Improving hand kinematic predictions from implanted EMG in humans and monkeys

Project Summary There are millions of people worldwide with debilitating upper limb amputations. While electrical signals from residual muscle can provide some function, every amputee is missing muscles, and therefore missing a variety of important functions. Our group has demonstrated a novel method for obtaining signals from independent nerve fascicles in humans, which we call the Regenerative Peripheral Nerve Interface (RPNI). The small muscle grafts degenerate, regenerate, revascularize, and reinnervate utilizing natural biologic processes. They also introduce a degree of conformity among prosthetic users, for example always having thumb muscles available for electromyography (EMG). Our long-term goal is to achieve able bodied performance for prosthetic hand movement. The objective of our current application, which represents the next step, is to develop reusable deep learning architectures for controlling wrist and finger movements. We will achieve this with the following specific aims. In Aim 1 we will utilize a range of deep learning techniques we developed for brain machine interfaces to use with implantable EMG signals for truly continuous control of finger movement. This will be done in monkeys and humans with similar implanted electrodes. In Aim 2 we will achieve simultaneous control of the wrist and fingers by learning to segregate stabilization related EMG from wrist movement related EMG, again in both humans and monkeys. Finally, in Aim 3, in humans, we will quantify the biomechanical efficiencies gained from using our novel prosthetic decoders testing the likely clinical impact of this approach. We believe that the demonstration of higher performance across the board will motivate widespread use of RPNI and implantable EMG for prosthetic control after upper limb amputation.