Improving EEG Source Localization with a Novel Regularization: Spatiotemporal Graph Total Variation (STGTV) Method.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalography (EEG) source localization reconstructs brain activity from scalp recordings, often using inverse problem-solving techniques.
• Existing spatial regularization methods like graph fractional-order total variation (gFOTV) process each time point independently, risking temporal discontinuity and noise overfitting.
• Realistic EEG data often suffers from low signal-to-noise ratio (SNR) and limited electrode counts, challenging current localization methods.

Purpose of the Study:

• To develop a novel spatiotemporal regularization method for EEG source localization that accounts for continuous temporal variations while allowing for abrupt changes.
• To improve the accuracy and robustness of EEG source localization, particularly under low SNR and limited electrode conditions.

Main Methods:

• Introduced a novel spatiotemporal graph total variation (STGTV) method for EEG source localization.
• STGTV integrates graph fractional-order total variation (gFOTV) for spatial regularization and standard total variation (TV) for temporal regularization.
• The combined approach encourages spatially smooth source distributions and enhances temporal consistency, facilitating noise cancellation and SNR improvement.

Main Results:

• Simulation studies demonstrated that STGTV significantly outperformed the standalone gFOTV method.
• STGTV achieved lower localization errors compared to gFOTV.
• The proposed method resulted in fewer spuriously discovered sources, indicating improved accuracy and reliability.

Conclusions:

• The proposed STGTV method offers a significant advancement in EEG source localization by incorporating spatiotemporal regularization.
• STGTV effectively addresses the limitations of purely spatial methods, providing more accurate and temporally consistent brain activity reconstructions.
• This method shows promise for improving the analysis of realistic EEG recordings, especially in challenging low-SNR scenarios.