Detecting Early-stage Muscle Fatigue to Improve Assist-as-Needed Control Strategies for Active Orthoses

Background: Muscle fatigue affects motor neuron function and, in turn, the electromyography\,(EMG) signals used to control wearable technologies. By accounting for this, we can more accurately regulate movement assistance. Muscle fatigue has been widely studied in the fields of sports science, rehabilitation, and occupational health. However, there are contradictory findings in the literature, and some aspects of muscle fatigue mechanisms are not yet well understood. Furthermore, none of the literature found focuses on the detection of early-stage muscle fatigue. Results: In this paper we specifically focus on early-stage muscle fatigue and show that the spectral variance (\((S_{var})\)) of the EMG signal is a more reliable measure of fatigue across participants and contraction levels compared to other widely used features. We then make recommendations on how these findings can be used in EMG-controlled assist-as-needed technologies. Experiments were carried out with 16 participants who performed intermittent isometric contractions of the biceps brachii at 20\,%, 50\,% and 75\,% MVC. The reliability of the motor unit action potential conduction velocity (CV) and the median/mean power frequencies (MDPF/MNPF) as measures of fatigue was tested. CV exhibited some linear correlation with perceived fatigue only at the higher contraction levels, for the majority of participants (R2=\((0.18\pm0.23)\)). Conclusions: CV was found to be an unreliable measure for muscle fatigue and no relationship was found between MDPF/MNPF and fatigue. However, for all participants and contraction levels, the fatigue-\((S_{var})\) relationship could be characterized using a Gaussian model\,(R2=\((0.87\pm0.12)\)).