Long-term Signal Stability of Regenerative Peripheral Nerve Interfaces (RPNIs) in humans with upper-limb amputations
Philip P Vu, PhD1, Alex K. Vaskov, MSE1, Scott R. Ensel, MSE1, Theodore A Kung, MD2, Cynthia A Chestek, PhD3, Paul S Cederna, MD4 and Stephen WP Kemp, PhD1, (1)University of Michigan, Ann Arbor, MI, (2)Section of Plastic & Reconstructive Surgery, University of Michigan, Ann Arbor, MI, (3)Biomedical Engineering, University of Michigan, Ann Arbor, MI, (4)Plastic Surgery, University of Michigan, Ann Arbor, MI
Introduction: State-of-the-art control systems can provide upper-extremity prosthetic users greater control, but rely on recording methods limited to low SNRs between 5-20. Consequently, the control systems require frequent recalibration, leading to user frustration and abandonment. To develop a more stable recording interface, we have developed the Regenerative Peripheral Nerve Interface (RPNI). A muscle graft reinnervated by a transected peripheral nerve, RPNIs have produced signal to noise ratios (SNR) 12x greater than other methods. In this study, we hypothesize that RPNI signal quality can remain stable over time, and allow control systems to maintain prediction performance without recalibration.
Materials and Methods: Two participants with transradial amputations (P1 and P2) underwent surgery to place RPNIs on each of their median, and ulnar nerves in 2015 and 2017, respectively. In 2018, intramuscular bipolar electrodes were placed into their RPNIs. Participants were instructed to volitionally move their phantom limb to mirror 3 hand postures. These tasks were repeated in 11 sessions over 276 days and 14 sessions over 422 days. SNRs were tracked across sessions, and if possible, motor unit action potentials were extracted using Offline Sorter (Plexon Inc.). Machine learning parameters calculated in the first session were reused in the subsequent sessions to predict hand postures, and no recalibration of the parameters was performed.
Results: In both participants, SNRs remained high ranging from 15-250 across sessions, and did not increase or decrease over time. However, signals did vary substantially from session to session. Specifically, P1's median RPNI SNRs ranged between 65-250 across days for thumb flexion, whereas the ulnar RPNI SNRs ranged between 15-55 for small finger flexion. P2's SNRs ranged between 20-40 for the median RPNI and 15-80 for the ulnar RPNIs. Contrarily, tracking a single motor unit action potential in P2 across 3 days showed an increase trend in peak-to-peak amplitude of 11.5%. When predicting 3 separate movements, the first session and last session had prediction accuracies greater than 95% with a few days in between predicting below 80% accuracy.
Conclusions: SNRs remained high with values that were 3x or up to 50x higher than other methods. Single motor unit analysis showed small changes in motor unit amplitudes, which suggest that electrode migration may be occurring but at a minimum. Control system prediction performance remained high regardless of changes in SNR. Overall, RPNIs have the potential to be a stable interface that minimizes recalibration, increasing patient satisfaction and decreasing user abandonment.
Back to 2021 ePosters