Spatiospectral Decomposition of Multi-subject EEG: Evaluating Blind Source Separation Algorithms on Real and...

Area of Science:

• Neuroscience
• Computational Neuroscience

Background:

• Electroencephalographic (EEG) oscillations (1–50 Hz) are linked to cognitive functions.
• Blind Source Separation (BSS) methods improve the isolation of distinct EEG processes.
• Multi-subject analysis enhances the understanding of brain networks.

Purpose of the Study:

• To demonstrate the feasibility of multi-subject BSS for deriving distinct EEG spatiospectral maps.
• To evaluate the performance of various BSS algorithms for EEG source separation.
• To identify optimal algorithms for analyzing brain networks in healthy and clinical populations.

Main Methods:

• Implemented multi-subject spatiospectral EEG decompositions using the EEGIFT toolbox.
• Utilized both real and realistic simulated EEG datasets.
• Evaluated twelve different decomposition algorithms, including WASOBI, COMBI, and various ICA methods.

Main Results:

• WASOBI and COMBI were the top-performing algorithms on simulated data, decomposing sources across varying noise levels.
• INFOMAX ICA, FAST ICA, WASOBI, and COMBI identified the most stable sources in real EEG data.
• The evaluated algorithms provided partially distinct spatiospectral maps of underlying brain activity.

Conclusions:

• Multi-subject BSS is a feasible approach for deriving distinct EEG spatiospectral maps.
• WASOBI and COMBI are recommended for their performance in EEG source decomposition.
• This methodology can advance the study of spatiospectral brain networks in diverse populations.