This item is under embargo and not available online per the author's request. For access information, please visit http://libanswers.wustl.edu/faq/5640.
Date of Award
Doctor of Philosophy (PhD)
Paralysis, due to spinal cord injury, amyotrophic lateral sclerosis (ALS), or stroke, is the result of severed communication between the brain and the motor periphery. Brain computer interfaces (BCIs) are neuroprosthetic devices that create novel communication pathways by measuring and transforming neural activity into operational commands. State of the art BCI systems measure brain activity using penetrating electrode arrays able to record from hundreds of individual cortical neurons simultaneously. Unfortunately, these systems are highly susceptible to signal degradation which limits their efficacy to 1-2 years. However, electrocorticography (ECoG) signals recorded from the surface of the brain deliver a more competitive balance between surgical risk, long-term stability, signal bandwidth, and signal-to-noise ratio when compared to both the aforementioned intracortical systems and the more common non-invasive electroencephalography (EEG) technologies.
Historically, neural signals for controlling a computer cursor or robotic arm have been mapped to extrinsic, kinematic (i.e. position or velocity) variables. Although this strategy is adequate for use in simple environments, it may not be ideal for control of real-world prosthetic devices that are subject to external and unexpected forces. When reaching for an object, the trajectory of the hand through space can be defined in either extrinsic (e.g. Cartesian) or intrinsic (e.g. joint angles, muscle forces) frames of reference. During this movement, the brain has to perform a series of sensorimotor transformations that involve solving a complex, 2nd order differential equation (i.e. musculoskeletal biomechanics) in order to determine the appropriate muscle activations. Functional neuromuscular stimulation (FNS) is a desirable BCI application because it attempts to restore motor function to paralyzed limbs through electrical excitation of muscles. Rather than applying the conventional extrinsic kinematic control signals to such a system, it may be more appropriate to map neural activity to muscle activation directly and allow the brain to develop its own transfer function.
This dissertation examines the application of intrinsic decoding schemes to control an upper limb using ECoG in non-human primates. ECoG electrode arrays were chronically implanted in rhesus monkeys over sensorimotor cortex. A novel multi-joint reaching task was developed to train the subjects to control a virtual arm simulating muscle and inertial forces. Utilizing a co-adaptive algorithm (where both the brain adapts via biofeedback and the decoding algorithm adapts to improve performance), new decoding models were initially built over the course of the first 3-5 minutes of each daily experimental session and then continually adapted throughout the day. Three subjects performed the task using neural control signals mapped to 1) joint angular velocity, 2) joint torque, and 3) muscle forces of the virtual arm. Performance exceeded 97%, 93%, and 89% accuracy for the three control paradigms respectively. Neural control features in the upper gamma frequency bands (70-115 and 130-175 Hz) were found to be directionally tuned in an ordered fashion, with preferred directions varying topographically in the mediolateral direction without distinction between motor and sensory areas. Long-term stability was demonstrated by all three monkeys, which maintained performance at 42, 55, and 57 months post-implantation. These results provide insights into the capabilities of sensorimotor cortex for control of non-linear multi-joint reaching dynamics and present a first step toward design of intrinsic, force-based BCI systems suitable for long-term FNS applications.
Daniel W. Moran
Dennis L. Barbour, Eric C. Leuthardt, Baranidharan Raman, Lawrence H. Snyder,
Available for download on Friday, April 19, 2019