Phasor-Based Myoelectric Synergy Features: A Fast Hand-Crafted Feature Extraction Scheme for Boosting Performance in...

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Human-Computer Interaction

Background:

• Surface electromyographic signals (EMGs) are vital for myoelectric control of lower limb assistive devices.
• Traditional machine learning models struggle with complex gait phase recognition beyond basic stance and swing phases.

Purpose of the Study:

• To introduce and evaluate a generalized phasor-based feature extraction approach (PHASOR) for enhanced gait phase recognition.
• To improve the performance of Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) models in recognizing five distinct gait phases.

Main Methods:

• Utilized a publicly available dataset from 40 subjects for evaluating the PHASOR feature extraction method.
• Compared PHASOR against existing state-of-the-art feature sets and deep learning architectures (Rocket, Mini-Rocket).
• Employed separability index (SI) and mean semi-principal axis (MSA) analyses to assess feature effectiveness.

Main Results:

• PHASOR demonstrated strong performance with mean SI of 7.7 and MSA of 0.5, indicating effective decoding of gait phases from EMG data.
• Support Vector Machine (SVM) achieved 82% accuracy using a five-fold leave-one-trial-out approach, surpassing Rocket and Mini-Rocket.
• The PHASOR approach proved competitive with deep learning methods, offering efficient feature extraction.

Conclusions:

• The PHASOR feature extraction method significantly enhances gait phase recognition accuracy using EMG signals.
• Novel feature extraction schemes like PHASOR can rival deep learning approaches in EMG-based gait analysis.
• This advancement holds promise for more sophisticated and responsive myoelectric control systems for assistive devices.