Real-Time Forecasting of sEMG Features for Trunk Muscle Fatigue Using Machine Learning

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Surface electromyography (sEMG) signals reflect muscle activity and fatigue.
• sEMG signal features are non-stationary and vary significantly between individuals.
• Accurate forecasting of sEMG features is crucial for monitoring muscle status.

Purpose of the Study:

• To investigate the efficacy of adaptive algorithms for forecasting trunk muscle sEMG features.
• To develop a real-time prediction system for muscle fatigue.
• To explore the application of deep learning in sEMG analysis.

Main Methods:

• Utilized shallow models and a deep Convolutional Neural Network (CNN).
• Simultaneously learned and forecasted 5 common sEMG features in real-time.
• Evaluated predictions up to a 25-second horizon across 14 trunk muscles in 13 healthy subjects performing various exercises.

Main Results:

• The CNN achieved a 25-second ahead forecast with 6.88% mean absolute percentage error and 3.72% standard deviation.
• CNN outperformed the best shallow model by at least 30% in accuracy and precision.
• Demonstrated accurate and precise forecasting despite non-stationary sEMG features and inter-subject variability.

Conclusions:

• Adaptive learning and forecasting, particularly with CNNs, provide reliable predictions for sEMG features.
• The findings support the development of wearable devices for muscle fatigue forecasting.
• The models offer a general framework for muscle activity monitoring applications in physical therapy and rehabilitation.