Developing Signal Processing and Feature Extraction Methods for Brain Computer Interface

Abstract

Brain Computer Interface (BCI) systems represent a revolutionary advancement in neural output manipulation, empowering individuals to interact with their environment solely through brain signals, bypassing conventional neuromuscular pathways. These systems primarily aim to facilitate control over assistive devices like speech synthesizers and robotic wheelchairs, facilitating seamless interaction with the surroundings. At the core of BCI systems lie preprocessing and feature extraction modules, which play pivotal roles in enhancing system performance. To this end, this work endeavors to pioneer novel techniques in both preprocessing and feature extraction for BCI systems. Beginning with the preprocessing phase, our focus centers on refining artifact removal from Electroencephalogram (EEG) data. We propose an automated approach leveraging Independent Component Analysis (ICA) guided by Sparse Entropy, effectively identifying and eliminating artifactual Independent Components without manual intervention. Furthermore, the methodology amalgamates Preconditional ICA for real data with Time-Reassigned Multisynchrosqueezing Transform, amplifying EEG artifact correction performance while preserving valuable neurological signals intertwined with artifacts. This correction-centric approach ensures a robust foundation for neuroscientific exploration and clinical applications by retaining pertinent neural activity. Transitioning to feature extraction, we introduce an innovative method tailored for Motor Imagery (MI) EEG signals within BCI systems. By integrating Multivariate Variational Mode Decomposition with Phase Space Reconstruction, we address the complexity of multivariate oscillatory patterns in MI-EEG signals, enhancing system accuracy and reliability. This method strategically utilizes data from a minimal set of EEG channels, optimizing computational efficiency without compromising performance. By reconstructing modes derived from Multivariate Variational Mode Decomposition into a 2D phase space, we gain deeper insights into the dynamic evolution of MI-EEG signals, facilitating comprehensive feature extraction encompassing statistical metrics and nonlinear characteristics. Remarkably, this method achieves high classification accuracy in both binary and multi-class tasks using five EEG channels, underscoring its viability for real-time BCI applications.

In a further advancement, we propose a novel feature extraction method focusing on MI-EEG signals, combining Local Maximum Synchrosqueezing Transform with Non-Negative Matrix Factorization. This approach presents substantial improvements in BCI technology, integrating a channel selection technique to enhance real-time feasibility while reducing computational burden and eliminating redundant information. The Local Maximum Synchrosqueezing Transform extracts temporal and spectral information from selected channels, transforming MI-EEG signals into time-frequency representations with enhanced resolution. Non-Negative Matrix Factorization further refines these representations, capturing meaningful patterns and reducing dimensionality for precise signal classification. By incorporating statistical and nonlinear features, this method achieves outstanding classification accuracies in binary and multi-class tasks, surpassing previous approaches and affirming its efficacy in advancing MI- EEG signal analysis for BCI systems. This comprehensive exploration delves into the forefront of BCI systems, revolutionizing neural output manipulation and enhancing user interaction with the environment. By focusing on innovative preprocessing and feature extraction techniques, we've significantly advanced the capabilities of BCI systems. Our automated artifact removal method, coupled with a correction-centric approach, ensures robust neural signal processing, preserving vital neurological activity for comprehensive analysis. Furthermore, our feature extraction methods for MI-EEG signals showcase remarkable accuracy and efficiency, leveraging minimal EEG channels while capturing complex temporal and spectral dynamics. These advancements underscore the potential for real-time BCI applications, paving the way for more feasible and effective BCI system designs with profound implications for neuroscientific research and clinical practice.