Brain Electric Model

The Brain Electric Model is a mathematical and physical framework used to describe how neural activity in the brain generates electric potentials that can be measured on the scalp (EEG) or intracranially (ECoG, sEEG).

This model is essential for understanding source localization, forward modeling, and inverse problem solving in EEG/MEG analysis. Below is a detailed explanation of the Brain Electric Model, including its principles, forward problem, inverse problem, and applications.

The Brain Electric Model describes how ionic currents in neurons (primarily pyramidal cells) generate electric potentials that propagate through the brain, cerebrospinal fluid (CSF), skull, and scalp.

It is based on Maxwell’s equations and the volume conductor theory, which models the head as a conductive medium with different tissue properties.

Key Concepts

Concept Description
Neural Sources Currents generated by pyramidal neurons in the cortex.
Volume Conduction Propagation of electric potentials through brain tissues to the scalp.
Forward Problem Calculating the scalp EEG given a known source configuration.
Inverse Problem Estimating the location and strength of neural sources from scalp EEG/MEG data.
Head Models Simplified representations of the head’s geometry and electrical properties (e.g., 3-layer sphere model, realistic head models).

Physics of the Brain Electric Model

Neural Sources

  • Pyramidal Neurons: The primary source of EEG signals is the synchronous activity of apical dendrites in pyramidal neurons.
  • Current Dipoles: The synchronous activity can be modeled as a current dipole, which has:
    • Location: Position in the brain.
    • Orientation: Direction of the current flow.
    • Strength: Magnitude of the current (in microamperes).

Volume Conduction

  • The brain and surrounding tissues (CSF, skull, scalp) act as a volume conductor.
  • The electric potential \(\phi(\mathbf{r})\) at a point \(\mathbf{r}\) in the volume conductor is described by Poisson’s equation:

\[ \nabla \cdot (\sigma(\mathbf{r}) \nabla \phi(\mathbf{r})) = -I(\mathbf{r}) \]

where:
\(\sigma(\mathbf{r}\) is the conductivity of the tissue at \(\mathbf{r}\),
\(I(\mathbf{r})\) is the current source density at \(\mathbf{r}\),

For EEG, the potential at the scalp is measured, and the forward problem involves solving Poisson’s equation for a given source configuration.

Conductivity Properties

Tissue Conductivity (\(\sigma\)) Notes
Gray Matter 0.33 S/m High conductivity.
White Matter 0.14 S/m Lower conductivity than gray matter.
CSF 1.79 S/m High conductivity (acts as a short circuit).
Skull 0.0042 S/m Low conductivity (high resistivity, attenuates signals).
Scalp 0.33 S/m Similar to gray matter.

The skull’s low conductivity is a major factor in attenuating and smearing EEG signals.

Forward Problem

The forward problem involves calculating the EEG/MEG signals that would be recorded on the scalp (or intracranially) given a known configuration of neural sources and a head model.

It is a well-posed problem (a unique solution exists for a given source configuration and head model).

Steps to Solve the Forward Problem

  1. Define the Head Model:
    • Use a simplified model (e.g., 3-layer sphere model for the head: brain, skull, scalp).
    • Use a realistic head model (e.g., finite element model (FEM) or boundary element model (BEM)) based on individual MRI.
  2. Define the Source Model:
    • Current Dipole: Single or multiple dipoles representing neural activity.
    • Distributed Source Model: Continuous distribution of current sources (e.g., minimum-norm estimate).
  3. Solve Poisson’s Equation:
    • Use analytical solutions (e.g., for spherical head models).
    • Use numerical methods (e.g., FEM, BEM, finite difference method (FDM)) for realistic head models.
  4. Compute the Scalp EEG:
    • The potential at each electrode is calculated as: \(\phi(\mathbf{r}_e) = \int_V \mathbf{J}(\mathbf{r}) \cdot \mathbf{G}(\mathbf{r}, \mathbf{r}_e) \, dV\) where:
      • \(\mathbf{J}(\mathbf{r})\) is the current source density.
      • \(\mathbf{G}(\mathbf{r}, \mathbf{r}_e)\) is the lead field (Green’s function) for the head model.

Example: 3-Layer Sphere Model

The 3-layer sphere model (brain, skull, scalp) is a simplified head model where:

  • The brain and scalp are modeled as homogeneous spheres with high conductivity.
  • The skull is modeled as a thin shell with low conductivity.

The lead field for a current dipole in a spherical head model is given by:

\[ \phi(\mathbf{r}_e) = \frac{1}{4\pi\sigma} \frac{\mathbf{p} \cdot (\mathbf{r}_e - \mathbf{r}_0)}{|\mathbf{r}_e - \mathbf{r}_0|^3} \]

where:
\(\mathbf{p}\) is the current dipole moment,
\(\mathbf{r}_0\) is the dipole location,
\(\mathbf{r}_e\) is the electrode location,
\(\sigma\) is the conductivity of the brain.

Inverse Problem

The inverse problem involves estimating the location, orientation, and strength of neural sources from scalp EEG/MEG data.

It is an ill-posed problem (many possible solutions exist, and small changes in data can lead to large changes in the solution).

