Author(s): COSTA DE SOUSA, L., DA COSTA-MACHADO, R., CONCEIÇÃO, A., PLÁCIDO DA SILVA, H., Institution: INSTITUTO SUPERIOR TÉCNICO /INSTITUTO DE TELECOMUNICAÇÕES, Country: PORTUGAL, Abstract-ID: 1696

INTRODUCTION:
Muscle fatigue (MF) is a pivotal factor in sports performance, yet its real-time estimation remains challenging as conventional methods rely on subjective scales or post-exercise analysis [1]. Surface electromyography (sEMG) provides objective insights into motor unit recruitment and conduction velocity [2], which correlate with MF, leading to the exploration of multimodal AI-driven algorithms that integrate mechanical and physiological data for real-time MF detection. The development of such algorithms requires large amounts of data, and establishing reliable protocols to induce and objectively quantify fatigue is crucial. To addresses this gap, the validation of an experimental protocol designed to induce MF in a consistent manner is presented, serving as a preliminary stage to ensure the data reliability of the future AI-driven system.
METHODS:
Six males (20.33 ± 1.37 years) participated in this proof-of-concept study. Using a variable-cam leg-extension machine, subjects performed two sets of repetitions until task failure. sEMG was recorded from the vastus lateralis and rectus femoris (PLUX Biosignals). Rating of Perceived Exertion (RPE) was collected every 5 repetitions (Borg-CR10). Signal processing included a 2nd order Butterworth bandpass filter (20-250Hz) and extraction of time-domain (Root Mean Square - RMS) and frequency-domain (Median Frequency - MDF, Mean Frequency - MNF, and Zero Crossing Rate - ZCR) features.
RESULTS:
A consistent upward trend in RMS was observed (p < 0.05) across all subjects, indicating increased motor unit recruitment, which is consistent with the onset of MF. Frequency-domain analysis showed a significant shift toward lower frequencies, with MDF and MNF decreasing by approximately 30% from the first to the last repetition. ZCR also decreased, confirming the slowing of muscle fiber conduction velocity. The RPE scores correlate with these findings, showing a consistent increase that reached the maximum level (10) at task failure.
CONCLUSION:
The results showed that this protocol effectively induces localized MF, with high convergence between neuromuscular indicators and subjective perception. The consistency of the signals shows that data originated from this protocol is suitable for training AI models for automated MF detection, potentially enabling the future transformation of conventional gym equipment into "smart" platforms capable of providing real-time estimation and biofeedback for performance optimization and injury prevention.

References:
[1] J. Yu, et al., “Exploration and Application of a Muscle Fatigue Assessment Model Based on NMF for Multi-Muscle Synergistic Movements,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol.32, 2024
[2] S. Wang, et al., “A Novel Approach to Detecting Muscle Fatigue Based on sEMG by Using Neural Architecture Search Framework,” IEEE Transactions on Neural Networks and Learning Systems, vol.34, 2023