Myoelectric pattern identification of stroke survivors using multivariate empirical mode decomposition

Area of Science:

• Biomedical Engineering
• Signal Processing
• Rehabilitation Technology

Background:

• Surface electromyogram (EMG) pattern recognition is crucial for prosthetic control and rehabilitation.
• Current methods often process EMG channels independently, potentially missing inter-muscle relationships.
• Accurate feature extraction is key to improving the performance of myoelectric control systems.

Purpose of the Study:

• To develop and evaluate a novel feature extraction method for multi-channel surface EMG signals.
• To utilize multivariate empirical mode decomposition (MEMD) for simultaneous analysis of multiple EMG channels.
• To assess the effectiveness of MEMD-based features for identifying functional movement patterns in stroke survivors.

Main Methods:

• A multivariate extension of empirical mode decomposition (MEMD) was applied to multi-channel surface EMG data.
• Mode-aligned intrinsic mode functions (IMFs) were extracted, representing signal components across multiple scales.
• Normalized amplitude distributions of IMFs across channels were computed as features.
• A linear discriminant classifier was used to identify 18 functional movement patterns.

Main Results:

• The MEMD-based feature set achieved an average classification error rate of 4.61 ± 4.70%.
• This performance was significantly better than the conventional time-domain feature set (7.14 ± 6.15%, p < 0.05).
• The method effectively captured relationships across multiple muscles via aligned intrinsic scales.

Conclusions:

• Multivariate EMD (MEMD) offers a promising approach for feature extraction from multi-channel EMG data.
• The proposed method enhances the modeling of muscle couplings for improved myoelectric pattern recognition.
• This technique shows potential for advancing rehabilitation and assistive technologies for individuals with motor impairments.