Paralysis resulting from spinal cord injury (SCI) is devastating, dramatically reducing the independence of affected individuals. Currently, functional electrical stimulation (FES), controlled by a patient’s residual movements, is used clinically to restore a limited range of voluntary movement. However, if FES could be controlled using signals recorded from the brain, it might allow patients with high-level SCI to regain even more natural and sophisticated movements. Cortically-controlled FES has been successfully used in animal experiments and in preliminary human clinical trials, but it needs refinement before it can be fully translated to the clinic. Here I present three distinct studies, each of which addresses the improvement of a system control strategy. Taken together, my three studies offer insights that will improve the future implementation of cortically-controlled FES. In my first study, I evaluated the ability to use peripheral nerve stimulation to selectively activate muscles for FES. I demonstrated that the Flat Interface Nerve Electrode (FINE) can selectively stimulate a subset of wrist and hand muscles, and that this stimulation is stable over a period of 4 months. In future implementations of FES, nerve stimulation can therefore be used to selectively stimulate a subset of muscles without the need to implant these muscles individually. This method may be especially useful for muscles which are difficult to individually implant and stimulate intramuscularly without current spillover. Cortically-controlled FES also relies on the ability to accurately predict muscle signals (EMG) from neural activity in motor cortex (M1) using a mathematical algorithm, or neural “decoder”. In my second and third studies, I address the question of how accurate a decoder needs to be, both for making accurate EMG predictions across behaviors, and for facilitating intuitive user control. No decoder can be expected to be perfect, but I also evaluate the brain’s ability to adapt to imperfect decoders, which may ultimately enable the successful restoration of movement. I first examine the accuracy of a single decoder for predicting actual wrist EMG across three highly varied dynamical conditions: isometric forces, unloaded movements, and movements against an elastic load. To allow a decoder to perform well across these tasks, it needs to be trained on data from all three, and furthermore, needs to be nonlinear. Second, I evaluate the ability of monkeys to learn two different kinds of altered decoders: one that preserved the natural coactivation patterns of muscles, and one that didn’t. The monkeys are better able to learn to use the former decoder, and never accomplish all task goals in the latter case. Taken together, my results suggest that neural decoders should include robust multi-task training, and should account for nonlinearities in the motor system. They also suggest that imperfect EMG decoders can be learned, as long as they take into account the natural activation patterns of muscles. Overall, the results presented in this dissertation offer insights and tools that will improve the future implementation of cortically-controlled FES.

Last modified

• 10/22/2018

Creator

• Stephanie Naufel Naufel

DOI

• https://doi.org/10.21985/N2WT8W

Subject

• Biomedical Engineering

Keyword

• brain-machine interfaces
• motor cortex
• neuromodulation

Date created

• 2017-01-01

Resource type

• Dissertation

Rights statement

• In Copyright