Machine Learning-Based Fatigue Detection in Neuromuscular Rehabilitation

Beschreibung

Functional electrical stimulation (FES) is established in neurorehabilitation, but its usefulness is limited by early muscle fatigue. Reliable, context-sensitive detection of functional fatigue is therefore needed, ideally based on forces and movements rather than muscle activity alone. This work develops a phase-oriented framework for machine-learning-based fatigue detection under FES, combines product development, CRISP-DM and maturity levels. In a main study, isometric and dynamic contractions of forearm muscles were recorded in ten healthy subjects under four stimulation protocols. Grip and finger forces, joint movements and sEMG were measured, and cycle-based parameters were derived. Conservative physiologically motivated criteria combining force decline and performance loss were used to generate robust class labels. A reliable ML pipeline with versioned data, defined preprocessing, leave-one-subject-out evaluation, and feature- and sequence-based models was implemented. Mechanical parameters were more sensitive to fatigue than sEMG amplitudes, and no universal fatigue criterion was identified. Fatigue detection at motor endpoints proved feasible, although target performance was not fully reached. The proposed framework provides a transparent basis for safety-critical, data-driven fatigue detection.

Autor*in

Sascha Selkmann

Themen in »Machine Learning-Based Fatigue Detection in Neuromuscular Rehabilitation«

Functional electrical stimulation Muscle fatigue Time series classification

Stimmen zu »Machine Learning-Based Fatigue Detection in Neuromuscular Rehabilitation«

Details

Link teilen