Variational Bayesian inversion of the equivalent current dipole model in EEG/MEG

Area of Science:

• Neuroscience
• Biophysics
• Computational Neuroscience

Background:

• Magneto- and electroencephalography (M/EEG) require spatial modeling for brain activity inference.
• Current source reconstruction methods include point dipole and distributed source models.
• Bayesian inversion schemes are well-developed for distributed models but less so for dipoles.

Purpose of the Study:

• To propose a fast variational Bayesian algorithm for inverting dipole models in M/EEG.
• To enable Bayesian confidence intervals for dipole parameters and model comparison via evidence.
• To improve the accuracy and flexibility of M/EEG source localization using dipole approximations.

Main Methods:

• Developed a variational Bayesian algorithm for rapid inversion of M/EEG dipole models.
• Incorporated specification of priors for all model parameters.
• Utilized Bayesian evidence for comparing models with varying numbers of dipoles.

Main Results:

• The proposed algorithm accurately localizes dipoles in synthetic M/EEG data.
• Demonstrated the utility of the Bayesian scheme in a multi-subject EEG auditory study.
• Successfully compared competing models for the generation of the N100 component.

Conclusions:

• The variational Bayesian algorithm offers a fast and accurate method for M/EEG dipole modeling.
• This approach enhances source reconstruction by providing robust parameter estimation and model selection.
• The method facilitates more reliable inferences about brain activity from M/EEG sensor data.