SyniEMG: An open-source platform for synthesizing intramuscular electromyography signals from kinematic inputs

Abstract

Abstract

Intramuscular Electromyography (iEMG) is a critical tool for neuromuscular diagnostics but is limited by its invasive nature, which causes patient discomfort and incurs significant costs. To address these challenges, we propose SyniEMG, an innovative, open-source platform that synthesizes iEMG signals from kinematic data using a hybrid generative model, providing a non-invasive and cost-effective alternative. SyniEMG integrates Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, attention mechanisms, and Generative Adversarial Networks (GANs) to effectively capture the spatial and temporal patterns necessary for accurate iEMG signal synthesis. The platform tackles the largely unexplored challenge of reverse neuromechanical modeling by converting kinematic data into continuous muscle activity signals, marking a significant advancement over traditional forward modeling techniques. We validated the platform using datasets, demonstrating its capability to accurately capture muscle activity dynamics. Moreover, SyniEMG requires only a brief training session to collect both iEMG and kinematic signals, enabling efficient synthesis of extended iEMG sequences. This approach reduces the risks associated with prolonged electrode use and provides a practical solution for continuous clinical monitoring and research. Quantitative evaluations in both time and frequency domains confirm the effectiveness of the platform, particularly in exoskeleton applications for motion enhancement and rehabilitation. By enabling indirect monitoring of muscle activity without the need for EMG sensors, SyniEMG has the potential to minimize injury risks, such as overstretching, during movement.