Online recursive independent component analysis for real-time source separation of high-density EEG

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Online Independent Component Analysis (ICA) is crucial for real-time biosignal processing and brain-computer interfaces.
• Existing online ICA methods often lack computational efficiency and are limited to low-density EEG data.
• Algorithmic convergence proofs dominate prior research, with limited practical application to complex datasets.

Purpose of the Study:

• To introduce an optimized online recursive ICA algorithm (ORICA) for efficient blind source separation.
• To implement ORICA with recursive least squares (RLS) whitening for high-density EEG data processing.
• To adapt ORICA for real-time applications using an open-source framework.

Main Methods:

• Developed an optimized online recursive ICA algorithm (ORICA).
• Integrated online recursive least squares (RLS) whitening for enhanced performance.
• Implemented ORICA as an online-capable plugin within the BCILAB (EEGLAB) framework.
• Derived and evaluated a block-update modification for the ORICA learning rule.

Main Results:

• Demonstrated accurate and efficient source identification in high-density (64-channel) simulated EEG data.
• Validated the algorithm's performance on real 61-channel EEG data from a wearable dry EEG system.
• Showcased the suitability of ORICA for real-time analysis of cognitive experiment data.

Conclusions:

• ORICA provides an efficient and accurate solution for blind source separation in high-density EEG.
• The algorithm is suitable for real-time applications, particularly with wearable EEG systems.
• The BCILAB plugin facilitates broader adoption and application of online ICA in neuroscience research.