Novel Muscle Sensing by Radiomyography (RMG) and Its Application to Hand Gesture Recognition.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Deep Learning

Background:

• Conventional electromyography (EMG) lacks explicit muscle contraction quantification.
• Mechanomyography (MMG) and accelerometers measure only surface motion.
• Imaging techniques like ultrasound, CT, and MRI provide only in-clinic snapshots.

Purpose of the Study:

• Introduce radiomyography (RMG) for continuous, wearable, or touchless muscle actuation sensing.
• Validate RMG's effectiveness in hand gesture recognition (HGR) and other muscle group monitoring.
• Explore RMG's potential in kinesiology, physiotherapy, rehabilitation, and human-machine interfaces.

Main Methods:

• Developed a novel radiomyography (RMG) sensor for continuous muscle actuation sensing.
• Utilized a wearable forearm sensor for experimental validation in hand gesture recognition.
• Applied time-frequency spectrogram conversion and a vision transformer (ViT) deep learning model for classification.
• Employed transfer learning to enhance adaptivity to user and sensor variations.

Main Results:

• Achieved up to 99% average accuracy in recognizing 23 hand gestures across 8 subjects.
• Demonstrated high adaptivity with up to 97% average accuracy using transfer learning.
• Successfully extended RMG to monitor eye and leg muscles, achieving high accuracy in eye movement and posture tracking.

Conclusions:

• Radiomyography (RMG) offers a novel solution for continuous muscle actuation sensing, overcoming limitations of existing methods.
• RMG, combined with deep learning, provides accurate and adaptive muscle activity monitoring for diverse applications.
• RMG holds significant potential for advancing fields such as kinesiology, physiotherapy, rehabilitation, and human-machine interfaces.