Musculoskeletal Modeling Based on Muscle Synergy for Prediction of Hand and Wrist Movements

Decoding human movements using electromyography (EMG) signals is important for the development of EMG-based human-machine interfaces (HMIs). This study proposed a novel muscle synergy-based musculoskeletal model (MM) for prediction of hand and wrist movements, including wrist flexion/extension, wrist adduction/abduction, wrist pronation/supination, and metacarpophalangeal (MCP) flexion/extension. Ten limb-intact subjects were recruited for the offline experiment, and 15-channel EMG signals from the subject’s forearm were recorded. Using the non-negative matrix factorization (NMF) algorithm, four pairs of excitation signals were extracted from the multi-channel EMG signals. Then the MM driven by the extracted muscle excitations was adopted to predict hand and wrist movements. The proposed method was compared with the NMF algorithm and artificial neural network (ANN), and the prediction performance of the three was evaluated with Pearson’s correlation coefficient (r) and normalized root mean square error (NRMSE). The total average r of the proposed MM was 0.8475 across all subjects and all movement types, approximately 0.123 higher than NMF algorithm and 0.106 higher than ANN. In addition, the total average NRMSE of the proposed MM was 0.16125 across all subjects and all movement types, approximately 0.074 lower than NMF algorithm and 0.037 lower than ANN. In brief, the proposed MM showed significantly improved prediction accuracy over the NMF algorithm and ANN. This study provides a promising approach for the control of robotic arms and prostheses in EMG-based HMIs.