Development of an Upper Limb Neuroprosthesis: Conquering Weakness and Fatigue

Abstract

Neuroprosthetics are devices that substitute for or supplant motor, sensory or cognitive modalities damaged as a result of spinal cord injury or stroke. Functional electrical stimulation (FES) neuroprosthetics utilize artificial stimulation to restore motor function in paralyzed muscles, where control exerted by higher nervous system centers over muscle may be impaired. Although promising, FES has failed to gain widespread acceptance due in part to weak contraction strength and rapid fatigue observed with artificial stimulation. This dissertation documents an attempt to create an upper limb FES neuroprosthetic and subsequently to address the issues of weakness and fatigue. To exploit the capabilities of the musculoskeletal system the neural drive to muscle first must be decoded. Decoding the neural drive for specific movements has been approached using either a deterministic (engineering) or machine learning model. While a deterministic model accounts for all components of a limb, number of joints, degrees of freedom, limb length, muscle length, etc, machine learning characterizes the relationship between select variables, in this case whole muscle electromyographic data (EMG) and limb kinematics. Ultimately, the output of both approaches is used to predict the neural drive required to generate movements. In this study we first attempt to build an upper limb FES neuroprosthetic. Utilizing machine learning, we characterize the relationship between limb kinematics and EMG. Then, predict EMG based solely on limb kinematics. Finally, stimulation pulses were generated and delivered via intramuscular electrodes to produce movement. Additionally, to address force generation we hypothesized that due to the distributed nature of motor axons within a muscle stimulating with multiple spatially distributed electrodes would activate a larger muscle volume thus generating additional force. This in turn would facilitate load sharing among muscle fibers, and reduce fatigue. To evaluate fatigue we compared interleaved and synchronous patterns of stimulation as well as single electrode vs multiple electrode stimulation. We approached these questions and aims with a combination of strategies and techniques including machine learning, implantation of stimulating electrodes in a non-human primate model and finally human subjects. While machine learning provided EMG predictions with high R values, we were unable to generate substantive movements activating all the muscle in a complete joint system. However, we were able to generate movements stimulating a single muscle in an intact joint system. We found that single electrode force could be augmented with multiple electrodes. Additional results indicate that multiple electrode stimulation was less fatiguing than single electrode stimulation. Interleaved stimulation however, did not result in less fatigue than synchronous stimulation.