Real-Time Wearable sEMG Onset Detection and Phase Discrimination of Sit-to-Stand Movement via a Compact Dual-Channel...

Area of Science:

• Biomechanics
• Wearable Technology
• Machine Learning

Background:

• Sit-to-stand and stand-to-sit transitions are crucial daily movements that can strain knee extensors.
• Work-related musculoskeletal disorders are linked to repetitive knee loading during these transitions.
• Effective monitoring and assistive devices require real-time detection and classification of these movements.

Purpose of the Study:

• To develop a wearable surface electromyographic (sEMG) system capable of real-time onset/offset detection and direction-aware classification of sit-to-stand and stand-to-sit transitions.
• To implement a deep learning method for accurate and immediate analysis of knee extensor muscle activity during these movements.
• To assess the system's performance in real-time applications for potential use in assistive technology and rehabilitation.

Main Methods:

• A wearable sEMG system was developed using two ESP32-S3 nodes and a hub to record muscle activity from vastus lateralis and vastus medialis.
• A deep learning approach was employed, featuring a per-burst convolutional detector and a dual-branch classifier with handcrafted features.
• The system was trained offline on the Gait120 dataset and deployed with embedded firmware for real-time operation, validated against a MATLAB reference.

Main Results:

• Offline evaluation on the Gait120 dataset demonstrated 99.6% accuracy for transition classification and 96.6% completeness for onset/offset detection.
• Real-time recordings from 30 healthy adults showed the system achieved 85.6% classification accuracy and 82.0% detection accuracy.
• The system exhibited approximately 100 ms latency and a compact 618 KB network footprint, indicating efficient real-time processing.

Conclusions:

• A low-power wearable sEMG system can successfully perform real-time detection and direction discrimination of sit-to-stand and stand-to-sit transitions.
• The developed deep learning method provides a robust solution for analyzing knee joint dynamics during functional movements.
• This technology holds significant potential for enhancing the control of assistive devices and improving the monitoring of knee function in clinical and daily settings.