Dear Tom,
The problem you found with EEG BEM is a known limitation of the method
used by the BEM code of Christophe Phillips. Vertices which are closer
to the inner skull surface than the edge length of that surface cannot
be handled correctly. There are just a few such vertices in the
standard template mesh and it doesn't seem to affect the SPM source
reconstruction methods (although may well affect other methods using
the same BEM). Also when transforming the meshes to fit individual
MRIs the problem might be aggravated but I haven't seen any examples
of that.
There is a more precise method implemented in the OpenMEEG library
(http://www-sop.inria.fr/athena/software/OpenMEEG/). This library is
supported by spm_eeg_inv_forward (as 'OpenMEEG BEM') but the option is
not in the GUI. The reason for this is that OpenMEEG is not trivial to
install and rather slow to run (e.g. it might take more than a day to
compute one set of leadfields). Therefore it is not practical for most
SPM users. However if you are very worried about EEG leadfield quality
this might be a good method for you. Also note that you can replace
the gain matrix file generated by SPM with your own file computed with
external software as long as the leadfields match.
For SPM12 we are planning to add the possibility to use pre-computed
leadfields stored as images. Then the pre-computing can be done with
OpenMEEG but the installation and slow running problems will be
circumvented. This will only work, however, for the standard EEG
template head and users who want to warp the meshes will have to run
OpenMEEG themselves.
Regarding your second issue, note that the mesh was computed from the
individual image stored in canonica/single_subj_T1.nii so you should
check for the correspondence with that image and not with the average
template which might be slightly different.
Best,
Vladimir
On Sun, Nov 13, 2011 at 10:24 PM, Rik Henson
<[log in to unmask]> wrote:
> ________________________________
> From: tom h [[log in to unmask]]
> Sent: 11 November 2011 21:43
> To: [log in to unmask]
> Cc: Rik Henson; [log in to unmask]
> Subject: Re: M+EEG+fMRI Face Dataset
>
> Hello Rik, SPM methods group,
> I'm interested in creating EEG lead fields, and have been trying SPM8 for
> this purpose. I noticed two curious features, perhaps you can comment on
> them -- I might have done some things incorrectly.
> 1. In this
> paper http://related.springerprotocols.com/lp/pubmed-central/eeg-and-meg-data-analysis-in-spm8-HDecQN27Jc
> you recommend using BEM. However, my BEM lead fields (and those created by
> some others in SPM8) all have the following property: some voxels have an
> order of magnitude higher norm than the rest. By norm I mean mean(G.^2,1),
> where G is the sensors X voxels lead field matrix. On the other hand, my
> 3-sphere lead fields created with SPM don't show this property.
> I'm attaching a figure (lead_field_norm) showing the norm of a BEM and a
> 3-sphere lead field created for the same 60-sensor system. I'm also
> attaching a mat file with the 2 lead field matrices and the sensor labels,
> in case you'd like to take a look.
> My question is, is the BEM result incorrect, and if so, what would you
> suggest doing to create an accurate BEM lead field?
> 2. I overlaid the voxels of my 3-sphere lead fields on the T1 MRI template.
> However, there seems to be some mismatch. I'm attaching 2 figures
> (spm_5124voxels_1,2) that show the voxels closest to each of the 91 surfaces
> along the x-axis (except the extreme ones that don't contain voxels). There
> seem to be some voxels in places you wouldn't expect, and the coverage is
> not quite uniform.
> For comparison, I'm attaching the same figures for a non-SPM lead field
> (nonspm_2400voxels_1,2). These voxels (admittedly there are about 1/2 as
> many as in the SPM case) seems to line up better.
> My question here is, am I correct in assuming that the voxel locations in
> SPM-created lead fields are given in MNI coordinates, and do these overlays
> indicate to you there was an error made using the software?
> I realize this is a long email, and will truly appreciate any advice you'd
> be able to offer. Overall I find SPM very helpful and enjoy using it.
> Have a good weekend!
> Best,
> Tom
>
>
> On Thu, Mar 10, 2011 at 11:23 AM, tom h <[log in to unmask]> wrote:
>>
>> Indeed I was using Chrome (doesn't everyone? :-)
>> Your solution works, thank you!
>> On Thu, Mar 10, 2011 at 3:12 AM, Guillaume Flandin
>> <[log in to unmask]> wrote:
>>>
>>> Hi Tom,
>>>
>>> I assume you are using Google Chrome, which seems to have trouble
>>> accessing some FTP servers. Try again with any other web browser or use
>>> this HTTP link:
>>> http://www.fil.ion.ucl.ac.uk/spm/ftp/data/mmfaces/
>>>
>>> Best regards,
>>> Guillaume.
>>>
>>>
>>> On 09/03/11 22:03, tom h wrote:
>>> > Hello, Rik, SPM Methods Group,
>>> >
>>> > I'm interested in the multimodal face
>>> > dataset http://www.fil.ion.ucl.ac.uk/spm/data/mmfaces/
>>> >
>>> > However, the data links don't work. Any advice?
>>> >
>>> > Thanks for your help!
>>> >
>>> > Best,
>>> > Tom Holdman
>>> >
>>>
>>>
>>> --
>>> Guillaume Flandin, PhD
>>> Wellcome Trust Centre for Neuroimaging
>>> University College London
>>> 12 Queen Square
>>> London WC1N 3BG
>>
>
>
|