Longevity of continuous prosthetic finger control using implanted electromyographic signals without recalibration

Background For decades, the field of robotic prosthetic hands has been hindered by a fundamental lack in effective, reliable control strategies. Currently, myoelectric hand prostheses are most often controlled with muscle activity recorded from the surface of the skin, or surface-electromyography (EMG). However, surface-based systems are mainly limited to grip control that requires frequent calibration. Methods In this study, we use EMG signals recorded from electrodes implanted in regenerative peripheral nerve interfaces and residual innervated muscles of the arm to decode continuous movements of two virtual finger groups in an able-bodied monkey and in a person with a transradial amputation. We compared the decoding performance of a linear and a nonlinear decoder against the participants' natural hand performance during a two-finger random target matching task. To assess the reliability of this control, we measured the participants' task performance over several months without recalibrating decoder parameters and investigated trends in online EMG-patterns when a static or same-day trained decoder was used. Results Here we demonstrate two critical milestones for prosthetic control simultaneously: high performance, continuous control of two finger groups and evaluate long-term stability of this control without decoder recalibration. Using a linear decoder, the monkey and the human performed the two-finger task with 63% and 56% task performance relative to their natural hand performance. Notably, when we tested these decoders without recalibration, we found that for both participants there was no significant decline in task performance (Monkey: p = 0.21, Human: p = 0.78) for several months (Monkey: 148 days, Human: 283 days). Interestingly, users adapted their muscle activation patterns to match the original decoder calibration, allowing for accurate finger control months after calibration even when the signals themselves changed day-to-day. Conclusions These results demonstrate that electromyographic signals recorded from intramuscular electrodes could serve as a stable neural interface for prosthetic control, and that long-term performance may depend on user adaptation rather than signal stability alone. Trial registration: Clinicaltrials.gov NCT03260400. Registered 25 July 2017.