Electromyography-Driven Force Prediction in Non-Disabled Individuals and Transradial Amputees

Accurate estimation of force from electromyography (EMG) signals plays a crucial role in advancing human-machine interactions. While most existing studies have focused primarily on non-disabled subjects, research on force estimation in amputees is very limited. Accurate force estimation is particularly important for enabling seamless control of prosthetic devices. In this study, we propose an LTCN, a hybrid Long-Term Convolutional Network (LTCN) for force estimation from EMG signals recorded from both non-disabled subjects and individuals with trans-radial amputation. Both surface EMG (sEMG) and intramuscular EMG (iEMG) modalities were systematically evaluated to assess their contributions to accurate force estimation. For healthy subjects, the proposed LTCN model demonstrated the highest accuracy for sEMG-based prediction (R ${}^{2} =0.99$ , MAE = 0.0022, RMSE = 0.0029), while the LTCN model achieved similarly strong performance for iEMG-based prediction (R ${}^{2} =0.99$ , MAE = 0.0560, RMSE = 0.0638). For the amputee participant, sEMG signals were used, and the hybrid LTCN model performed well, showing good agreement with the actual force values and low prediction errors (R ${}^{2} = 0.86~\pm ~0.01$ ; MAE = 0.8575; RMSE = 1.3362). The LTCN demonstrated robust prediction performance across both subject groups, highlighting its potential for improving prosthetic control and rehabilitation strategies. Future work may focus on integrating the LTCN model directly into prosthetic systems to enhance intuitive and responsive control for users.