Decoding phantom limb movements from intraneural recordings

Limb loss leads to severe sensorimotor deficits and requires the use of a prosthetic device, especially in lower-limb amputees. While direct recording from residual nerves offers a biomimetic route for an effective prosthetic control, the low amplitude and noisy nature of these neural signals together with the challenge of establishing a reliable nerve interfacing, have hindered its adoption. Intraneural multichannel electrodes could potentially establish an effective interface with the nerve fibers, enabling access to motor signals even from muscles lost after the amputation. In this study, we report the direct neural recordings of two transfemoral amputees using transversal intrafascicular multichannel electrodes (TIME) implanted in the tibial nerves. We observed multiunit activity associated with volitional phantom movements of the knee, ankle and toes flexion and extension, with joint- and direction-specific neural modulation in both participants. The motor signals were distributed across all the electrodes, showing both single-joint and multi-joint selectivity, as well as direction selectivity for limb flexion and extension. After characterizing the neural evoked activity, we developed a Spiking Neural Network (SNN)-based decoder that outperform conventional motor decoders in predicting attempted phantom leg movements. Decoding accuracy improved further by including a broader signal bandwidth that captured both intraneural (ENG) and inter-muscular (imEMG) activity. Finally, comparing motor maps (recording) with sensory maps (stimulation) revealed a minimal overlap, suggesting early segregation of motor and sensory fibers within the tibial nerve before the knee bifurcation. Our findings demonstrate the feasibility to record motor signal and decode lower-limb movements directly from the nerves in amputees using intraneural interfaces. This paves the way for bidirectional, neurally-controlled prosthetic limbs combining natural control with somatosensory feedback through a single implanted interface.