Source Analysis with DIPFIT

Equivalent-dipole source localization of ICA components with EEGPrep DIPFIT. Mirrors EEGLAB’s “Source analysis” tutorial workflow: choose a head model, coarse-fit component dipoles on a grid, refine them, then read the resulting EEG.dipfit.model entries.

EEGPrep’s bundled DIPFIT backend is a single-sphere analytic leadfield, so positions are approximate and will not match FieldTrip BEM coordinates.

channels=32 trials=80 components=32
EEG = pop_dipfit_settings(EEG, model='standardBESA')
coordformat=Spherical hdmfile=standard_BESA/standard_BESA.mat
EEG = pop_dipfit_gridsearch(EEG, [1, 2, 3], [-85, -68, -51, -34, -17, 0, 17, 34, 51, 68, 85], [-
EEG = pop_dipfit_nonlinear(EEG, component=1, nonlinear='yes')
EEG = pop_multifit(EEG, [1, 2, 3], threshold=40, rmout='off', dipoles=1, dipplot='off')
IC1: posxyz=(  22.8,   -0.3,  -13.4) rv=2.1%
IC2: posxyz=(  -5.3,  -22.3,    1.1) rv=4.4%
IC3: posxyz=(  28.6,    0.7,  -23.9) rv=0.3%
pop_dipplot(EEG, [1, 2, 3], cornermri='on')

import matplotlib

matplotlib.use("Agg")

from pathlib import Path

import numpy as np

import eegprep
from eegprep import (
    pop_dipfit_gridsearch,
    pop_dipfit_nonlinear,
    pop_dipfit_settings,
    pop_dipplot,
    pop_loadset,
    pop_multifit,
)

# sphinx-gallery defines no __file__, so anchor on the installed package.
REPO_ROOT = Path(eegprep.__file__).resolve().parents[2]
SAMPLE = REPO_ROOT / "sample_data" / "eeglab_data_epochs_ica.set"

# 1. Load a dataset that already carries channel locations and an ICA decomposition.
EEG = pop_loadset(str(SAMPLE))
print(f"channels={EEG['nbchan']} trials={EEG['trials']} components={EEG['icaweights'].shape[0]}")

# 2. Head model and settings (Tools > Source localization using DIPFIT > Head model and settings).
EEG, com = pop_dipfit_settings(EEG, model="standardBESA", return_com=True)
print(com)
print(f"coordformat={EEG['dipfit']['coordformat']} hdmfile={EEG['dipfit']['hdmfile']}")

# 3. Coarse fit on a grid, then fine (nonlinear) fit of component 1.
EEG, com = pop_dipfit_gridsearch(EEG, [1, 2, 3], gui=False, return_com=True)
print(com[:96])
EEG, com = pop_dipfit_nonlinear(EEG, 1, gui=False, return_com=True)
print(com)

# 4. Autofit components 1-3 and drop fits above 40% residual variance.
EEG, com = pop_multifit(EEG, [1, 2, 3], threshold=40, gui=False, return_com=True)
print(com)

for index, model in enumerate(EEG["dipfit"]["model"][:3], start=1):
    pos = np.atleast_2d(np.asarray(model["posxyz"], dtype=float))
    rv = float(np.asarray(model["rv"], dtype=float).ravel()[0])
    coords = "; ".join(f"({x:6.1f}, {y:6.1f}, {z:6.1f})" for x, y, z in pos)
    print(f"IC{index}: posxyz={coords} rv={100 * rv:.1f}%")

# 5. Dipole plot over the packaged MNI MRI slices (Component dipole plot).
_, com = pop_dipplot(EEG, [1, 2, 3], gui=False, cornermri="on", return_com=True)
print(com)