Muscle synergies to reduce the number of electromyography channels in neuromusculoskeletal modelling: a pilot study

Background: Surface electromyography driven models are effective in physical rehabilitation, yet their application out-of-the-lab is hampered due to extensive electromyography recordings requirement. Muscle synergy analysis has been proposed to predict missing muscles excitations from the measured electromyographic signals. Here is proposed a calibrated muscle-synergies approach to predict unmeasured muscle excitations, considering few surface electromyographic signals as motor control primitives to drive electromyography-based neuromusculoskeletal models. Methods: Data from walking trials, including kinematics, kinetics, and surface electromyographic signals of 15 lower-limb muscles, were collected from four individuals. The data was split into calibration and test sets. The calibration set served to determine the number of synergies (2 < k < 6) via non-negative matrix factorization and to calibrate the electromyography-driven neuromusculoskeletal models. The test set was used to predict full muscle excitations based on k-selected surface electromyography signals, identified during calibration. Findings: Using 4 < k < 6 experimental surface electromyography signals to reconstruct muscle excitations resulted in low normalized root mean squared differences (minimum 0.04 peroneus-longus, maximum 0.34 gracilis) and high determination coefficients (minimum 0.73 gracilis, maximum 0.99 peroneus-longus) in muscle forces when compared to fully experimental sEMG-driven models. Interpretation: The proposed method could serve to create reliable and efficient electromyography-driven models, transferable in ecological conditions, where reducing sensor number while maintaining a reliable characterization of subjects neuromuscular status is fundamental.