Coefficient-of-variation-based channel selection with a new testing framework for MI-based BCI.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Multi-channel electroencephalogram (EEG) is crucial for motor-imagery (MI) based brain-computer interfaces (BCIs).
• EEG signals contain redundant information and noise, complicating feature extraction.
• Existing channel selection algorithms, particularly filtering techniques, face limitations in efficiency and effectiveness.

Purpose of the Study:

• To propose improved filtering channel selection methods for MI-BCIs.
• To develop a novel channel classification method based on signal characteristics.
• To introduce a new testing framework for evaluating channel selection algorithms' generalization ability.

Main Methods:

• A novel channel classification method was designed using coefficient of variation and inter-class distance.
• A filtering channel selection algorithm was developed based on the proposed classification.
• A new testing framework was established to assess the generalization performance of filtering channel selection algorithms.

Main Results:

• The proposed channel classification method effectively categorizes channels based on their contribution.
• The new testing framework demonstrates superior evaluation capabilities compared to existing methods.
• The developed channel selection algorithm achieved high accuracy (87.7% and 81.7%) on BCI competition datasets.

Conclusions:

• The proposed channel classification and filtering selection algorithm significantly improve MI-BCI performance.
• The novel testing framework provides a more robust evaluation of channel selection algorithm generalization.
• These advancements contribute to more effective and practical brain-computer interface systems.