A New Labeling Approach for Proportional Electromyographic Control.

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Human-Computer Interaction

Background:

• Electromyography (EMG) is used for human-machine interfaces, enabling control of devices via muscle signals.
• Complex systems require proportional and simultaneous control schemes, often achieved through machine learning and regression.
• Current training methods for EMG control rely on visual tracking or additional sensors, posing challenges for individuals with disabilities.

Purpose of the Study:

• To develop a new, simplified labeling approach for proportional EMG control.
• To eliminate the need for complex training procedures and external sensor data.
• To investigate the impact of muscle contraction's transient phase on EMG control accuracy.

Main Methods:

• A novel method was developed to generate continuous labels directly from the EMG-feature stream during a simple training procedure.
• The approach avoids synchronization issues and does not require additional sensors like cameras or force sensors.
• A user study with 10 subjects evaluated five labeling methods, including variations of the new approach and binary label baselines, using 2D goal-reaching and tracking tasks.

Main Results:

• The new labeling approach, particularly when incorporating the transient phase of muscle contraction, resulted in more accurate proportional control compared to binary labeling methods.
• The proposed method successfully generated continuous labels without synchronization mismatches.
• No additional sensor data was required for the decoder calibration.

Conclusions:

• The introduced labeling approach offers a simplified and effective solution for proportional EMG control.
• This method is advantageous for individuals with physical disabilities who may not be able to utilize complex training setups.
• The findings suggest that considering the transient phase of muscle activity enhances EMG-based control performance.