Estimating proximity to muscular failure using surface EMG and deep learning

Area of Science:

• Biomedical Engineering
• Exercise Physiology
• Machine Learning

Background:

• Personalizing resistance training requires real-time monitoring of muscular fatigue.
• Surface electromyography (sEMG) offers a non-invasive method to assess muscle activation.

Purpose of the Study:

• To develop and evaluate a deep learning model for real-time estimation of proximity to muscular failure using sEMG signals.
• To create a novel dataset for training and validating such models.

Main Methods:

• Collected a dataset of 192 sEMG recordings from isometric biceps brachii holds to failure.
• Preprocessed sEMG signals and converted them into spectrograms.
• Trained deep learning models (MLP, Transformer, LSTM) and regression baselines to predict a Proximity to Failure Index (PFI).

Main Results:

• Deep learning models significantly outperformed linear and support vector regression baselines.
• The Long Short-Term Memory (LSTM) network achieved the lowest mean squared error (49.44±18.34).
• Demonstrated accurate estimation of PFI from sEMG spectrograms.

Conclusions:

• Proximity to muscular failure can be reliably estimated from sEMG during isometric contractions.
• The findings support the development of real-time biofeedback systems for adaptive resistance training.
• This technology has the potential to optimize training personalization and performance.