Postdoctoral Research Associate, School of Engineering
Abstract Title: Patient Derived and Neural Network Driven Epidural Electrical Stimulation Supports Sensorimotor Function in Humans with Complete Paraplegia
Abstract: Spinal cord injury (SCI) is a debilitating condition that can result in permanent loss of sensorimotor function below the SCI lesion. Although there is currently no cure for chronic SCI, epidural electrical stimulation (EES) of the lumbosacral spinal cord below the SCI lesion has emerged as a research technique to enable intentional control of motor functions that were lost after SCI. EES is hypothesized to function via activation of lumbosacral sensorimotor networks combining with descending signals crossing the SCI lesion via remaining neural fibers. Therefore, EES requires intact spinal cord tissue crossing the lesion and does not result in any proprioceptive feedback to participants during functional tasks. Herein, we demonstrate first-in-human results from a clinical study pursuing to bridge the SCI lesion gap via implantation of two EES electrode arrays, one rostral to the SCI lesion and one caudal. Two study participants with chronic, anatomically complete thoracic SCI were implanted with two EES electrode arrays, spanning the T1-T3 and T11-L1 vertebrae, respectively. Following 1-2 days of recovery, the study participants performed 12 days of experiments. Stimulation sweeps of varying stimulation locations, amplitudes, and frequencies were applied across both EES arrays. Motor activation was quantified via electromyography (EMG) of the upper and lower extremity, and perception was quantified via a tablet interface that reported location, intensity, and description of the elicited sensation. These results were used to train a neural network model that inferred optimal stimulation parameters based on desired EMG or perception patterns. Additionally, the participants were given direct control over stimulation parameters via user interfaces that enabled exploration of the parameter space to identify areas of perception and motor activation. Based on the identified EES parameters, stimulation was spatiotemporally applied on the rostral electrode array based on leg position, and the participants were able to accurately identify their leg position without visual feedback. Also, the participants were able to control flexion and extension of their lower limb via spatiotemporal stimulation on the caudal electrode array synchronized to movement of each upper limb. Finally, simultaneous rostral and caudal spatiotemporal stimulation was applied during treadmill walking to produce lower extremity motor activation and proprioceptive feedback regarding leg position during gait. These results demonstrate the ability to leverage state-of-the-art machine learning techniques and patient-derived feedback to derive optimal EES parameter selection to enable motor function below the SCI lesion with simultaneous EES above the SCI lesion to enable haptic proprioceptive feedback.