A mesostate-space model for EEG and MEG.

Area of Science:

• Computational Neuroscience
• Biophysics
• Signal Processing

Background:

• Electroencephalography (EEG) is crucial for understanding brain function.
• Existing models often oversimplify or overcomplicate the spatiotemporal dynamics of neural activity.
• Bridging the gap between distributed source models and equivalent current dipole models is needed.

Purpose of the Study:

• To introduce a novel multi-scale generative model for EEG data.
• To model brain activity using a limited number of dynamic 'mesostates'.
• To provide a framework for inferring mesostate dynamics and their functional connectivity.

Main Methods:

• Developed a multi-scale generative model for EEG.
• Employed a Variational Bayesian learning scheme for parameter inference and model evidence calculation.
• Introduced a mesostate-space representation of brain activity dynamics.
• Integrated stochastic dynamical causal models as priors for mesostate evolution.

Main Results:

• The model successfully captures the spatiotemporal dynamics of EEG signals with minimal assumptions.
• Model evidence allows for determining the optimal number of mesostates.
• Posterior probability maps reveal dipole activity within mesostates.
• The model demonstrates added value compared to standard inverse EEG techniques on synthetic and real data.

Conclusions:

• The proposed mesostate-space model offers a flexible yet constrained approach to EEG source analysis.
• It effectively bridges the gap between distributed and dipole models.
• This framework enhances the understanding of functional brain connectivity and dynamics.