Stroke is the number one cause of movement disability in the world. In recent years, robotic assistance has empowered people with stroke to complete intensive movement therapy in motivating environments, thus matching or bettering the motor recovery attainable with traditional therapy. Yet, motor deficits remain stubbornly persistent, especially for those with severe impairments.
Brain-computer interfaces (BCI) are a technology that can facilitate direct communication between the brain and an external device. BCIs have already been used to control robotic prostheses to replace lost function. The premise of this dissertation is that, with the right tools and knowledge, BCIs could also help restore function to those with movement disability after a neurologic injury. In this dissertation, I investigate use of a BCI to help individuals with a stroke shape their brain activity while moving the fingers with assistance from a robotic orthosis, with the goal of guiding activity-dependent plasticity in the brain to drive motor recovery. The working hypothesis is that appropriately shaping brain activity will improve finger movement ability and provide a therapeutic benefit after stroke.
First, I present a computational model of motor learning that uses a neural network to simulate the motor cortex after a stroke and during subsequent finger force recovery. These simulations suggested that BCI-based interventions should target perilesional motor areas, thus restoring normative network recruitment during finger movement, and that targeted training should make up about 20% of total limb use to maximize recovery.
In a study of unimpaired people completing a robot-assisted movement task, I identified a key confound of BCI-contingent robot-assisted therapy, showing that robot assistance can affect the BCI even when the participant is passive, which may hinder motor learning. I also present a potential design approach for both the robot and the BCI to avoid this confound.
I then explore BCI methodological considerations in two experiments with impaired and unimpaired people moving in a robot-assisted environment. Key results included that bipolar EEG recordings and finger extension movements led to the best models correlating brain state with ensuing movement and are thus most conducive to BCI-based training.
The culmination of this work is the design of a BCI-robot rehabilitation paradigm, which I tested in a study with eight people with severe impairment after a chronic stroke. Participants participated in four weeks of a therapy protocol that determined the effect of BCI-based sensorimotor rhythm control on finger extension performance. Here, we found that BCI training can improve subsequent movement performance – a result never before found for individuals with a stroke. The training also produced therapeutic benefits, indicating its viability as a future rehabilitation intervention. Finally, looking to the future of BCI-robot therapy, I present low-cost alternatives for BCI signal acquisition and wearable robotic devices.