Edge AI-Based Closed-Loop Peripheral Nerve Stimulation System for Gait Rehabilitation after Spinal Cord Injury

Recently, it has been revealed that Vsx2 neurons in the lumbar of the spinal cord are essential to restore locomotor function after spinal cord injury (SCI). However, Vsx2 neurons still need afferent feedback from peripheral nerves for locomotor rehabilitation. Also, movement-dependent electrical stimulation after SCI can solidify synaptic connections in a directed way. Based on these two facts, we hypothesized that providing movement-dependent electrical stimulation to hindlimb nerves may facilitate the rehabilitation process for restoring locomotion function after SCI while preventing aberrant remodeling of denervated spinal circuits. To evaluate our hypothesis, we developed an edge artificial intelligence (AI)-based closed-loop peripheral nerve stimulation system which can timely generate stimulation pulses for distal-tibial and peroneal nerves based on stance phase detection. The main parts of the system consist of a multi-site EMG electrode, neural amplifiers, an edge AI processing circuit, and neural stimulators. Medial gastrocnemius (MG) electromyography (EMG) was used as a input source of the AI model to detect the stance phase. The AI model was deployed on a dual-core 32-bit microprocessor. The whole system was evaluated with two SCI rats walking bipedally on a motorized treadmill. The accuracy of the presented AI model was calculated as 97.19%. In the animal experiments with SCI rats, stimulation pulses for the distal-tibial and peroneal nerves were timely generated for 200 ms and 100 ms, respectively based on the stance phase detected by the AI model. The experimental results suggest that the presented system can be a powerful neural interface tool to investigate the efficacy of edge AI-based closed-loop peripheral nerve stimulation on restoring locomotion function after SCI.