concept of size in M/EEG is
problematic since it's difficult to separate effects of size from
effects of amplitude and SNR.
> What is the radius around the coordinate taken as a source?
See above. With dipoles your sources are effectively point sources but
it is known that a single dipole can fit well activity generated by a
path of something like 2 cm radius.
> Is it possible to mark an ROI around an area to be defined as a source?
>
Not at the moment but I don't think it's really necessary for the reasons above.
> 5. Electromagnetic model:
> How many sources can be listed? I assumed 5 or 8, but I managed to write
> many more. Is it a limit set only by the PC's ability to perform the
> computations? Or is there a limit set by the DCM?
There is no hard limit but models with a lot of sources will be
difficult to invert.
>
> 6. Wavelet number:
> What should be the considerations when specifying the wavelet number at the
> bottom of the DCM window? Is it a function of the frequency band taken? How
> (and if) would it differ in different data models- ERP/SSR/CSD/IND/PHA ?
This parameter is only used for DCM-IND. The best way to set it is to
start with the default and then try increasing and decreasing it and
looking at the resulting TF plots to see which one looks more sensible
to you. That's what 'Wavelet transform' button is for.
Best,
Vladimir
|