Bayesian Algorithms for Joint Estimation of Brain Activity and Noise in Electromagnetic Imaging

Area of Science:

• Neuroscience
• Biophysics
• Signal Processing

Background:

• Simultaneously estimating brain source activity and noise in electromagnetic brain imaging (EBI) is a significant challenge.
• Reconstructing brain activity from limited sensor data involves solving an NP-hard inverse problem complicated by noise and interference.

Purpose of the Study:

• To present a novel generative model and associated Bayesian inference algorithms for simultaneous brain source activity and sensor noise estimation in EBI.
• To evaluate the performance, robustness, and computational efficiency of the proposed algorithms.

Main Methods:

• Developed a generative model incorporating an augmented leadfield matrix for EBI.
• Derived three Bayesian inference algorithms: expectation-maximization (EBI-EM), convex bounding (EBI-Convex), and fixed-point (EBI-Mackay).
• Conducted comprehensive simulations to assess algorithm performance using Gaussian and real brain noise.

Main Results:

• EBI-Convex and EBI-Mackay algorithms demonstrated superior performance over EBI-EM.
• EBI-Convex and EBI-Mackay showed robustness to initialization and fast convergence.
• These algorithms effectively reconstructed complex brain activity from limited sensor data and resting-state data, improving source reconstruction and noise learning.

Conclusions:

• The EBI-Convex and EBI-Mackay algorithms offer significant advancements for simultaneous brain activity and noise estimation in EBI.
• These methods provide robust, efficient, and accurate solutions for source reconstruction, even with limited data.
• The developed approach enhances the capabilities of magneto- and electroencephalography for brain imaging research.