Simultaneous and proportional control of wrist and finger force via motor unit activity

• Published in: Biomedical signal processing and control Vol. 103; p. 107399
• Main Authors: Xia, Miaojuan, Chen, Chen, Li, Dongxuan, Sheng, Xinjun, Ding, Han
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.05.2025
• Subjects: Hand-wrist system; High-density electromyography; Motor unit; Neural interfacing; High-density electromyography; Hand-wrist system; Motor unit; Neural interfacing
• ISSN: 1746-8094
• DOI: 10.1016/j.bspc.2024.107399

Myoelectric interfaces hold promise for restoring motor functions in amputees by enabling natural and intuitive control of daily-life reaching and grasping tasks. However, current myoelectric interfaces still face challenges in predicting both wrist and finger movements, thereby restricting the dexterity of prosthetic hand control. In this study, we proposed a novel strategy for simultaneous and proportional control of wrist and pinch force. We collected motor unit filters (MU filters) during single degree-of-freedom (DOF) tasks and applied them to multi-DOF tasks, which significantly expanded the dimensionality of the neural feature space. Additionally, motor unit spike counts were integrated into a ridge regression model to address the multicollinearity issues that we detected. We evaluated the model performance using the normalized root mean square error (nRMSE) and the coefficient of determination (R2). Ten subjects were recruited to perform combinations of different DOF tasks, including wrist flexion/extension and two different pinches, with varying force levels. We compared the proposed neural feature to two other neural features: direct decomposed motor units and cumulative spike trains (CST), and one traditional feature: root mean square (RMS). The experimental results demonstrated that the proposed method outperformed comparative approaches, achieving an nRMSE of 12.1% ± 2.2% and an R2 of 0.81 ± 0.05 on average across all DOFs and subjects. In addition, the proposed method maintained comparable accuracy across different linear models compared to other input features. This study has the potential to empower prosthesis users to achieve more natural and dexterous control. •Proposes a novel strategy for simultaneous and proportional control of wrist and finger forces using motor unit (MU) activity.•MU filters collected during single-DOF tasks are successfully applied to multi-DOF tasks, expanding the neural feature space.•Ridge regression addresses multicollinearity in sEMG signals and neural firings, improving control precision.•Experimental results outperform baseline models and features, achieving accurate force estimation across multiple DOFs.•The proposed method shows potential for enhancing the dexterous control of myoelectric prostheses in daily activities.