Quaternion Common Spatial Patterns.

Area of Science:

• Signal Processing
• Biomedical Engineering
• Machine Learning

Background:

• Co-channel coupling in multi-dimensional processes requires advanced modeling techniques.
• Quaternion-valued data presents unique challenges due to its non-circular nature.
• Existing methods may not fully capture the complexities of quaternion-valued signals.

Purpose of the Study:

• To introduce a novel quaternion-valued common spatial patterns (QCSP) algorithm.
• To propose a generalized QCSP (G-QCSP) for non-circular quaternion-valued data, incorporating power differences.
• To demonstrate the application of G-QCSP in high-accuracy classification tasks, even at low signal-to-noise ratios (SNR).

Main Methods:

• Development of the generalized quaternion-valued common spatial patterns (G-QCSP) algorithm.
• Incorporation of power difference information between real and imaginary parts of data channels.
• Application of G-QCSP for feature extraction in electroencephalography (EEG) analysis.

Main Results:

• G-QCSP achieves high classification rates at signal-to-noise ratios (SNR) as low as -10 dB.
• Classification accuracy of 70% for discriminating between imagery left and right hand movements using EEG data.
• Classification accuracy of 87% for distinguishing between Parkinson's disease (PD) patients and healthy control subjects.

Conclusions:

• The proposed G-QCSP algorithm is effective for modeling co-channel coupling in multi-dimensional processes.
• G-QCSP demonstrates significant utility in EEG analysis for both motor imagery and disease diagnosis.
• The method offers robust performance even in challenging low SNR conditions, highlighting its potential for real-world applications.