[Mne_analysis] Lead field spikes using sample BEM - bug or explainable?

[Taylor Williams]

        External Email - Use Caution        

I've updated the shared folder to include a binary version of the Info and
src objects used to create this forward model. To create the forward model,
you can use the following:

import mne
from mne.datasets import sample
import pickle

bem_fname = sample.data_path() +
'/subjects/sample/bem/sample-5120-5120-5120-bem-sol.fif'
trans_fname = sample.data_path() + '/MEG/sample/sample_audvis_raw-trans.fif'

#Compute src for ico4 spacing from scratch
src = mne.setup_source_space('sample', spacing='ico4')
#Or, read src from file, pre-calculated using the command above
#src = pickle.load(open('/LOCAL_PATH_HERE/ico4_src.fif','rb'))

info =
pickle.load(open('/LOCAL_PATH_HERE/sample_bem_1632mag_info.pkl','rb'))

fwd = mne.make_forward_solution(info, trans_fname, src, bem_fname)

On Fri, Feb 7, 2020 at 3:58 PM Taylor Williams <williams.taylor at gmail.com>
wrote:

> Eric,
>
> Please find the forward model and script for the 1632 magnetometer array
> in the following folder. Script is also attached.
>
> https://drive.google.com/drive/folders/1dX7YZSCa4RAcn_b4JH3I0CADe2ZOQIM-?usp=sharing
>
> A bit more context on the forward model: It places point magnetometers on
> 102 simulated rectangular 24mm x 24mm chips that are projected towards the
> scalp from the standard Neuromag sensor locations. The center point of each
> sensor cluster is 1.5mm above the nearest outer skin surface from the BEM.
> All sensors are sensitive in the z-direction and preserve the orientation
> vector from the derived Neuromag magnetometer.
>
> This script generates the full set of lead fields in the Evoked object Y0.
> There are plenty of other instances where this occurs, not just dipole
> #364. Let me know if you have any questions about the construction of the
> forward model that I can answer for context. I'm curious to see what you
> think.
>
> Thanks,
> Taylor
>
> On Thu, Feb 6, 2020 at 4:12 PM Eric Larson <larson.eric.d at gmail.com>
> wrote:
>
>>         External Email - Use Caution
>>
>> I'm not sure offhand what would cause this. Can you share a script to
>> produce the bottom plot in the source 364, 1632-point-magnetometer case?
>> That would help us look into what specifically what (at least: where in the
>> code) is causing the value to become so large.
>>
>> Eric
>>
>>
>> On Thu, Feb 6, 2020 at 3:51 PM Taylor Williams <williams.taylor at gmail.com>
>> wrote:
>>
>>>         External Email - Use Caution
>>>
>>> I've found something unusual in the lead fields computed using
>>> MNE-Python that I can't explain. When arrays containing a large number of
>>> magnetometer sensors are used with the 'sample' subject BEM
>>> (sample-5120-5120-5120-bem.fif) and the ico4 source space to create a
>>> forward model, there are some unusual strong spikes in the lead field.
>>> These spikes are not present when I use a spherical BEM model instead.
>>>
>>> To demonstrate, I created three different theoretical arrays of point
>>> magnetometers (coil_id: 2000) measuring roughly the normal component of the
>>> magnetic field with respect to the scalp surface. Each model contains an
>>> increasing number of total sensors (102 sensors, 1632 sensors, and 3672
>>> sensors).
>>>
>>> Figure 1 shows the lead field for a particular dipole (#364 using ico4
>>> source space) for each array using a spherical BEM model. A smooth dipolar
>>> response is shown for each that is roughly equivalent over the arrays. The
>>> peak response for each array doesn't change appreciably, but there is an
>>> improvement in detail of the lead field. All evoked topomap plots are fixed
>>> to the same color bar limits (400 fT).
>>>
>>> Figure 2 shows the same set of lead fields but with the 'sample' subject
>>> BEM instead. It shows curiously strong spikes present in the two high
>>> density arrays. These spikes vary in their location and magnitude. The main
>>> dipolar response is still present, and appears to be fairly consistent.
>>> However, due to the spikes, the peak response for each array is vastly
>>> different. Again, evoked topomap plots use a fixed color bar limit at 400
>>> fT.
>>>
>>> Are these lead field patterns in Figure 2 explainable? They certainly
>>> seem like erroneous values, but perhaps something about the BEM model
>>> creates a focusing effect for the magnetic field response when we sample at
>>> different spatial frequencies? I'm happy to provide some sample forward
>>> models for inspection.
>>>
>>> Thanks,
>>> Taylor Williams
>>> _______________________________________________
>>> Mne_analysis mailing list
>>> Mne_analysis at nmr.mgh.harvard.edu
>>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>>
>> _______________________________________________
>> Mne_analysis mailing list
>> Mne_analysis at nmr.mgh.harvard.edu
>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20200212/00666088/attachment.html