A linearly extendible multi-artifact removal approach for improved upper extremity EEG-based motor imagery decoding

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Non-invasive multichannel Electroencephalography (EEG) is crucial for decoding motor imagery (MI) for brain-computer interfaces (BCIs) in rehabilitation robotics.
• Artifacts in EEG signals pose significant challenges for BCI system processing and practical application.
• Existing artifact removal methods often require calibration or reference electrodes and struggle with simultaneous multi-artifact cancellation in MI-EEG.

Purpose of the Study:

• To propose a semi-automatic EEG preprocessing method for simultaneous elimination of multiple artifacts from MI-EEG signals.
• To evaluate the effectiveness of the proposed method in improving motor imagery decoding for transhumeral amputees.
• To assess the feasibility of the method for real-time control in miniaturized rehabilitation robotic interfaces.

Main Methods:

• A semi-automatic EEG preprocessing method combining Generalized Eigenvalue Decomposition (GEVD) with low-rank approximation and a Multi-channel Wiener Filter (MWF) was developed.
• The GEVD-MWF method employs a learning technique for simultaneous artifact elimination.
• The method was applied to 64-channel EEG data from transhumeral amputees performing upper limb MI tasks, followed by feature selection and machine learning for movement intent decoding.

Main Results:

• The GEVD-MWF method significantly improved MI decoding accuracies by 13.23%-41.21% compared to existing artifact removal algorithms.
• High decoding accuracies (90.44% - 99.67%) were achieved using single-trial EEG recordings, reducing the need for large data ensembles.
• A subset of 9 channels yielded a decoding accuracy of 93.73%±1.58% using a sequential forward floating selection algorithm.

Conclusions:

• The proposed GEVD-MWF method demonstrates superior performance in artifact removal and MI decoding.
• Reduced data requirements and feasibility with fewer channels make the method practical for real-time applications.
• The GEVD-MWF method has the potential to advance the development of effective control strategies for EEG-based rehabilitation robotics.