ECoG correlates of visuomotor transformation, neural plasticity, and application to a force-based brain computer interface

Electrocorticography: ECoG) has gained increased notoriety over the past decade as a possible recording modality for Brain-Computer Interface: BCI) applications that offers a balance of minimal invasiveness to the patient in addition to robust spectral information over time. More recently, the scale of ECoG devices has begun to shrink to the order of micrometer diameter contacts and millimeter spacings with the intent of extracting more independent signals for BCI control within less cortical real-estate. However, most control signals to date, whether within the field of ECoG or any of the more seasoned recording techniques, have translated their control signals to kinematic control parameters: i.e. position or velocity of an object) which may not be practical for certain BCI applications such as functional neuromuscular stimulation: FNS). Thus, the purpose of this dissertation was to present a novel application of ECoG signals to a force-based control algorithm and address its feasibility for such a BCI system. Micro-ECoG arrays constructed from thin-film polyimide were implanted epidurally over areas spanning premotor, primary motor, and parietal cortical areas of two monkeys: three hemispheres, three arrays). Monkeys first learned to perform a classic center-out task using a brain signal-to-velocity mapping for control of a computer cursor. The BCI algorithm utilized day-to-day adaptation of the decoding model to match the task intention of the monkeys with no need for pre-screeening of movement-related ECoG signals. Using this strategy, subjects showed notable 2-D task profiency and increased task-related modulation of ECoG features within five training sessions. After fixing the last model trained for velocity control of the cursor, the monkeys then utilized this decoding model to control the acceleration of the cursor in the same center-out task. Cursor movement profiles under this mapping paralleled those demonstrated using velocity control, and neural control signal profiles revealed the monkeys actively accelerated and decelerated the cursor within a limited time window: 1-1.5 seconds). The fixed BCI decoding model was recast once again to control the force on a virtual cursor in a novel mass-grab task. This task required targets not only to reach to peripheral targets but also account for an additional virtual mass as they grabbed each target and moved it to a second target location in the presence of the external force of gravity. Examination of the ensemble control signals showed neural adaptation to variations in the perceived mass of the target as well as the presence or absence of gravity. Finally, short rest periods were interleaved within blocks of each task type to elucidate differences between active BCI intention and rest. Using a post-hoc state-decoder model, periods of active BCI task control could be distinguished from periods of rest with a very high degree of accuracy: ~99%). Taken together, the results from these experiments present a first step toward the design of a dynamics-based BCI system suitable for FNS applications as well as a framework for implementation of an asyncrhonous ECoG BCI