EMG-Based Continuous Estimation of Index Finger Movements With Varying Interjoint Coordination Patterns by Modeling Musculoskeletal Dynamics

Surface electromyography (EMG) is widely used for predicting kinematics of intended finger movements in applications including teleoperation. Generally, ‘black-box’ models with high complexity such as neural networks (NN) were used to improve prediction accuracy, which may not reproduce other important movement characteristics such as smoothness or kinematic similarity. The goal of this study is to develop a novel EMG-based approach that models impact of finger intersegmental dynamics to reproduce ‘physiologic’ characteristics of coordinated finger movements. Performance of the proposed dynamic model was compared with a polynomial model with the same level of complexity (no dynamics considered) and NN models, based on (A) simulation data from four musculoskeletal systems with varying parameters; and (B) experimental data from 10 subjects performing finger movements with four distinct coordination patterns. Performance of the proposed dynamic model in predicting simulated movements was significantly better than the polynomial model with the same level of complexity (18 parameters). In predicting experimental data, performance of the dynamic model was significantly better than that of the NN model of lower complexity (480 parameters), and similar to that of the NN model of higher complexity (2976 parameters). Furthermore, movement quality produced by the dynamic model, quantified by jerk (smoothness) and kinematic similarity, and its computational efficiency were significantly better than other models. The proposed technique, which captures the impact of musculoskeletal dynamics in a compact form, can accurately reproduce physiologic finger movements with a higher computational efficiency than existing models, thus could serve as a robust tool for teleoperation.