The objective of this proposal is to examine the time course of skill learning in the context of a brain computer interface (BCI) toward developing optimal training regimens not only for nave BCI subjects but for rehabilitation purposes in general. Toward this overall goal, this proposal will track the learning process of three male rhesus monkeys nave to BCI control as they learn to perform a 2D, 3D, and 4D BCI center-out task with target size as the main variable of difficulty. Using this experimental setup, the first im of this proposal seeks to develop a BCI training framework utilizing an adaptive task difficulty schedule. Previous BCI experiments have often altered difficulty for training purposes on an ad hoc basis, making evaluation and comparison of performance during learning problematic. We seek to address this problem by first defining a task's difficulty empirically through hardware and software simulations of the task parameters to determine chance performance levels for a given target size. These simulations are followed by simulations to determine a theoretical subject's utility curve (success rate vs. difficulty) for a given skill level, and are then fit to a parameteized model. With this model, we may estimate a user's current utility curve and skill level on the fly using catch trials to sample task difficulties outside of the subject's current training difficulty This new estimate is used to place the current task difficulty at a point expected to yield 70% success in order to main subject motivation while allowing sufficient feedback of errors to promote further learning. We will alternate this training schedule on a daily basis for comparison of within-day learning day and between day skill retention with a fixed difficulty regimen. Our second objective in this proposal is to identify the time course of improvements in BCI psychophysical metrics during long-term BCI skill acquisition. Borrowing from studies of natural motor skill learning, we hypothesize there will be distinguishable phases of BCI skill learning identified by the time course of improvements in different movement qualities. To test this, we will track average trial movement times, speed profiles, trajectory path lengths, trajectory smoothness, path variance, directional error, and directional bias over the course of weeks to months of training. Finally, th third objective of this proposal is to identify neural correlates of BCI skill acquisition. Recent studies have suggested similarities between BCI use and natural motor control, while others have demonstrated neural single-unit and population level changes that appear to coincide with native motor skill learning. Thus, we will track changes in neural activity at the level of the single-unit (directional tuning fit, depth of modulation, preferred direction variance), neuronal population (intrinsic dimensionality of activity, variance in active population membership, network structure), and local field potential (LFP directional tuning and modulation, spike-field coherence, cross- frequency coupling). These psychophysical and neural markers of learning could be used to further refine our proposed training algorithm and improve existing rehabilitation practices such as for stroke rehab.
This proposal investigates how the brain learns a new motor skill over time. Understanding how we progress from identifying the general movements appropriate for a task to refining their execution with practice is important for developing successful training regimens. These training principles may then be directly applied to rehabilitation of patients with motor impairment seeking to learn or relearn a motor skill.
|Williams, Jordan J; Tien, Rex N; Inoue, Yoh et al. (2016) Idle state classification using spiking activity and local field potentials in a brain computer interface. Conf Proc IEEE Eng Med Biol Soc 2016:1572-1575|