Temporal Context Informed Myoelectric Feature Extraction Uncovers Frequency Invariance in EMG-based Gesture...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Human-Computer Interaction

Background:

• Electromyographic (EMG) armbands are key for gesture recognition.
• Current methods often ignore temporal activity trends, limiting context capture and leading to errors.
• Existing spatial feature extraction methods can be enhanced by incorporating temporal dynamics.

Purpose of the Study:

• To develop a context-aware Electromyographic (EMG) feature extraction framework.
• To improve gesture recognition accuracy by integrating short-term and long-term temporal activity trends.
• To evaluate the proposed method's performance across different sampling frequencies and in clinical populations.

Main Methods:

• Developed a temporal context framework encapsulating Phasor-based Multi-signal Waveform Length (MSWL) features.
• Concatenated short-term memory (partial correlation) and long-term memory (trend) information streams.
• Evaluated the method on EMG datasets from healthy subjects with varying sampling frequencies and transradial amputees (NinaPro protocol).

Main Results:

• The context-aware EMG feature extraction demonstrated sampling frequency invariance in gesture recognition.
• Achieved similar average accuracy (91%) across high- and low-frequency armbands, outperforming 58 existing methods.
• Showcased efficacy in context-sensitive EMG pattern recognition using amputee data.

Conclusions:

• Context-aware EMG feature extraction enhances gesture recognition accuracy and robustness.
• The proposed method challenges the conventional preference for higher sampling frequency devices in EMG applications.
• This approach holds significant clinical relevance for advanced human-machine interfaces.