EEG Feature Extraction and Classification for Upper Limb Flexion and Extension Motor Imagery Based on Discriminative...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Traditional common spatial pattern (CSP) algorithms struggle with overlapping cortical representations and frequency sensitivity in motor imagery (MI) electroencephalogram (EEG) signals for upper limb neural rehabilitation.
• These limitations hinder the decoding performance of MI-EEG signals, necessitating improved methodologies.

Purpose of the Study:

• To enhance the decoding performance of upper limb MI-EEG signals by employing an improved discriminative filter bank common spatial pattern (DFBCSP) framework.
• To evaluate the classification performance of the DFBCSP framework combined with different machine learning models for MI tasks.

Main Methods:

• EEG data were collected from 16 participants performing two-class and three-class upper limb MI tasks.
• EEG signals were decomposed into nine sub-bands, and mutual information-based feature selection was applied for optimization.
• Optimized features were classified using multilayer perceptron (MLP), support vector machine (SVM), and linear discriminant analysis (LDA).

Main Results:

• The DFBCSP + MLP method demonstrated superior performance compared to traditional CSP algorithms.
• Achieved 94.83% accuracy (Kappa: 0.890) for two-class MI tasks and 86.20% accuracy (Kappa: 0.775) for three-class MI tasks.

Conclusions:

• The DFBCSP + MLP framework offers a robust and effective approach for decoding upper limb MI-EEG signals.
• This framework provides a strong technical foundation for future research in upper limb motor dysfunction rehabilitation.