Shallow neural networks and high-density electromyography for proportional control of hand kinematic state: applications in healthy subjects and upper-limb impaired patients

Thesis

Upper limb impairments significantly compromise functional independence and quality of life for individuals, yet current assistive technologies often fail to meet expectations of users for intuitive, effective control. This thesis addresses this gap through the development of the RPC-Net/HDE-Array system, a unified platform combining High-Density surface Electromyography (HD-sEMG) with a physiologically inspired regression-based neural controller. The system includes a dry-electrode, circumferentially distributed bracelet (HDE-Array) designed for user-friendly signal acquisition, and RPC-Net, a recursive, shallow neural network that decodes HD-sEMG into online, multi-degree-of-freedom (DoF) hand kinematics. Six experimental studies validate the accuracy of the system, its robustness to electrode variability, and its usability by both able-bodied and tetraplegic participants. Findings demonstrate that the platform matches or exceeds state-of-the-art decoding performance, maintains stability under non-ideal conditions, and enables effective real-time control using both forearm and neck muscles. Novel contributions include recursive signal integration, PCA-driven dimensionality reduction, and the demonstration of neck-based control for inclusive rehabilitation. The thesis provides open-source code and datasets to support reproducibility and outlines pathways for future clinical deployment, including integration with prosthetic and FES devices. Collectively, this work establishes a viable, user-centred control interface for upper-limb rehabilitation, bridging the gap between laboratory-grade systems and practical, real-world applications.

Authors: Rolandino, G

• Alternative title: Shallow neural networks and high-density electromyography for proportional control of hand kinematic state
• DOI: 10.5287/ora-rybddmaed
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: electromyography; machine learning; rehabilitation device; brain machine interface
• Subjects: Biomedical engineering