Evaluating the robustness of an ultrasound based sensing strategy for intuitive control of upper extremity prosthetics

Abstract

Current commercially available prostheses based on myoelectric control have limited functionality, leading to many amputees abandoning use. Myoelectric control using surface electrodes has a number of limitations and lacks specificity for deep muscles, presenting a continued need for more robust strategies. We propose a new strategy for sensing muscle activity based on real-time ultrasound imaging. Results from our previous work demonstrate that complex motions could be classified with 92% accuracy in real-time. However, arm and hand repositioning during natural movements tend to alter the geometry of forearm musculature, possibly affecting performance. In this study, we evaluated the robustness of the image-based control strategy in the presence of varied forearm positions on able-bodied subjects. Ultrasound images of the forearm muscles were collected during two different scenarios using a Sonix RP with a 5–14 MHz linear probe. The subject was asked to perform four hand motions at eight different arm positions and three levels of wrist pronation. Images were analyzed to generate activity patterns for each motion and then classified. Results demonstrate that forearm positions do not significantly compromise reliability. We also show that performance could be further improved by including additional training activity patterns corresponding to motions performed in a few selected arm positions.