Bayesian estimation of evoked and induced responses

Area of Science:

• Neuroscience
• Biophysics
• Signal Processing

Background:

• Magnetoencephalography (MEG) and electroencephalography (EEG) are crucial for studying brain activity.
• Source reconstruction in MEG/EEG aims to identify the neural origins of measured signals.
• Existing empirical Bayes approaches have primarily focused on spatial aspects of source reconstruction.

Purpose of the Study:

• To extend empirical Bayes source reconstruction for MEG/EEG to encompass both evoked and induced neural responses.
• To develop an estimation scheme capable of distinguishing between phase-locked (evoked) and non-phase-locked (induced) neural activity.
• To enable Bayesian estimation of changes in power or energy of wavelet coefficients for both response types.

Main Methods:

• Utilized a restricted maximum likelihood (ReML) framework for covariance component estimation.
• Incorporated temporal basis functions to constrain the temporal characteristics of neural responses.
• Developed distinct hierarchical multitrial models for evoked and induced responses.
• Applied the extended scheme to estimate evoked and induced changes in power/energy of wavelet coefficients.

Main Results:

• Successfully extended the empirical Bayes approach to jointly estimate evoked and induced responses in MEG/EEG.
• Demonstrated the capability of the ReML-based scheme to differentiate between phase-locked and non-phase-locked neural activity.
• Showcased efficient estimation of evoked and induced changes in power or energy of wavelet coefficients.

Conclusions:

• The developed method provides a unified framework for analyzing both evoked and induced neural responses in MEG/EEG source reconstruction.
• The ReML-based approach offers an efficient and robust method for Bayesian estimation of neural activity.
• This extension enhances the analytical power of MEG/EEG for investigating complex brain dynamics.