Based on the linked source; automatically prepared and checked against the original report.
Study design
Researchers collected surface electromyography (sEMG) recordings from eight upper-limb muscles, alongside oxygen-uptake measurements, from nine wheelchair users—three women and six men. Participants completed incremental wheelchair propulsion on an ergometer. The research record summarizes the available abstract rather than the full paper, so the findings should be interpreted as preliminary study results.
How fatigue was defined
The investigators used ventilatory threshold, derived from oxygen-uptake data, as the label for fatigue onset. They then developed a dynamic weighted attention long short-term memory model, or DWA-LSTM, to distinguish propulsion cycles classified as non-fatigued from cycles transitioning toward fatigue. The approach also assessed which muscles contributed most strongly to prediction.
Reported model performance
Using sEMG intensity from all eight muscles, the model achieved an average classification accuracy of 94.82%. A single pectoralis major recording produced 89% accuracy, suggesting that a future wearable system might not require signals from every measured muscle. The authors indicate this could support single-muscle monitoring of propulsion-related fatigue onset, but the abstract does not establish clinical use, broader reliability, or treatment guidance.
