WaveCSP: a robust motor imagery classifier for consumer EEG devices.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Consumer-level brain-computer interfaces (BCI) require robust motor imagery (MI) classification.
• Device limitations and subject variability pose significant challenges for MI classification algorithms.

Purpose of the Study:

• To investigate the impact of electrode limitations, signal quality, and user variability on MI classification performance.
• To propose and evaluate a novel machine learning approach, WaveCSP, to mitigate these challenges.

Main Methods:

• Developed WaveCSP, a machine learning algorithm utilizing 24 features from EEG signals.
• Employed wavelet transform and common spatial pattern (CSP) filtering for feature extraction.
• Evaluated WaveCSP on the Physionet MI database and data from a commercial EEG headset.

Main Results:

• WaveCSP demonstrated improved performance regarding subject variability compared to existing methods.
• Over 50% of subjects in the Physionet MI database achieved >64% accuracy.
• Four out of five subjects with prior BCI experience using a commercial headset achieved >64% accuracy.

Conclusions:

• WaveCSP offers a robust solution for motor imagery classification in BCI applications.
• The proposed methods effectively address limitations in hardware and user factors, enhancing BCI usability.