Muscle Fatigue Assessment in Healthcare Application by Using Surface Electromyography: A Transfer Learning Approach

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Gerontology

Background:

• Muscle fatigue monitoring is crucial for elderly safety and activity support.
• Existing methods for fatigue assessment have limitations in real-time application.
• Ambient Assisted Living (AAL) requires non-invasive, reliable monitoring solutions.

Purpose of the Study:

• To introduce a novel deep learning framework for classifying muscle fatigue levels.
• To utilize wireless surface electromyographic (sEMG) data for fatigue detection.
• To support the development of AAL applications for elderly care.

Main Methods:

• A new dataset of sEMG signals was collected from elderly and non-elderly adults.
• One-dimensional sEMG signals were transformed into two-dimensional time-frequency images (scalograms) using Continuous Wavelet Transform.
• Pre-trained Convolutional Neural Networks (CNNs) were fine-tuned for image classification, including binary and multiclass fatigue level detection.

Main Results:

• The deep learning framework achieved 98.6% accuracy in binary classification (fatigued vs. non-fatigued).
• A multiclass classification achieved 95.6% accuracy (No Fatigue, Moderate Fatigue, Hard Fatigue).
• The proposed transfer learning pipeline outperformed traditional Machine Learning methods (max 92% accuracy).

Conclusions:

• The proposed deep learning framework demonstrates robust and generalizable performance for muscle fatigue monitoring.
• This approach offers a potential real-time, non-invasive solution for AAL scenarios.
• The findings support the integration of advanced AI for enhanced elderly care and safety.