Effects of Different Preprocessing Pipelines on Motor Imagery-Based Brain-Computer Interfaces.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) utilize electroencephalography (EEG) signals for device control.
• Improving information transfer rates in BCIs is crucial for practical applications.
• Optimal preprocessing pipelines for EEG signals in BCIs require further investigation.

Purpose of the Study:

• To explore and evaluate various EEG preprocessing techniques for motor imagery-based BCIs.
• To determine the most effective preprocessing methods and their optimal sequence.
• To identify preprocessing pipelines suitable for real-time online BCI implementation.

Main Methods:

• Rigorous testing of multiple preprocessing pipelines (e.g., independent component analysis, surface Laplacian, bandpass filtering, baseline correction) across four EEG datasets.
• Integration and evaluation of five EEG machine learning models with different preprocessing methods.
• Analysis of time complexity for assessing suitability for online deployment.

Main Results:

• Baseline correction and bandpass filtering consistently yielded the most significant preprocessing benefits.
• A recommended pipeline for online implementation includes baseline correction, bandpass filtering, and surface Laplacian.
• The surface Laplacian algorithm showed enhanced performance when combined with spatial information algorithms.
• Achieved superior results (92.91%, 88.11%) compared to state-of-the-art feature extraction methods in specific cases.

Conclusions:

• The study provides critical insights into selecting effective EEG preprocessing pipelines for signal decoding.
• The findings contribute to the advancement and refinement of brain-computer interface technologies.
• Identified specific preprocessing methods and sequences that enhance BCI performance and enable online implementation.