Independent Component Analysis (ICA)

Many M/EEG signals including biological artifacts reflect non-Gaussian processes. Therefore PCA-based artifact rejection will likely perform worse at separating the signal from noise sources. MNE-Python supports identifying artifacts and latent components using temporal ICA. MNE-Python implements the mne.preprocessing.ICA class that facilitates applying ICA to MEG and EEG data. It supports FastICA, the infomax, and the extended informax algorithm. It allows whitening the data using a fast randomized PCA algorithmd. Furthermore, multiple sensor types are supported by pre-whitening / rescaling. Bad data segments can be excluded from the model fitting by reject parameter in mne.preprocessing.ICA.fit.

For convenience, mne.preprocessing.ICA implements methods for

Concepts

ICA finds directions in the feature space corresponding to projections with high non-Gaussianity.

  • not necessarily orthogonal in the original feature space, but orthogonal in the whitened feature space.

  • In contrast, PCA finds orthogonal directions in the raw feature space that correspond to directions accounting for maximum variance.

  • or differently, if data only reflect Gaussian processes ICA and PCA are equivalent.

Example: Imagine 3 instruments playing simultaneously and 3 microphones recording mixed signals. ICA can be used to recover the sources ie. what is played by each instrument.

ICA employs a very simple model: $X = AS$ where $X$ is our observations, $A$ is the mixing matrix and $S$ is the vector of independent (latent) sources.

The challenge is to recover A and S from X.

First generate simulated data

import numpy as np
import matplotlib.pyplot as plt
from scipy import signal

from sklearn.decomposition import FastICA, PCA

np.random.seed(0)  # set seed for reproducible results
n_samples = 2000
time = np.linspace(0, 8, n_samples)

s1 = np.sin(2 * time)  # Signal 1 : sinusoidal signal
s2 = np.sign(np.sin(3 * time))  # Signal 2 : square signal
s3 = signal.sawtooth(2 * np.pi * time)  # Signal 3: sawtooth signal

S = np.c_[s1, s2, s3]
S += 0.2 * np.random.normal(size=S.shape)  # Add noise

S /= S.std(axis=0)  # Standardize data
# Mix data
A = np.array([[1, 1, 1], [0.5, 2, 1.0], [1.5, 1.0, 2.0]])  # Mixing matrix
X = np.dot(S, A.T)  # Generate observations

Now try to recover the sources

# compute ICA
ica = FastICA(n_components=3)
S_ = ica.fit_transform(X)  # Get the estimated sources
A_ = ica.mixing_  # Get estimated mixing matrix

# compute PCA
pca = PCA(n_components=3)
H = pca.fit_transform(X)  # estimate PCA sources

plt.figure(figsize=(9, 6))

models = [X, S, S_, H]
names = ['Observations (mixed signal)',
         'True Sources',
         'ICA estimated sources',
         'PCA estimated sources']
colors = ['red', 'steelblue', 'orange']

for ii, (model, name) in enumerate(zip(models, names), 1):
    plt.subplot(4, 1, ii)
    plt.title(name)
    for sig, color in zip(model.T, colors):
        plt.plot(sig, color=color)

plt.tight_layout()
plt.show()
../../_images/ICA_primer.png

\(\rightarrow\) PCA fails at recovering our “instruments” since the related signals reflect non-Gaussian processes.