EEG artifact removal using sub-space decomposition, nonlinear dynamics, stationary wavelet transform and machine...

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Electroencephalogram (EEG) artifact elimination is crucial for accurate neural signal analysis.
• Existing Blind Source Separation (BSS) methods require effective artifact identification and removal strategies.
• Complete removal of artifactual components can lead to loss of valuable neural information.

Purpose of the Study:

• To introduce a novel method for detecting and suppressing artifacts in EEG signals.
• To preserve neural information during artifact removal.
• To evaluate the proposed method's effectiveness using simulated, semi-simulated, and real EEG data.

Main Methods:

• Utilized Poincare planes in phase space for artifact detection.
• Employed Second-Order Blind Identification (SOBI) for artifact component estimation.
• Combined conventional classifiers (MLP, KNN, Naïve Bayes, SVM) with Stationary Wavelet Transform (SWT) for artifact suppression and neural information preservation.

Main Results:

• Achieved 98% average accuracy and 97% average sensitivity in artifactual EEG component detection.
• Obtained a mean square error of approximately 2% in EEG reconstruction after artifact removal.
• Demonstrated effectiveness across simulated, semi-simulated, and real-world EEG scenarios.

Conclusions:

• The proposed method effectively detects and removes EEG artifacts while preserving neural activity.
• The combination of signal processing and machine learning offers a robust solution for EEG artifact management.
• The method shows promise for future applications in EEG analysis and research.