Towards an EMG-based Decoder of Natural Motor Intentions in Amputees

Individuals with upper limb amputations or congenital defects face substantial challenges in performing daily tasks, often exacerbated by limited functionality in existing bionic prostheses. To address this, we present an AI-powered virtual reality (VR) system designed to train patients for controlling bionic prostheses with a high number of degrees of freedom using electromyographic activity decoding. Current prosthetic control systems lack sophistication, relying on discrete muscle activity decoding and manual mode switching. Our system employs a novel neural network architecture pre-trained on healthy subjects, facilitating continuous decoding of residual muscle activity to control a virtual limb in VR. Through more than three months of training, participants with congenital limb absence achieved precise finger-specific control, demonstrating a low error rate and strong correlation between target and decoded kinematics. Importantly, the system’s adaptability and user-friendly interface positively impacted patient experience, offering potential psychological benefits and preparing individuals for advanced prosthetic devices. This innovative approach represents a significant step toward enhancing prosthetic functionality and improving the quality of life for upper limb amputees and congenital defect patients.