s-SMOOTH: Sparsity and Smoothness Enhanced EEG Brain Tomography

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Computational Neuroscience

Background:

• Electroencephalography (EEG) source imaging reconstructs brain activity with high temporal resolution but faces an ill-posed inverse problem.
• Infinite solutions arise from limited sensors and signal noise, necessitating regularization for accurate brain imaging.

Purpose of the Study:

• To introduce a novel method, Sparsity and SMOOthness enhanced brain TomograpHy (s-SMOOTH), for improved EEG source reconstruction accuracy.
• To integrate Total Generalized Variation (TGV) and L1-2 regularization into a unified framework for enhanced brain imaging.

Main Methods:

• Proposed a voxel-based Total Generalized Variation (vTGV) for 3D EEG source images, extending 2D TGV to irregular cortical surfaces.
• Utilized L1-2 regularization to promote sparsity in current density, offering computational advantages over L1 regularization.
• Employed an efficient algorithm combining Difference of Convex functions Algorithm (DCA) and Alternating Direction Method of Multipliers (ADMM) to solve the proposed model.

Main Results:

• The s-SMOOTH method demonstrated superior performance in total reconstruction accuracy, localization accuracy, and focalization degree compared to existing methods using synthetic data.
• vTGV effectively preserves source edges and enhances image smoothness, reducing staircasing artifacts.
• L1-2 regularization accelerated computations and improved sparsity.

Conclusions:

• The proposed s-SMOOTH method significantly enhances EEG source imaging reconstruction accuracy.
• The novel vTGV regularization and L1-2 combination provide a robust framework for 3D brain imaging.
• The method shows strong potential for real-world applications, including event-related potential source localization.