Automated Diagnosis of Muscle Fatigue Using Surface Electromyography (sEMG): A Review

Abstract

Muscle fatigue — the progressive decline in a muscle's ability to sustain a required force — is of central interest in clinical diagnosis, rehabilitation, sports science, and ergonomics. Surface electromyography (sEMG) is the most widely used non-invasive means of assessing muscle activity, since the amplitude and spectral content of the acquired motor unit action potentials (MUAPs) change measurably as fatigue develops. This paper reviews the state of the art in sEMG-based muscle fatigue and neuromuscular-disease detection, tracing the evolution from time- and frequency-domain statistical descriptors through wavelet/time-frequency representations to machine learning (SVM, KNN, ensemble/multi-classifier fusion) and deep learning approaches (CNNs on time-frequency scalograms, LSTMs on raw or transformed sequences). We present a generic system architecture and a step-by-step methodology flow for an sEMG fatigue-diagnosis pipeline, compare feature-extraction domains and classifier families, and summarise reported classification accuracies from recent literature. The review also discusses persistent challenges — inter-subject variability, electrode placement sensitivity, the trade-off between multi-classifier accuracy and system complexity, and the need for reliable MUAP decomposition — and outlines directions for future work, including transfer learning, multimodal sensing, and real-time embedded deployment for clinical and sports applications.