Alarmingly similar results for subcortical source activity

✉️ Mailing List Archive (read-only)
1

Hello,

I've been looking at right amygdala activity in the spm faces dataset as a
means of developing subcortical source localization tools. I've tried two
different methods so far.

1.) I did a whole brain volume analysis (dSPM) and then averaged together
vertices belonging to the right amygdala (code here
<https://github.com/leggitta/PythonEEG/blob/master/spm_volume.py>).

2.) I merged the left and right cortical surface source spaces with a right
amygdala volume source space, performed dSPM and then averaged together the
volume sources (code here
<https://github.com/leggitta/mne-python/blob/gsoc-subcortical/examples/inverse/plot_subcortical_activation.py>).

I've attached the evoked plots for each condition (grey bars indicate
regions of significant difference using independent t-tests per time point
with fdr correction).

The results of each method are so similar that I'm concerned they may be
derived from a common source of error. Can anyone think of think of any
probable sources of error and suggest other ways to validate these methods?
I haven't yet ventured into simulated data, but I think that's going to be
the next step.

Thanks!
Alan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140602/375bb7ce/attachment.html
-------------- next part --------------
A non-text attachment was scrubbed...
Name: merge.png
Type: image/png
Size: 60443 bytes
Desc: not available
Url : http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140602/375bb7ce/attachment.png
-------------- next part --------------
A non-text attachment was scrubbed...
Name: wholebrain.png
Type: image/png
Size: 60746 bytes
Desc: not available
Url : http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20140602/375bb7ce/attachment-0001.png

2

Hi Alan,

I don't see any obvious mistake in your scripts and the results
don't surprise me much due to field spread.

The line that bothers me is:

forward = mne.convert_forward_solution(forward, surf_ori=True)

There should not be any source orientation constraint for
deep structures.

Simulation is indeed the way to go now I would say.

Alex

3

Hi Alan,

I am not sure why you would expect to get something different for the two
cases. We have compared volume based analyses with cortex only analyses
and they are nearly identical (as they should be assuming little of the
sensor signal is coming from non-cortical sources). Thus, I would agree
with Alex, the key now is simulations. The critical questions are:

1. if there is a source in the amygdala, but not on the cortex, can you
see it and how much cortical bleed is there?

2. If there is a source nearby in the cortex, but not in the amygdala, how
much bleed into the amygdala is there?

Hope that is helpful.

Best wishes,
Avniel

4

Hi Junpeng,

Yes, the function mne.source_space.get_segment_positions in on my current
development branch.

https://github.com/leggitta/mne-python/tree/gsoc-subcortical

I modified the file /mne/source_space.py to include this additional
function.

I think you can checkout my development branch and then you should be able
to run /mne/examples/inverse/plot_subcortical_activation.py to get the same
results.

Cheers,
Alan

5

Thanks Alex and Avniel,

I'll get started on trying to generate simulated data.

Alex,

I think I was getting an error when I didn't include the line

forward = mne.convert_forward_solution(forward, surf_ori=True)

But I can dig deeper into that.

-Alan