American Society for Peripheral Nerve

Back to 2018 ePosters

Extracting Regenerative Peripheral Nerve Interface Signals from Human Subjects for Neuroprosthetic control
Carrie A Kubiak, MD; Philip Vu, MSE; Zachary T Irwin, PhD; Chrono Nu, BSc; Troy Henning, MD; Deanna Gates, PhD; RB Gillespie, PhD; Theodore A. Kung, MD; Paul S. Cederna, MD; Cynthia Chestek, PhD; Stephen W. P. Kemp, PhD, MSc
University of Michigan, Ann Arbor, MI

Introduction: Peripheral nerves provide a promising source for neuroprosthetic control given their functional selectivity and relative ease of accessibility. However, current interface methods, such as penetrating electrodes, are limited in a clinical setting either by low signal amplitude or interface instability. Here, we address these issues by extracting hand level prosthetic control signals from Regenerative Peripheral Nerve Interfaces (RPNI) implanted within 3 human subjects.

Materials and Methods: RPNIs are constructed by suturing a graft of devascularized, denervated muscle to the residual end of a severed nerve. The graft then revascularizes, regenerates and becomes reinnervated by the transected nerve, creating a stable bioamplifier that produces recordable electromyography (EMG) signals. In addition, nerves can be surgically subdivided into individual fascicles to construct multiple RPNIs and independent signal sources. Here, 2 distal transradial (P1, P2) and 1 proximal transradial (P3) amputees were implanted with RPNIs. P1 and P2 were each implanted with 3 RPNIs with a single graft placed on each of the median, ulnar, and dorsal radial sensory nerve. P3 had 9 RPNIs implanted, with the median, ulnar, and radial nerves subdivided into 4, 3, and 2 branches, respectively. During acute recording sessions, we used ultrasound to locate and implant percutaneous fine-wire bipolar electrodes within the RPNIs. When available, relevant finger-related residual muscles were also implanted with control electrodes. A total of 3-8 electrodes were inserted per recording session.

Results: P1 and P2 produced EMG signals within a range of 20-270ÁVp-p from the median RPNI, with a signal-to-noise ratio (SNR) between 2.2-31.4 and 50-140ÁVp-p from the ulnar RPNI with a SNR between 3.3-11.6. P3's 3 ulnar RPNIs produced signals with a range of 30-50ÁVp-p and SNR of 1.6-5.6, and the 2 radial RPNIs produced signals with a range of 14.5-28.3 ÁVp-p and SNR of 2.0-3.0. Using a combination of RPNI and residual muscle signals, subjects successfully controlled a virtual prosthesis in real-time. For P1, a Na´ve Bayes classifier was able to classify movements as either rest, index, middle, or thumb opposition in a 212-trial session with 96.2% accuracy using temporal features of the EMG waveform within 300-1500Hz and binned at 50ms. Similarly, P2 could control rest, thumb, or index with 100% accuracy in a 115-trial session, while P3 could control between little and fist movements with 96.2% accuracy.

Conclusions: Overall, we have demonstrated that RPNIs may provide a clinically viable strategy for producing high amplitude signals from severed nerves for neuroprosthetic control.

Back to 2018 ePosters