Learning spatial filters from EEG signals with Graph Signal Processing methods.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Electroencephalography (EEG) is crucial for brain activity monitoring.
• Traditional EEG processing methods like surface Laplacian filters face challenges with noise and electrode defects.
• Improving topographical localization and attenuating volume-conducted features are key goals in EEG analysis.

Purpose of the Study:

• To develop a novel spatial filtering method for EEG signals using graph signal processing.
• To enhance topographical localization and reduce artifacts in EEG data.
• To provide a robust alternative to the surface Laplacian filter, especially in noisy conditions or with electrode issues.

Main Methods:

• Learning a spatial filter directly from EEG signals.
• Utilizing graph signal processing tools, including a graph learning algorithm.
• Combining a graph learning algorithm with a high-pass graph filter to remove large spatial signals.
• Applying the method to raw EEG data.

Main Results:

• The proposed method effectively increases topographical localization.
• Volume-conducted features are attenuated by the graph filter.
• The approach demonstrates robustness to noise and defective electrodes.
• Results are comparable to the surface Laplacian filter in noiseless scenarios.

Conclusions:

• The developed graph-based spatial filter is a viable alternative to the surface Laplacian for EEG processing.
• This method offers improved performance in scenarios with low signal-to-noise ratios or defective electrodes.
• The technique holds clinical relevance for enhancing EEG signal quality in practical settings, particularly with electrode artifacts.