Recurrent Fusion of Time-Domain Descriptors Improves EMG-based Hand Movement Recognition.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Robot Interaction

Background:

• Myoelectric pattern recognition (PR) is key for natural control of powered prostheses.
• Current limitations in clinical translation stem from the quality of extracted Electromyogram (EMG) features.
• Accurate and reliable feature extraction is vital for effective EMG PR systems.

Purpose of the Study:

• To introduce a novel feature extraction method, Recurrent Fusion of Time Domain Descriptors (RFTDD) combined with Range Spatial Filtering (RSF).
• To enhance classifier performance for more appropriate prosthetic hand control in clinical applications.
• To improve the capture of spatio-temporal dependencies in EMG signals.

Main Methods:

• Proposed a hybrid approach using RSF for spatial information and RFTDD for temporal information from EMG signals.
• RSF increases available EMG data by analyzing spatial information across physical channels.
• RFTDD applies recurrent fusion to time-domain features, using a sigmoidal function to manage feature range.

Main Results:

• The RFTDD method demonstrated a significant reduction in classification errors, averaging approximately 12% across subjects.
• Benchmarking against traditional methods on two EMG datasets with 31 subjects confirmed RFTDD's effectiveness.
• The method successfully captured complex temporal-spatial dependencies in EMG signals, reducing classification errors.

Conclusions:

• The proposed RFTDD method, utilizing simple time-domain features, offers a robust solution for clinical applications.
• This approach enhances the accuracy and reliability of myoelectric control for prosthetic devices.
• RFTDD presents a promising avenue for low-cost, high-performance prosthetic control systems.