Hello! I am computing a source localisation using EEG data I have collected, and the free surfer average MRI template for all participants. I have used the code below to calculate a custom volume source space before computing the forward solution. On different computers, this has produced both a different number of sources and different spacing between them in the setup_volume_source_space() function, and a different source rejection rate in make_forward_solution(). I have detailed this issue below, starting with the code I have used across all computers.
### see potentially relevant additional information below for packages imported
### load in fsaverage files
fs_dir = fetch_fsaverage(verbose=True)
subjects_dir = fs_dir[:-10] #take off the last "\fsaverage" to remove duplicate folder
# specify file locations within
subject = 'fsaverage'
trans = mne.read_trans(f'{fs_dir}\\\\bem\\fsaverage-trans.fif')
bem_file = f"{fs_dir}\\bem\\fsaverage-5120-5120-5120-bem-sol.fif"
### load in one subject's epochs
os.chdir('F:\\DNap\\EEG\\classified\\processed')
epoch_files = glob.glob("*.fif.gz")
epochs = mne.read_epochs(epoch_files[0],preload=True)
# set montage
montage = mne.channels.make_standard_montage('standard_1020')
# apply montages to epochs
epochs.set_montage(montage)
# reset reference
epochs.set_eeg_reference('average', projection=True)
### make custom source space with less voxels #1844 (divisible evenly by 4)
src = setup_volume_source_space(subject, subjects_dir=subjects_dir, bem=bem_file, pos=10, mindist=4.75) #4.75 to get an even number of sources to split later
### plot source coordinates before fwd - Figure 1 (computer 1), Figure 3 (computers 2 and 3)
coords = np.vstack([s['rr'][s['inuse'].astype(bool)] for s in src])
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.scatter(coords[:, 0], coords[:, 1], coords[:, 2], s=1, color='blue')
### compute the forward solution
fwd = mne.make_forward_solution(
epochs.info,
trans=trans,
src=src,
bem=bem_file,
eeg=True,
mindist=4.75, #min distance of sources from inner skull surface
n_jobs=None
)
### plot sources after fwd - Figure 2 (computer 1), Figure 3 (computers 2 and 3)
src_fwd = fwd['src']
coords_fwd = np.vstack([s['rr'][s['inuse'].astype(bool)] for s in src_fwd])
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
ax.scatter(coords_fwd[:, 0], coords_fwd[:, 1], coords_fwd[:, 2], s=1, color='blue')
I first ran this code on one computer (windows 10, MNE 1.7.0, python 3.12.2) and it produced the following spacing for my sources, visualised on the 3d grids below (positioned from a view slightly diagonally up and left of a sagittal plane. The right of the graph is anterior of the brain). Figure 1 is the in-use source coordinates before the forward solution, Figure 2 is the in-use source coordinates after the forward solution. The number of initial sources in-use and plotted are 807 (Figure 1), which the forward solution reduces to 567 (Figure 2). In both versions, but especially the second, there is uneven spacing between the sources with a disproportionately high number in the middle of the brain, and significant gaps throughout the neocortex.
Figure 1
Figure 2
This is the relevant output from computing the forward solution, describing the reduction in sources:
“Checking surface interior status for 807 points…
Found 456/807 points inside an interior sphere of radius 47.7 mm
Found 0/807 points outside an exterior sphere of radius 98.3 mm
Found 0/351 points outside using surface Qhull
Found 240/351 points outside using solid angles
Total 567/807 points inside the surface
Interior check completed in 109.6 ms
240 source space points omitted because they are outside the inner skull surface.”
I have run this exact code on a second computer (windows 11, MNE 1.7.0, python 3.11.3) and third computer (windows 11, MNE 1.5.1, python 3.11.5). On both the second and third computers, the number of in-use sources is 1844, which is not reduced by computing the forward solution (all sources are identified as within the inner skull surface). The in-use source coordinates when plotted look as below (Figure 3), which are the same both before and after the forward solution, and across these 2 computers.
Figure 3
I assume firstly that the first computer is calculating the source space incorrectly, and the second and third computer are doing this correctly. Secondly, I assume that this discrepancy is caused by an issue of environment difference between the computers and/or a setting that has been specified somewhere, but am unaware of what difference could be causing this.
Thank you in advance for any help or advise on how to troubleshoot this. I am relatively new to mne python and this is my first time posting a topic here 
Potentially relevant additional information:
additional information
- The packages I have imported at the top of the script are:
import mne
import os
import os.path as op
import glob
from mne.datasets import eegbci, fetch_fsaverage
from mne.beamformer import apply_lcmv_epochs, make_lcmv
import numpy as np
import matplotlib.pyplot as plt
from mne import setup_volume_source_space, write_forward_solution
from mne import read_labels_from_annot
from nilearn.plotting import plot_glass_brain, plot_markers
%matplotlib qt
- I have already completely deleted and redownloaded the fsaverage data on both computers, in case there was a problem with the downloaded files. This has not resolved the issue.
- the issue replicates no matter what I set the pos and mindist arguments to when setting up the source space, only the number of sources changes.
- I have tried loading the forward solution and src calculated on all computers onto the other computers, and this does not change the spacing of the sources, so it is not a plotting issue on one computer.
- I use spyder as my python IDE.
- I have not updated MNE to the newest version, as this breaks my code in a few places before I even reach the above issue. I can probably fix these issues if someone identifies the version to be the problem, but I see no reason to do this otherwise.
- One of my supervisors suggested that the difference may be due to a background setting specifying something to do with including sources based on whether they are in grey or white matter. I have not been able to find such a setting.