Challenges of the Inverse Problem

  1. Non-Uniqueness: Multiple source configurations can produce the same scalp EEG.
  2. Noise and Artifacts: EEG data is noisy and contaminated by artifacts (e.g., eye blinks, muscle activity).
  3. Volume Conduction Effects: The skull smears and attenuates the signal, making localization difficult.
  4. Computational Complexity: Realistic head models require high computational resources.

Approaches to Solve the Inverse Problem

  1. Dipole Fitting

Assumption: The EEG is generated by a small number of focal sources (e.g., dipoles).

Method:

  1. Initialization: Guess the number and initial locations of dipoles.
  2. Optimization: Use non-linear optimization (e.g., Levenberg-Marquardt algorithm) to fit the dipole parameters to the EEG data.
  3. Validation: Check the goodness-of-fit (e.g., residual variance).

Limitations:

  • Assumes few sources (may not capture distributed activity).
  • Sensitive to noise and initialization.
  1. Distributed Source Models

Assumption: The EEG is generated by a continuous distribution of current sources.

Methods:

  1. Minimum-Norm Estimation (MNE):
    • Finds the smoothest current distribution that explains the EEG data.
    • Minimizes the L2 norm of the source estimate.
    • Formula: \(\mathbf{J} = \arg\min_\mathbf{J} \left( \|\mathbf{M} - \mathbf{LJ}\|^2 + \lambda \|\mathbf{J}\|^2 \right)\) where:
      • \(\mathbf{M}\) is the measured EEG data.
      • \(\mathbf{L}\) is the lead field matrix.
      • \(\mathbf{J}\) is the source current density.
      • \(\lambda\) is the regularization parameter.
  2. Low-Resolution Brain Electromagnetic Tomography (LORETA):
    • A variant of MNE that localizes sources in depth by imposing smoothness constraints.
  3. Weighted Minimum-Norm (wMNE):
    • Applies depth weighting to reduce bias toward superficial sources.
  4. Bayesian Methods (e.g., sLORETA, eLORETA):
    • Use Bayesian inference to estimate source distributions.
    • Provide statistical significance for source localization.

Advantages

  • Captures distributed activity.
  • Does not require a priori assumptions about the number of sources.

Limitations:

  • Low spatial resolution (blurred images).
  • Sensitive to noise and regularization parameters.
  1. Beamforming

Assumption: The EEG is generated by multiple independent sources.

Method:

  1. Construct a spatial filter (e.g., Linearly Constrained Minimum Variance (LCMV) beamformer) to pass signals from a specific location while suppressing others.
  2. Scan the brain with the filter to estimate source activity at each location.

Advantages:

  • High spatial resolution.
  • Robust to noise.

Limitations:

  • Assumes uncorrelated sources.
  • Computationally intensive.
  1. Machine Learning Approaches

Use deep learning to map EEG patterns to source locations.

Examples:

  • Convolutional Neural Networks (CNNs) for source localization.
  • Autoencoders for feature extraction.

Example:

  1. Input: Scalp EEG data \(\mathbf{M}\) and lead field matrix \(\mathbf{L}\).
  2. Regularization: Choose \(\lambda\) to balance data fitting and smoothness.
  3. Solve: \(\mathbf{J} = (\mathbf{L}^T \mathbf{L} + \lambda \mathbf{I})^{-1} \mathbf{L}^T \mathbf{M}\)
  4. Output: Source current density \(\mathbf{J}\) at each brain location.

Head Models

Simplified Models

  1. 3-Layer Sphere Model:
    • Brain, skull, and scalp modeled as concentric spheres.
    • Analytical solution for Poisson’s equation.
    • Limitations: Poor representation of realistic head geometry.
  2. Multishell Spherical Model:
  • More shells (e.g., CSF, gray matter, white matter) for better accuracy.

Realistic Head Models

  1. Finite Element Model (FEM):
    • Tetrahedral mesh of the head based on individual MRI.
    • Accurate conductivity assignment to each tissue type.
    • High computational cost.
  2. Boundary Element Model (BEM):
    • Surface-based model (e.g., scalp, skull, brain).
    • Lower computational cost than FEM but less accurate for deep sources.
  3. Finite Difference Model (FDM):
    • Cartesian grid for the head.
    • Easier to implement but less accurate for complex geometries.

Applications of the Brain Electric Model

Application Description
Source Localization Identifying epileptic foci, eloquent cortex, or functional brain networks.
EEG/MEG Analysis Interpreting brain activity from scalp recordings.
Brain-Computer Interfaces (BCIs) Using source estimates for neural decoding (e.g., motor imagery).
Clinical Diagnostics Assessing brain function in disorders like epilepsy, Alzheimer’s, or stroke.
Neurostimulation Guiding transcranial magnetic stimulation (TMS) or deep brain stimulation (DBS).
Cognitive Neuroscience Studying brain connectivity and network dynamics.

Forward vs. Inverse Problem

Aspect Forward Problem Inverse Problem
Definition Calculate EEG/MEG from known sources. Estimate sources from EEG/MEG data.
Well-Posedness Well-posed (unique solution). Ill-posed (many solutions, unstable).
Input Source configuration, head model. EEG/MEG data.
Output Scalp EEG/MEG signals. Source locations and strengths.
Methods Analytical/numerical solutions. Dipole fitting, MNE, beamforming, ML.
Challenges Computational complexity (realistic models). Non-uniqueness, noise, volume conduction.
Applications Simulations, forward modeling. Source localization, diagnostics.