Bispectrum-Based Channel Selection for Motor Imagery Based Brain-Computer Interfacing.

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Motor imagery (MI) Brain-computer interfaces (BCI) performance is hindered by noise and redundant information in multi-channel electroencephalogram (EEG) data.
• Existing temporal and spatial feature-based channel selection methods do not fully capture oscillatory EEG signal power changes.
• Spectral features offer a promising alternative for improving channel selection in MI-based BCI.

Purpose of the Study:

• To propose and validate a novel bispectrum-based channel selection (BCS) method for MI-based BCI.
• To leverage spectral features from bispectrum analysis for more effective EEG channel selection.
• To improve classification accuracy in MI-based BCI by reducing noise and redundant information.

Main Methods:

• Utilized bispectrum analysis to extract non-linear and non-Gaussian signal information from EEG.
• Employed sum of logarithmic amplitudes (SLA) and first-order spectral moment (FOSM) features derived from bispectrum analysis.
• Validated the BCS method on three public BCI competition datasets (BCI competition IV dataset 1, BCI competition III dataset IVa, and BCI competition III dataset IIIa).

Main Results:

• The BCS method significantly outperformed using all channels across the tested datasets (e.g., 83.8% vs 69.4%).
• Achieved higher classification accuracies compared to other state-of-the-art methods.
• Demonstrated statistically significant improvements (p < 0.05) via Wilcoxon signed test.

Conclusions:

• The proposed bispectrum-based channel selection (BCS) method is effective for MI-based BCI.
• BCS successfully reduces noise and redundant information, leading to enhanced classification performance.
• This spectral feature-based approach offers a valuable advancement for BCI technology.