Improved Gait Posture Prediction in Transfemoral Amputees with Reconstructed Shank EMG Signals

Electromyographic signals (EMGs) have been widely used in robotic prostheses control in order to achieve the natural gait posture as an actual lower limb. However, especially for patients with high-level amputations, such as transfemoral amputees, the lack of EMGs in the amputated part of the body poses a challenge in gait posture prediction and makes the remaining accessible EMGs difficult to be used as a reliable control source. In this study, we propose to reconstruct the shank EMGs for transfemoral amputees, using the EMGs from the thighs. Then, the augmented EMGs (from both thighs and shanks) is used to predict feet position in gait cycles. We compare the prediction results using five different learning methods. It is found that with the reconstructed shank EMGs, the results show improved performance comparing to using the thigh EMGs to predict feet location directly. We also confirm that the prediction system with augmented EMGs recovers 95.28% of the performance on average, comparing to prediction systems using both thigh and shank EMGs.