Improved functions to get leadfields and LCMV weights

[cgohil8]

Changes:

• Added a function to extract LCMV beamformer weights.
• Simplified the function to get leadfields.
• Improved docs for these functions.

Tested the beamforming projection with:

import mne
import numpy as np
from osl_ephys.source_recon import beamforming

subjects_dir = "data/raw/src"
subject = "sub-CC320608_task-rest"
preproc_file = "data/raw/preproc/mf2pt2_sub-CC320608_ses-rest_task-rest_meg/mf2pt2_sub-CC320608_ses-rest_task-rest_meg_preproc_raw.fif"

raw = mne.io.read_raw_fif(preproc_file, preload=True)
raw = raw.filter(l_freq=1, h_freq=45, method="iir", iir_params={"order": 5, "ftype": "butter"})
sensor_data = raw.get_data(picks="meg", reject_by_annotation="omit")
sensor_cov = np.cov(sensor_data)

voxel_data, _ = beamforming.voxel_timeseries(
    subjects_dir=subjects_dir,
    subject=subject,
    preproc_file=preproc_file,
    chantypes="meg",
    freq_range=[1, 45],
    reject_by_annotation="omit",
)
voxel_cov = np.cov(voxel_data)

weights, _ = beamforming.get_lcmv_weights(subjects_dir, subject)  # projects from sensor to MNI voxel space
weights_inv = np.linalg.pinv(weights)

print("Going forward")
test_cov = weights @ sensor_cov @ weights.T
print(voxel_cov)
print(test_cov)
print()

print("Going backward")
test_cov = weights_inv @ voxel_cov @ weights_inv.T
print(sensor_cov)
print(test_cov)

The covariances going forward match. The covariances going backwards don't because you removed information (power, i.e. variance) when you applied the beamformer to the sensor data.