Demographic-Driven Electromyography Analysis: Advancing Personalized Biosignal Interpretation

The integration of rehabilitation robotics and surface electromyography (sEMG) offers a powerful approach for monitoring and enhancing recovery in patients with neuromuscular disorders. However, variability in baseline sEMG readings across individuals can limit its effectiveness. Factors such as age, height, and weight influence these baselines, and there is a lack of personalized baselines that account for demographic differences. This study proposes a novel model to estimate individualized baselines for one important sEMG parameter, Root Mean Square (RMS). Demographics and physiological data were collected from 30 healthy participants, and sEMG signals were recorded using four electrodes on the forearm muscles during a pushing task at two wrist positions. A Decision Tree Regression model was developed for each combination of the two wrist postures and four electrode locations, resulting in eight combinations, with optimal features identified using the Recursive Feature Elimination method. The regression models achieved accuracies ranging from 88.81% to 95.6%. A global sensitivity analysis using the Sobol method evaluated the importance of each input feature. Results indicate that gathering more comprehensive sEMG data for the most influential factors could improve model generalizability. The findings of this study offer a promising approach for individualized sEMG baselines, with potential applications in rehabilitation robotics to enable personalized recovery strategies for neuromuscular disorders.